Chapter 2: The Epistemology of Cognitive Neuroscience

Chapter Introduction

The Turtle has walked with you a long way.

In K-12 you met your brain — its general anatomy and function, the basic neurotransmitter systems, the broad strokes of attention, learning, sleep, stress, and reward. At Associates you went into neuroscience proper — neurons and glia, ion-channel biophysics, the major neurotransmitter systems at receptor depth, the limbic and cortical anatomy, LTP and the BDNF cascade, the HPA axis and allostatic load, Posner's attention networks, the reward circuitry. At Bachelor's you went deeper — Hodgkin-Huxley as mathematical model, SNARE-complex molecular detail, AMPAR trafficking and PKMζ; dorsolateral prefrontal persistent activity, the Schultz dopamine prediction error unified with Sutton-Barto temporal-difference learning, addiction at Berridge wanting/liking and Nestler ΔFosB depth; the Ogawa BOLD signal, the Eklund 2016 cluster-correction crisis, the optogenetic and chemogenetic revolution. At Master's you went translational — antidepressant pharmacology and the ketamine paradigm shift, neuroimaging methodology at graduate depth, computational psychiatry, the inflammatory hypothesis of depression, and the bench-to-bedside pipeline's notorious difficulty in this field.

This chapter is the fourth and final step of the upper-division spiral.

At the Doctorate level, Coach Brain goes meta. The translational engagement of Master's is the substrate of this chapter, not its content. What this chapter asks is the next question: how does the field of cognitive and clinical neuroscience know what it thinks it knows, where do its unresolved questions live, what theoretical frameworks compete for the field's allegiance, and what original research would advance the science beyond its present limits? This is the doctoral question. Where Master's read the field's published clinical trials and translational programs and learned to evaluate them, Doctorate reads the field as an epistemological enterprise — a community of researchers operating under specific methodological constraints (small samples, indirect measurement, reverse-inference difficulty), structural incentives (publication, novelty, the funding cycle's tempo), theoretical commitments (information-processing, predictive processing, embodied cognition, integrated information), and historical conditions (a young field, an emergent set of measurement technologies, a maturing relationship to philosophy of mind). The substance of cognitive neuroscience is no longer the content of this chapter. The character of cognitive neuroscience as a knowledge-producing system is.

The voice is the same Turtle. Patient. Methodical. Slow and deep. What changes once more is the depth. At Doctorate you are no longer reading the published intervention trials, the imaging studies, and the computational models and weighing them against one another. You are reading them, and reading the methodological commentaries on them, and reading the philosophical literature on what they can and cannot establish about the brain and the mind, and reading the historical archives that document how the field arrived where it has arrived. You learn to read cognitive neuroscience as a doctoral student in any natural science learns to read a field: as something that was made under conditions, that could have been made differently, that will be remade by the work you and your peers go on to do, and that exists in a particular relationship to its parent disciplines (psychology, philosophy of mind, biology, computer science, medicine) that the contemporary researcher must navigate with awareness.

A word about prescriptions, before you begin. The rule has not changed and does not change at Doctorate. The Turtle teaches cognitive neuroscience as a research enterprise, not as personal prescription. Nothing in this chapter is clinical advice. The research methodology engaged here — the strengths and weaknesses of small-sample fMRI inference, the structural critique of the cognitive neuroscience literature's published findings, the theoretical-framework debate between predictive processing and traditional information-processing accounts of cognition, the philosophical literature on the hard problem of consciousness — is presented at research-track depth so that you can read the methodological and theoretical literature in its own form and contribute to it as you go on to do original work. None of it is a recommendation about pharmacotherapy, neurostimulation, psychedelic-assisted therapy, or any other clinical intervention. Any such decision — yours, a research participant's, a patient's — is the proper subject of a clinical conversation with appropriately licensed and trained clinicians, never the conclusion of a chapter.

A word about being a doctoral-level reader in this field, before you begin. This audience reads the chapter from a different position than the Master's audience did. Some of you are training to do original research in cognitive neuroscience, computational neuroscience, clinical neuropsychology, psychiatry research, or neuro-philosophy. Some of you are physician-scientists, clinical psychologists, or computational scientists working at the intersection of neuroscience and your home discipline. Some of you are philosophers of mind or philosophers of cognitive science reading neuroscience as a case in the structure of contemporary scientific knowledge. The chapter is written for that audience. The framing throughout remains research-descriptive and methodologically careful, never diagnostic or prescriptive. The work of this chapter is to build the meta-understanding that informs original research — that allows you to choose your research questions wisely, design your studies well, read your discipline's literature with appropriate skepticism, and contribute methodology and theory and not only findings to the field.

A word about mental health, before you begin. The methodological content in this chapter engages the literature on depression, anxiety, addiction, suicide research, and the broader clinical neuropsychiatric landscape at methodology-critique depth. The doctoral neuroscience and clinical-psychology training years remain a real-incidence window for several of the conditions the chapter examines. Caring for the populations you will go on to serve requires that you be well yourself. The verified crisis resources at the end of this chapter are real. The 988 Suicide & Crisis Lifeline is real. Your program's counseling and student wellness resources are real. If anything in this chapter — methodological, theoretical, philosophical, or substantive — surfaces patterns that feel anxious, rigid, or out of proportion to ordinary intellectual engagement, pause. The Turtle is patient with you.

This chapter has five lessons.

Lesson 1 is The Epistemology of Cognitive Neuroscience — the philosophical question of what cognitive neuroscience as a field is in a position to know, the levels-of-explanation question (David Marr's three levels — computation, algorithm, implementation — and the contemporary debate over levels integration), the mind-brain explanatory gap and its various proposed bridges, the limits of inference from neural measurement to cognitive function (reverse inference at PhD depth), and the historical contingency of the contemporary cognitive-neuroscience paradigm.

Lesson 2 is Open Research Frontiers in Cognitive Neuroscience — connectomics at frontier depth (the BRAIN Initiative, the Allen Brain Atlas, the Human Connectome Project, the C. elegans complete connectome as historical exemplar and contemporary limit case), single-cell and spatial transcriptomics applied to brain, large-scale brain network dynamics and the dynamical-systems framing, predictive coding and active inference as a frontier theoretical framework (Friston's free energy principle read at the depth of its actual mathematical formulation), and the consciousness research program as a doctoral-research enterprise.

Lesson 3 is Methodological Critique at Expert Depth — the foundational anchor for this chapter: Button et al. 2013 Nature Reviews Neuroscience, Power failure: why small sample size undermines the reliability of neuroscience — read at the depth of its actual statistical argument and its application to the broader field; the reverse inference problem (Poldrack 2006, 2011) at PhD depth; the multiple-comparisons problem in fMRI and the Eklund 2016 PNAS cluster-correction crisis at design-decision depth; the Marek et al. 2022 Nature sample-size analysis for brain-wide association studies; the voodoo-correlations critique (Vul, Harris, Winkielman, and Pashler 2009 Perspectives on Psychological Science); preregistration, registered reports, and the Open Science Collaboration replication work; the broader Ioannidis 2005 framework's application to cognitive neuroscience (and the lateral reference to Food Doctorate Lesson 3).

Lesson 4 is Theoretical Frameworks in Cognitive Neuroscience — predictive processing and active inference (Karl Friston's free energy principle) versus the traditional information-processing framework, presented at theoretical-debate depth; the consciousness theory landscape — Integrated Information Theory (Tononi 2004, 2016), Global Workspace Theory (Baars 1988, Dehaene and colleagues), Higher-Order Theories (Rosenthal, Lau), and the recent adversarial collaboration findings (Cogitate Consortium 2025); the hard problem of consciousness (Chalmers 1995) and the philosophical literature on what it means for science; the embodied-cognition, extended-cognition, and 4E-cognition frameworks (Clark, Chemero, Thompson) versus the computational-cognitivist mainstream; the framework debate as a curriculum subject rather than a tribal commitment.

Lesson 5 is The Path Forward and Original Research Synthesis — open research problems in cognitive neuroscience at PhD depth, the methodological infrastructure the field most needs, the basic-science-to-clinical-practice translation pipeline's failure modes in mental health specifically (Insel 2009 and 2022, the RDoC framing, the persistent gap between mechanistic neuroscience and clinical psychiatry), the methodological-evidence-threshold framework (Master's) applied at doctoral neuroscience research-design depth, and the integration of the chapter's frameworks into the doctoral student's research-question selection and study-design discipline. The Turtle's Cognition position held, deepened to research-track responsibility.

The Turtle is in no hurry. Begin.

---

Lesson 1: The Epistemology of Cognitive Neuroscience

Learning Objectives

By the end of this lesson, you will be able to:

• Articulate, at the level of the field's structural conditions and disciplinary history, why cognitive neuroscience as a knowledge-producing enterprise differs from older biomedical sciences, and identify the methodological and epistemological consequences of those differences
• Read David Marr's three-levels framework (computational, algorithmic/representational, implementational) at the depth of its original Vision (1982) articulation, and identify the contemporary debates over levels integration in cognitive neuroscience
• Articulate the mind-brain explanatory gap at the depth at which contemporary philosophy of mind engages with it — distinguishing the ontological question (is the mind reducible to the brain) from the explanatory question (can our current scientific tools, given current methods, produce a satisfying explanation of cognitive function from neural mechanism) from the methodological question (what kinds of inferential bridges does the field actually deploy between neural measurement and cognitive function)
• Read the reverse inference problem at doctoral depth — Poldrack 2006 Trends in Cognitive Sciences and Poldrack 2011 Neuron — and identify what conditions are required for a reverse inference to be epistemologically warranted, what the field's typical practice does and does not satisfy of those conditions
• Engage the historical contingency of the contemporary cognitive-neuroscience paradigm — recognizing that the contemporary configuration of neuroimaging-driven cognitive neuroscience emerged from specific historical and technological conditions, could have been configured differently, and may be reconfigured by the field's coming methodological and theoretical developments

Key Terms

Why Begin a Doctoral Chapter with Epistemology

A doctoral chapter on cognitive neuroscience does not begin with the substantive content of cognitive neuroscience. It does not even begin with the methodology, though methodology is where the chapter has its center of gravity. It begins with the epistemology, because at this level of study you are not learning what cognitive neuroscience says — you have learned that — and you are not even only learning how cognitive neuroscience knows what it says — you have learned that at Master's depth too — you are learning what kind of knowing the field engages in, what kind of object that knowing produces, and what the structural conditions of that knowing are. Doctoral engagement with any field begins here, and cognitive neuroscience in particular requires it — the field has a uniquely complex relationship to its parent disciplines (psychology, philosophy of mind, biology, computer science, medicine), uniquely complex inferential bridges between its primary measurements and its claims of interest, and a uniquely public set of theoretical commitments that make its inferential structure especially worth examining.

Cognitive neuroscience as a field is structurally young. The Bachelor's chapter walked you through its emergence — from Hubel and Wiesel's single-unit recordings in cat visual cortex in the 1960s through PET imaging in the 1980s and the Ogawa BOLD signal in 1990 to the contemporary multi-modal computational-imaging-and-genetics research enterprise. The field is approximately three decades old at its contemporary scale and configuration. Its parent disciplines — psychology, neurology, neuroanatomy, philosophy of mind — are older and have shaped the questions cognitive neuroscience inherits. The field's working tools — fMRI, EEG, MEG, intracranial recording, optogenetics in animals, computational modeling — each have their own measurement-error structures, their own inferential warrants, and their own historical trajectories. A doctoral reader engages all of this with awareness.

The Turtle's posture in this lesson is the same posture the Turtle has held throughout the curriculum: patient, methodical, slow and deep. The lesson is not designed to settle any of the philosophical or epistemological questions it raises. It is designed to introduce the questions in a form that the doctoral reader will recognize when they encounter them in the published literature, will understand when they shape research-design decisions, and will engage with the seriousness they deserve as the doctoral student goes on to do original work.

Marr's Three Levels and the Levels-Integration Problem

David Marr's 1982 Vision introduced what has become the canonical organizing framework for cognitive neuroscience. The framework distinguishes three levels of analysis for any information-processing system [1].

The computational level asks: what is the system computing, and why? For visual recognition, the computational-level description is, roughly, "the system computes a representation of objects in the visual scene that supports recognition, navigation, action selection, and learning." The computational level is concerned with the function of the system in the abstract — what it does and why it does that thing, given the structure of the input and the demands of the task.

The algorithmic / representational level asks: how does the system perform the computation? With what representations and what procedures? For visual recognition, the algorithmic-level description specifies the representational primitives (edges, contours, surfaces, objects), the algorithmic operations (edge detection, surface integration, object identification), and the temporal sequence of these operations. The algorithmic level is concerned with how the computation is realized in principle, independent of any specific physical substrate.

The implementational level asks: how is the algorithm physically realized? In the brain, the implementational level concerns specific neural circuits, specific cell types, specific neurotransmitter systems, specific anatomical pathways. For visual recognition, the implementational-level description includes V1 simple and complex cells, the dorsal-ventral stream distinction, inferotemporal cortex object selectivity, and the specific neural mechanisms by which these areas perform their algorithmic roles.

Marr's framework is not a hierarchy in which one level reduces to the next. It is a tripartite analysis in which each level provides a different kind of explanatory information about the same system. A complete cognitive-neuroscience account of visual recognition would integrate findings at all three levels into a coherent picture in which the computational, algorithmic, and implementational descriptions cohere. This integration is the levels-integration problem, and it is one of the field's defining intellectual projects.

The contemporary debate over levels integration is substantial. The integration project is rarely fully successful for any cognitive function. We have detailed implementational knowledge of some cortical regions (V1 single-unit physiology) without a complete algorithmic account of the operations they perform; we have detailed algorithmic theories of some cognitive functions (Bayesian models of perception) without complete implementational accounts; we have computational-level accounts of some cognitive functions that the algorithmic and implementational levels have not yet caught up with. Some researchers (e.g., the Tom Griffiths laboratory work) argue that the computational level provides the right starting point and that algorithmic and implementational details should follow from computational analysis [2]. Others (e.g., neural-network-implementation-first approaches) argue that algorithmic-and-implementational details constrain the computational level and that the levels are best worked from the bottom up. Others again (e.g., the recent dynamical-systems framing of motor cortex by Mark Churchland and Krishna Shenoy) argue that Marr's levels are not the right organizing framework at all for some neural phenomena, which may not have a clean computational/algorithmic description [3][4].

The doctoral reader engages this debate as an active intellectual landscape. Marr's framework remains the dominant organizing structure for most of the field's published work, but it is not unchallenged, and the methodological and theoretical commitments a doctoral researcher makes within this framework substantially shape what their original research will look like.

The Mind-Brain Explanatory Gap

The mind-brain explanatory gap, named by Joseph Levine in 1983 [5] and extensively engaged by the contemporary philosophy-of-mind literature, is the condition in which a satisfactory explanation of cognitive or phenomenal experience from neural mechanism appears unavailable on currently understood principles. The gap is explanatory rather than ontological — Levine and most contemporary philosophers do not argue that the mental is irreducible to the physical in some metaphysically deep way; they argue that the explanatory project of accounting for mental in terms of physical encounters specific challenges that other reductive explanations in the natural sciences do not face in the same way.

Three distinctions are worth holding clearly at doctoral depth.

The ontological question asks: are mental events, properties, or processes anything over and above physical (neural) events, properties, or processes? Most contemporary philosophers and neuroscientists hold a non-dualist position — the mental is somehow constituted by, or supervenient on, the physical. The contemporary mainstream view is that mental properties supervene on neural properties: no change in mental state without a change in neural state. Supervenience does not require reduction, but it does require dependence.

The explanatory question asks: given that the mental is somehow physically realized, can our current science actually produce a satisfying explanation of mental phenomena from neural mechanism? Here the picture is much more complicated. We have substantial neural explanations of specific cognitive functions — visual feature detection, working memory maintenance, reinforcement-learning prediction errors. We have substantially less for the phenomenal character of conscious experience, for the global integration of cognition across modalities, and for the unity of the perceiving subject. The explanatory question is the question of how good our explanations actually are, not the question of whether the entities being explained are reducible in principle.

The methodological question asks: what inferential bridges does the field actually deploy between its primary measurements (neural activity, behavior, computational simulation) and its claims of interest (cognitive function, mental state, phenomenal experience)? The methodological question is more tractable than the philosophical ones because it is amenable to empirical examination. The reverse-inference problem (next section) is one of the methodological-question structures.

David Chalmers in 1995 [6] articulated a stronger version of the explanatory question that has become known as the hard problem of consciousness. Chalmers distinguished the "easy problems" of consciousness — explaining specific cognitive functions like attention, working memory, perceptual integration — from the "hard problem" of explaining why physical processing is accompanied by phenomenal experience at all. On Chalmers's analysis, even a complete neuroscientific account of all of cognition would not by itself answer the question of why there is something it is like to undergo that cognition. The hard problem is structurally distinct from the explanatory gap because it concerns the existence of phenomenal experience as such.

The hard problem is contested. Some philosophers and neuroscientists (e.g., the Daniel Dennett and Patricia Churchland traditions, the more recent Andy Clark predictive-processing literature) argue that the hard problem dissolves under sufficient scientific progress — that what looks now like an irreducible gap will close as the field develops, much as vitalist objections to biological reduction closed in the early twentieth century. Others (e.g., the David Chalmers and Frank Jackson traditions) argue that the hard problem reflects a structural feature of phenomenal experience that no further empirical progress will close. Others again (e.g., Galen Strawson, Philip Goff, and the contemporary panpsychism literature) argue that the hard problem is best addressed by revising our background metaphysics rather than by extending empirical neuroscience [7].

The Turtle's doctoral posture on the hard problem is the same posture the Bear takes on the carbohydrate-insulin/energy-balance debate in Food Doctorate: read each position's strongest case in primary form, recognize that competing positions can be consistent with the available evidence, identify what would discriminate between positions, and engage descriptively. The hard problem is the curriculum content, not the conclusion. Doctoral students who do research that touches consciousness must navigate this terrain whether they intend to or not; navigating it with awareness is preferable to inheriting commitments from one's training program without examination.

Reverse Inference at Doctoral Depth: Poldrack 2006 and 2011

The reverse-inference problem is the most consequential single methodological-epistemological issue in fMRI-based cognitive neuroscience. Doctoral students must understand it at structural depth.

The structure is the following. In a typical fMRI study, the researcher manipulates a cognitive task (presents stimuli, asks the participant to perform an operation, presents a control condition) and measures brain activity (BOLD signal). The forward inference is: when participants perform task A versus task B, brain region X is more active in A than in B. The data support the forward inference directly: the contrast of A versus B is the experimental design's primary product.

The reverse inference is: because brain region X is more active in task A than in task B, and because brain region X is known from prior literature to be involved in cognitive function F, therefore task A engages cognitive function F. Reverse inference moves from the brain-activity pattern to the cognitive function. It is the inferential direction that most popular and some published interpretations of fMRI findings depend on. It is also the inferential direction that is most often unwarranted under the conditions in which it is deployed.

Russell Poldrack's 2006 Trends in Cognitive Sciences paper articulated the problem at Bayesian depth [8]. The warrant for reverse inference depends on the conditional probabilities: P(cognitive function F | activity in region X) and the relevant prior probabilities. By Bayes's theorem:

P(F | X) = [ P(X | F) × P(F) ] / P(X)

That is: the probability that task A engages cognitive function F, given activity in region X, depends on (1) how often region X is active when F is engaged (the forward conditional, typically what the prior literature has established); (2) the base rate of F across all studied tasks (the prior); and (3) the overall activation rate of region X (the marginal, integrated across all studied tasks). Reverse inference is warranted to the degree that P(X | F) is high and P(X | not F) is low — that is, region X is reliably active when F is engaged and reliably inactive when F is not. Many fMRI papers infer reversely from region X's activity to cognitive function F when P(X | not F) is in fact substantial — that is, region X is active across many tasks, not just F-engaging tasks.

The amygdala is the canonical example. Many fMRI papers, especially popular-press reports of fMRI papers, interpret amygdala activity as evidence of fear, anxiety, or emotional processing. The amygdala is in fact active across a wide range of tasks — emotional, attentional, learning-related, value-based, motivationally salient — and is not specifically active during fear. The conditional probability P(amygdala active | fear) is substantial, but so is P(amygdala active | not fear). Reverse inference from amygdala activity to fear is therefore weak. Poldrack's analysis demonstrated this empirically and called for the field to constrain reverse-inference practice [8].

Poldrack 2011 Neuron extended the framework [9], introducing the Cognitive Atlas project, which formalizes a cognitive ontology relating tasks, cognitive functions, and brain regions. The Cognitive Atlas project is structurally aimed at giving the field a shared vocabulary in which reverse-inference claims can be evaluated at scale. The project has been productive but also exposes a deeper problem: the field's cognitive ontology is inconsistent. Many task paradigms do not cleanly map onto single cognitive functions, and many cognitive functions are operationalized by multiple non-equivalent task paradigms across the literature [10]. The reverse-inference problem rests on a cognitive ontology that itself contains substantial inconsistency.

A doctoral reader of an fMRI paper engages reverse-inference claims explicitly. Does the paper make a reverse-inference claim? If so, on what cognitive ontology, with what selectivity assumptions, with what consideration of the marginal? Are alternative cognitive functions that could account for the same brain-activity pattern considered and ruled out? The discipline of running this analysis in real time as one reads the literature is the discipline that distinguishes doctoral engagement from earlier modes of engagement with the imaging literature.

Historical Contingency of the Contemporary Cognitive-Neuroscience Paradigm

The cognitive neuroscience paradigm of 2026 — neuroimaging-driven, computationally augmented, increasingly integrated with genomics and large-scale cohort data, organized around the academic publication-and-grant cycle, working under Marr's three-levels framework with predictive-processing pressure from one direction and dynamical-systems pressure from another — is a historically contingent configuration. It emerged from specific historical and technological conditions. It could have been configured differently. It may be reconfigured by the field's coming methodological and theoretical developments. A doctoral reader engages this contingency with awareness rather than naturalizing the contemporary configuration as the way cognitive neuroscience must be.

The technological conditions that shaped the field's emergence are well known. The PET imaging revolution of the 1980s, the fMRI revolution following the Ogawa BOLD signal in 1990 [11], the explosion of in vivo human brain imaging that followed, the parallel development of optogenetics in the 2000s and chemogenetics in the 2010s, the more recent expansion of intracranial recording from epilepsy-monitoring populations and the development of brain-machine interfaces have all shaped which research questions became answerable at scale and which remained out of reach. The field's research-question structure in 2026 is largely the research-question structure that fMRI made tractable at large sample size.

The disciplinary conditions are equally consequential. Cognitive neuroscience as an organized field emerged from the convergence of cognitive psychology (which had developed information-processing models of cognition without strong neural grounding), neuropsychology (which had developed lesion-based inferences without strong cognitive models), behavioral neuroscience (which had developed animal-model neural mechanisms without strong cognitive translation), and computational neuroscience (which had developed mathematical models without strong empirical anchor in either cognition or neural mechanism). The convergence of these parent disciplines into a unified cognitive-neuroscience program shaped the field's intellectual structure in ways that remain consequential.

The historical contingency of these conditions means that the field's current questions, methods, and findings are partly artifacts of the conditions under which the field emerged. Original doctoral research is an opportunity to recognize the contingency and, where possible, work outside its strongest constraints. Researchers asking questions that fMRI cannot easily answer, theoretical commitments that the dominant frameworks do not easily accommodate, methodological tools that the discipline has not yet incorporated — they are often the researchers who reconfigure the field for the next generation.

The Turtle's posture on this is the patient long-view posture. Cognitive neuroscience in 2050 will not look like cognitive neuroscience in 2026; the doctoral research being designed now will, in part, determine what it looks like. Choose your contribution with awareness of the long view.

What This Lesson Built

You should leave this lesson able to do something specific: read a piece of cognitive neuroscience research, or a piece of cognitive neuroscience commentary, or a piece of cognitive-neuroscience-adjacent philosophy, and place it in the field's structural-epistemological context. What Marr-level claim is the work making, on what inferential warrants, with what consideration of reverse inference, against what historical configuration of the field, within what theoretical framework, with what awareness of the levels-integration problem? This is the doctoral reading. It is the precondition of doctoral research-question selection, doctoral study design, and doctoral contribution to the field.

The remainder of the chapter rests on this lesson. Lesson 2 moves to the open research frontiers where the field is currently doing its most interesting work. Lesson 3 moves to the methodological tools and the foundational anchor — the Button 2013 Nature Reviews Neuroscience power-failure analysis — at the depth needed for doctoral methodological engagement. Lesson 4 moves to the theoretical-framework debates. Lesson 5 moves to the path forward and to the research-design discipline that integrates the chapter's frameworks. None of those make sense without the structural reading developed here.

Lesson Check

1. Marr's three-levels framework distinguishes computational, algorithmic/representational, and implementational analyses of an information-processing system. Define each. Apply the framework to a specific cognitive function of your choosing (visual object recognition, working memory, reinforcement learning, decision-making) and articulate, for that function, what is well-understood at each level and what remains open. Then articulate one open question about the integration across levels for the function you chose.
2. The mind-brain explanatory gap (Levine 1983) and the hard problem of consciousness (Chalmers 1995) are structurally distinct but related problems. Articulate the distinction. For a doctoral researcher whose primary research program is in cognitive neuroscience proper (not philosophy of mind), how should the existence of these problems shape research-question selection, research interpretation, and communication of findings to the broader scientific and public audiences?
3. Reverse inference is warranted to the degree that the relevant Bayesian conditionals support it: P(X | F) is high and P(X | not F) is low. The amygdala-fear inference is a canonical example of a reverse inference that is weak under the actual conditionals because P(amygdala active | not fear) is substantial. Identify two additional examples of reverse-inference claims in the published or popular cognitive neuroscience literature that you would evaluate as weak under the same analysis, and articulate why.
4. Russell Poldrack's Cognitive Atlas project formalizes the field's cognitive ontology and exposes its inconsistencies. Identify one inconsistency — a cognitive function operationalized by multiple non-equivalent task paradigms, or a task paradigm interpreted as engaging multiple distinct cognitive functions — that you have encountered in the published literature, and articulate why the inconsistency is consequential for the reverse-inference claims that depend on it.
5. The contemporary cognitive-neuroscience paradigm is historically contingent. Articulate two technological conditions and two disciplinary conditions that shaped its emergence. Identify one element of the contemporary paradigm that you anticipate might be reconfigured by methodological or theoretical developments in the next decade, and articulate what the reconfiguration would require.

---

Lesson 2: Open Research Frontiers in Cognitive Neuroscience

Learning Objectives

By the end of this lesson, you will be able to:

• Characterize the contemporary connectomics research program at frontier depth — including the BRAIN Initiative, the Human Connectome Project, the Allen Brain Atlas, MICrONS, and the C. elegans complete connectome as historical exemplar and contemporary limit case — and articulate what each connectomic level (synaptic, mesoscale, macroscale) is positioned to answer
• Read the single-cell and spatial transcriptomics frontier in brain research at the depth at which Allen Brain Atlas, Lein et al. work, and Tasic et al. cortical-cell-type characterization have framed the cell-type-identity question, and articulate what doctoral research is positioned to contribute at this frontier
• Characterize the large-scale brain network dynamics research program (default-mode network, salience network, fronto-parietal control network) at frontier depth and the dynamical-systems framing of cortex (Churchland, Shenoy, Vyas) as an alternative or complement to information-processing framings
• Read Karl Friston's free energy principle and the predictive processing / active inference framework at the depth of the actual mathematical formulation, distinguishing the empirically testable claims from the principles that operate at the level of theoretical commitment, and articulate where the framework has been productively applied and where its claims have been contested
• Engage the consciousness research program as a doctoral-research enterprise — distinguishing the neural-correlate-of-consciousness research program (Crick and Koch 1990 onward) from the consciousness-theory program (IIT, GWT, HOT, recent adversarial collaborations) — and identify how a doctoral student interested in consciousness research would position their work

Key Terms

The Connectomics Frontier

The connectomics research program asks a question of substantial doctoral consequence: given a complete map of an organism's neural connections, what could we know about its cognitive and behavioral capacities that we cannot know without one? The question is empirical in form but its full answer is methodological and theoretical: complete connectomic knowledge would constrain the algorithmic and implementational explanations of cognitive function but would not by itself produce them, and the distinction between connectome-as-data and connectome-as-explanation is one of the most important conceptual moves the doctoral student must absorb.

The historical exemplar is the C. elegans connectome. White, Brenner, Southgate, and Thompson published in 1986 Philosophical Transactions of the Royal Society B the complete neuronal connectivity of C. elegans (302 neurons in the hermaphrodite, all synapses) [12]. Subsequent decades have refined the connectome [13], characterized cell-type-specific gene expression across the connectome, and built increasingly sophisticated computational models. C. elegans has had a complete connectome for forty years. The field does not have a complete behavioral or cognitive model of C. elegans. The connectome is necessary but not sufficient for that model. This is the canonical doctoral lesson on what complete connectomic knowledge does and does not deliver.

The contemporary mammalian connectomics enterprise extends across multiple scales. The macroscale connectome — region-to-region white-matter pathways — is mapped routinely by diffusion MRI in human populations; the Human Connectome Project (HCP) imaged approximately 1,200 healthy young adults with multimodal MRI as a foundational resource for macroscale connectomic analysis [14]. The HCP dataset has been extended by lifespan HCP (HCP-Aging, HCP-Development) and by disease-focused HCP projects covering Alzheimer's, anxiety and depression, psychosis, and substance use [15]. The UK Biobank imaging substudy (approximately 100,000 participants imaged) extends macroscale connectomics at unprecedented sample size [16]. Macroscale connectomic data is plentiful, methodologically mature, and increasingly integrated with genomic, behavioral, and clinical data.

The mesoscale connectome — cell-type-to-cell-type and region-to-region projections — has been advanced substantially by the Allen Mouse Brain Connectivity Atlas, which mapped axonal projections from defined source regions to their targets across the mouse brain [17][18]. The mesoscale picture in mouse is now substantially complete and increasingly available; mesoscale connectomic data in non-human primates is more limited but growing.

The microscale / synaptic connectome — individual neuron-to-neuron synaptic connections — is mapped by serial-section electron microscopy with computational reconstruction. The MICrONS consortium, a BRAIN Initiative project, produced in the 2020s the largest-scale electron-microscopy reconstruction of a mammalian cortical volume: approximately 1 cubic millimeter of mouse visual cortex, with all neurons reconstructed and their synaptic connections mapped, combined with functional characterization by two-photon calcium imaging [19][20]. Comparable efforts are underway at the Janelia Research Campus and at the Drosophila FlyEM project, which produced the first complete adult Drosophila brain connectome (FlyWire) in 2024 [21][22]. The doctoral student in cognitive neuroscience increasingly engages connectomic data as primary empirical material and must understand the methodological and analytical landscape that frames it.

What connectomic knowledge enables and what it does not enable is the doctoral question. Complete connectomic data, by itself, does not deliver a cognitive function. The functional consequence of a given connection pattern depends on the dynamical properties of the neurons involved (cell type, intrinsic excitability, ion-channel composition), the synaptic strengths and dynamics (which are typically not directly readable from connectomic data alone), the activity patterns the system actually exhibits in vivo, and the cognitive or behavioral task the system is engaged in. A connectome is a structural constraint on possible function, not a function. Doctoral research that integrates connectomic data with functional, transcriptomic, and behavioral data is the structurally productive direction.

Single-Cell and Spatial Transcriptomics in Brain

The single-cell and spatial transcriptomics frontier has substantially reshaped the cell-type-identity question in cognitive neuroscience over the past decade. Where the older field characterized cortical neurons by morphology and electrophysiology into approximately a dozen broad categories (pyramidal cells of various layers, basket cells, chandelier cells, Martinotti cells, others), the single-cell transcriptomic frontier has revealed that within each category there are typically many transcriptomically distinct subtypes, each with distinct gene expression profiles, distinct connectivity preferences, distinct intrinsic properties, and distinct functional roles.

The Allen Brain Cell Type Atlas is the contemporary foundational resource. Tasic et al. 2018 Nature characterized approximately 130 transcriptomically distinct cell types in mouse visual cortex [23]; Hodge et al. 2019 Nature extended the characterization to human cortex [24]. Subsequent atlases have characterized cell types across the mammalian nervous system at finer and finer resolution. The Yao, van Velthoven, and colleagues 2023 Nature paper produced a whole-mouse-brain cell-type atlas integrating transcriptomic identity with spatial location [25].

The methodological consequence for cognitive neuroscience is substantial. Cell-type identity is no longer a coarse category that can be invoked casually; cell types are increasingly precise empirical entities with specific gene expression, specific connectivity, and specific function. Cell-type-targeting tools — Cre driver lines, AAV vectors with cell-type-specific promoters, intersectional genetic strategies — have made cell-type-resolved functional manipulation routine in mouse research. The translation to human neuroscience, where direct cell-type-resolved manipulation is not generally possible, is more limited but increasingly informed by mouse-to-human cell-type homology analyses.

The theoretical consequence is equally substantial. The classical reduction of cognitive function to brain region — visual cortex does vision, hippocampus does memory, prefrontal cortex does executive function — was always known to be coarse. The cell-type-resolved picture replaces it with a substantially finer story in which the same brain region contains many distinct cell types performing distinct computational and behavioral roles. The doctoral student must develop fluency with this finer story.

The spatial transcriptomics extension — preserving anatomical position while measuring transcriptomic identity — is the integrative methodology that connects cell-type and connectomic frameworks. MERFISH, Slide-seq, Visium, and other spatial transcriptomic platforms have matured rapidly in the early 2020s [26][27][28]. The doctoral career-research opportunity in this space is substantial: doctoral students who develop computational and experimental fluency across single-cell, spatial transcriptomic, and connectomic data are positioned to do the integrative work the field most needs over the next decade.

Large-Scale Brain Network Dynamics

The large-scale brain network research program has substantially shifted the field's framing of cognitive function over the past two decades. The classical brain-region-by-cognitive-function mapping has been complemented (and in some accounts replaced) by a brain-network-by-cognitive-function mapping in which distributed networks of regions co-activate or de-activate together in characteristic patterns supporting cognitive operations.

The Marcus Raichle et al. 2001 PNAS discovery of the default mode network (DMN) is the canonical foundational finding [29]. Raichle and colleagues observed that a specific set of brain regions — medial prefrontal cortex, posterior cingulate, precuneus, angular gyrus, parahippocampal cortex — was more active during rest than during externally focused cognitive tasks. The pattern was reliable across individuals, characteristic, and (initially surprisingly) inverse to the pattern of activation during attention-demanding tasks. Buckner et al. 2008 Annals of the New York Academy of Sciences extended the DMN characterization and connected it to self-referential thought, autobiographical memory, future thinking, and theory-of-mind cognition [30].

The DMN finding catalyzed a broader research program on intrinsic brain networks. Resting-state fMRI — measuring spontaneous BOLD fluctuations at rest, independent of any specific task — became a primary methodology of the field. The Yeo, Krienen, et al. 2011 Journal of Neurophysiology network parcellation identified approximately seven major intrinsic networks in the human cortex (DMN, salience, fronto-parietal control, dorsal attention, ventral attention, somatomotor, visual) that were reliable across individuals and connected to characteristic cognitive functions [31]. The Power, Cohen, et al. 2011 Neuron parcellation extended the analysis [32]. Subsequent work has characterized the developmental trajectory of these networks, their alteration in psychiatric and neurological conditions, and their relationship to white-matter connectomic structure.

The large-scale brain network framing has been productive but is not unchallenged. The networks are statistical patterns rather than discrete anatomical entities; their boundaries depend on the analytic methodology used to identify them; the relationship between within-network and between-network connectivity is the subject of active methodological debate; and the cognitive-function interpretation of any given network rests on reverse-inference structures that are themselves under question. The doctoral reader engages the network literature with the methodological caution Lesson 1 developed.

The dynamical-systems neuroscience framing is, in some accounts, an alternative or complement to the network framing. The Mark Churchland and Krishna Shenoy laboratory work on motor cortex over the past two decades has reframed motor cortical function in terms of high-dimensional population-level dynamics — the evolution of neural state trajectories through high-dimensional space — rather than in terms of feature-coding by individual neurons [33][34]. The Churchland-Shenoy framing has been extended to broader cognitive function in active research and constitutes a substantial theoretical alternative to information-processing framings of cortex.

The Krishna Shenoy laboratory at Stanford and the Mark Churchland laboratory at Columbia, working in collaboration and with multiple extensions to other laboratories, have produced a substantial body of work demonstrating that motor preparation, motor execution, and certain cognitive operations are better described as low-dimensional dynamical trajectories through neural state space than as feature-coding by individual neurons [35][36]. The framing has substantial implications for how the field thinks about the relationship between neural activity and cognitive function: it suggests that the relevant explanatory units may be dynamical objects (manifolds, trajectories, fixed points) rather than functional units (regions, networks, feature detectors). Whether the dynamical-systems framing scales to higher cognitive functions or remains specific to motor and motor-adjacent cortex is an active research question.

The Free Energy Principle and Predictive Processing at Frontier Depth

Karl Friston's free energy principle (FEP) and the closely related predictive processing framework constitute one of the most influential and most contested theoretical frameworks in contemporary cognitive neuroscience. The doctoral reader engages it at the depth of its actual mathematical formulation, distinguishes its empirically testable claims from its more abstract principle status, and identifies where it has been productively applied and where its claims have been contested.

The FEP, in compact form, holds that biological systems minimize a quantity called variational free energy over time. Variational free energy is a mathematical bound on the negative log-likelihood (surprise) of sensory input given a generative model — the model the system has built of the causes of its sensory input. Minimizing free energy is mathematically equivalent, under specific assumptions, to maximizing the evidence for the generative model (Bayesian model evidence). Friston's principle holds that any system that persists through time, in the face of an environment that would otherwise dissipate it, must operate to minimize free energy [37][38].

Applied to brain, the FEP implies that perception, action, and learning are all forms of active inference operating to minimize prediction error against the generative model. Perception is inference about the causes of sensory input. Action is selection of behaviors that the agent expects will reduce future prediction error. Learning is updating of the generative model to reduce future prediction error. Attention is precision-weighting of prediction errors. Emotion, motivation, and decision-making can all be reframed within the active-inference vocabulary.

The framework's reach is substantial. Friston and colleagues have applied it to perception [39], action [40], psychiatric conditions [41], embodied cognition [42], and the philosophical foundations of cognitive science [43]. Andy Clark's Surfing Uncertainty (2016) and The Experience Machine (2023) provide the philosophical articulation [44][45]. Anil Seth's Being You (2021) extends the framing to consciousness research [46]. The framework has been productively applied across a range of domains and is widely taught in contemporary cognitive science.

The contested questions are equally substantial. The FEP's status as a principle — a universal characterization of biological systems — versus a framework — a productive way to think about specific cognitive functions — is itself contested. Critics including Colombo and Wright [47], Bruineberg, Kiverstein, and Rietveld [48], and Sun and Firestone [49] have argued at varying depths that the FEP's universal claims rest on assumptions that, when examined, either become trivially true (because the formal mathematics allows reinterpretation of any system as free-energy-minimizing under suitable transformation) or empirically false (because actual biological systems do not in fact minimize the specific quantity the FEP names). The methodological consequence is that the framework's predictions, in many applications, are not uniquely identifiable — multiple incompatible generative models can produce the same observable behavior, and the framework's empirical content is correspondingly diffuse.

The doctoral reader approaches the FEP with the underdetermination posture Lesson 1 developed. The framework has been productive — many specific empirical findings have been clarified by predictive-processing analyses, the active-inference framing has motivated specific experiments, and the integrative reach across perception, action, and learning is genuinely useful. The framework has also been over-claimed in some contexts as a universal theory of brain and mind, and the empirical content of its strongest claims is harder to pin down than its advocates sometimes communicate. Doctoral research that engages the FEP carefully — neither dismissing it nor accepting it as universal — is well-positioned to advance the field's understanding of when and where the framework is genuinely explanatory.

The Consciousness Research Program as Doctoral Enterprise

The consciousness research program is, at present, one of the most active and most theoretically structured doctoral-research areas in cognitive neuroscience. Doctoral students with interest in consciousness research enter a field with substantial empirical infrastructure, multiple competing theoretical frameworks, and a recent history of methodologically ambitious work.

The neural correlate of consciousness (NCC) research program was articulated as a research target by Francis Crick and Christof Koch in 1990 [50]. The NCC is defined as the minimal set of neural events sufficient for a specific conscious experience. The research program asks, for any given conscious experience, what neural events are jointly sufficient and how. The program has been pursued across multiple paradigms — binocular rivalry, masking, attentional blink, anesthesia, brain injury — and has accumulated substantial empirical findings [51][52]. Whether the successful execution of the NCC research program would close the hard problem (Lesson 1) is itself contested; the NCC program is well-defined empirically regardless of the philosophical question.

The consciousness-theory research program is the more theoretically structured enterprise. Three major theoretical families currently dominate.

Integrated Information Theory (IIT), articulated by Giulio Tononi since 2004 [53] and substantially extended in IIT 4.0 [54], holds that consciousness is identical to integrated information (Φ) — a measurable mathematical property of a system's causal structure characterizing the degree to which the whole is more than the sum of its parts in causal terms. IIT makes specific empirical predictions: the cerebral cortex should support high Φ; the cerebellum, despite its substantial neuron count, should support low Φ (because of its more independent processing architecture); anesthesia should reduce Φ; brain injury should preserve consciousness in proportion to preserved Φ. IIT has been productive in motivating empirical work, and the perturbational complexity index (PCI) developed by the Massimini and Tononi laboratories operationalizes a Φ-related measure that distinguishes conscious from unconscious states with substantial accuracy across anesthesia, vegetative state, minimally conscious state, and brain-injury populations [55][56].

Global Workspace Theory (GWT), articulated by Bernard Baars in 1988 [57] and developed into the Global Neuronal Workspace (GNW) model by Stanislas Dehaene, Jean-Pierre Changeux, and colleagues [58][59], holds that consciousness corresponds to the broadcast of information across a global workspace — a distributed cortical system, particularly involving fronto-parietal regions, that makes information available to a wide range of cognitive systems (working memory, verbal report, action selection, memory consolidation). GWT makes specific empirical predictions: the P3b event-related potential at approximately 300 ms post-stimulus, the late ignition dynamics of conscious access, the dependence of consciousness on top-down attentional processing. The Dehaene laboratory work has substantially advanced the empirical case [60].

Higher-Order Theories (HOT), articulated in a family by David Rosenthal, Hakwan Lau, Joseph LeDoux, and others [61][62], hold that a mental state is conscious in virtue of being represented by a higher-order mental state — a thought about, or perception of, the first-order state. HOT predictions differ from IIT and GWT in identifiable ways: HOT predicts that consciousness should depend on prefrontal cortical regions that implement the higher-order representation, that conscious access without higher-order representation should be impossible, and that disruption of higher-order representation (by TMS, lesion, or specific cognitive manipulation) should disrupt consciousness independent of first-order processing.

The Cogitate Consortium adversarial collaboration, with results published in 2025 [63], represents a methodologically important step. Cogitate brought together proponents of IIT and GWT to design experiments together — with prespecified hypotheses, prespecified analyses, and prespecified discrimination criteria — that would adjudicate between the theories' predictions. The results, broadly: some predictions of each theory were supported, some predictions of each theory were not supported, the theories were not cleanly decided in favor of one or the other, and the methodology of adversarial collaboration itself emerged as a substantial methodological advance for theoretically contested fields. The Cogitate Consortium model is being extended to additional theoretical contrasts in consciousness research and to other contested areas in cognitive neuroscience [64].

A doctoral student interested in consciousness research enters a field with substantial empirical infrastructure, multiple competing theoretical frameworks (each of which has its strongest case and its limits), and a recent methodological advance (adversarial collaboration) that the field is actively integrating. The Turtle's posture on consciousness research is the same posture the Turtle has held throughout: choose your theoretical commitments with awareness, design hypothesis-discriminating experiments where the methodology permits, engage the literature descriptively, and contribute work that the field's diverse theoretical commitments can integrate. The hard problem (Lesson 1) sits over the entire program; you will not solve it in your dissertation, but you will contribute to the field's slow accumulation of empirical and conceptual progress on consciousness as the field defines it.

Frontier Questions a Doctoral Student is Positioned to Engage With

A short list, by no means exhaustive, of frontier questions in cognitive neuroscience that the field's current methodology is in principle capable of addressing and that would constitute meaningful original contribution:

1. The connectome-function gap question. Given increasingly complete connectomic data at multiple scales, what additional measurement and analysis is needed to bridge from structure to function? Specifically: how do connectomic, transcriptomic, dynamical, and behavioral data integrate into functional models with predictive power?

2. The cell-type-to-cognition question. Given the cell-type diversity revealed by single-cell transcriptomics, which cell types are causally relevant to which cognitive functions, and how does cell-type-resolved manipulation in animal models translate to human cognitive function?

3. The large-scale-network-meaning question. The intrinsic networks identified by resting-state fMRI are reliable and reproducible, but their cognitive-function interpretation rests on reverse-inference structures that the methodology of Lesson 3 will examine critically. What experimental designs would close the inferential gap between network statistical structure and cognitive function?

4. The predictive-processing empirical content question. Predictive processing has been productively applied across many domains, but the framework's empirical content is sometimes diffuse. What experimental designs would produce hypothesis-discriminating evidence between predictive processing and alternative framings of specific cognitive functions?

5. The consciousness-theory adjudication question. The IIT, GWT, and HOT frameworks make distinct predictions; the Cogitate Consortium adversarial collaboration is a methodological advance, but many specific theoretical contrasts remain unadjudicated. Original research that contributes to hypothesis-discriminating evidence between theories of consciousness is among the most consequential work the field is currently positioned to do.

6. The clinical translation question. The gap between basic cognitive neuroscience findings and clinical translation in psychiatric and neurological practice remains substantial. Implementation science, biomarker development, and computational psychiatry are areas where doctoral research can contribute work that the clinical field will be able to build on. The Insel and Cuthbert Research Domain Criteria (RDoC) framing [65][66] is one institutional response; whether it succeeds depends on the original research that operationalizes it.

The doctoral career-orienting move is to identify a frontier question — one of these, or a related one — and develop a sustained research program oriented toward it. The Turtle's posture is patient: research questions in this field unfold across years; the work that advances the field is the work that holds steady on a well-chosen question through the long iteration that doctoral research actually requires.

Lesson Check

1. The connectome is necessary but not sufficient for cognitive function — the C. elegans complete connectome, mapped in 1986, has not yielded a complete behavioral model in forty years. Articulate what additional data and analysis is required to bridge from connectome to function. Identify three specific data types (transcriptomic, dynamical, behavioral, others) and the role each plays in the bridging.
2. The Allen Brain Cell Type Atlas reveals that within previously homogeneous cell categories (cortical pyramidal cells, GABAergic interneurons) there are many transcriptomically distinct subtypes. For a cognitive function of your choosing (working memory, attention, reinforcement learning, perception), articulate why cell-type-resolved characterization matters for the function's mechanistic explanation. Identify one specific cell-type-targeted experiment you would design to advance the field's understanding.
3. The dynamical-systems framing of motor cortex (Churchland, Shenoy) reframes neural function in terms of high-dimensional state trajectories rather than feature-coding by individual neurons. Articulate the framing. For a cognitive function beyond motor preparation (working memory, decision-making, perception), what would it mean for the dynamical-systems framing to apply, and what experimental evidence would adjudicate between the dynamical-systems and the feature-coding framings?
4. Karl Friston's free energy principle has been productive in motivating empirical work and has also been criticized for diffuse empirical content. Articulate one specific empirical prediction of the predictive-processing framework that has been productively tested. Articulate one application of the FEP where its predictions are not uniquely identifiable. What does the difference between these two cases suggest about how doctoral researchers should engage the framework?
5. The IIT, GWT, and HOT theories of consciousness make distinct predictions. Identify two empirical predictions on which the theories differ. The Cogitate Consortium adversarial collaboration (results 2025) tested some such predictions and produced mixed results. As a doctoral researcher, what additional adversarial-collaboration design would you propose to advance the field's adjudication? What specific theoretical contrast and what empirical paradigm would your design address?

---

Lesson 3: Methodological Critique at Expert Depth

Learning Objectives

By the end of this lesson, you will be able to:

• Read Button et al. 2013 Nature Reviews Neuroscience, Power failure: why small sample size undermines the reliability of neuroscience, at the depth of its actual statistical argument — including the positive predictive value derivation, the effect-size inflation argument, the systematic-review evidence on median power across neuroscience subfields, and the proposed reforms — and apply the framework to specific scenarios in the cognitive neuroscience literature
• Critique an fMRI study at peer-reviewer depth across the structural constraints of brain imaging — sample size and statistical power, multiple-comparisons correction at the Eklund 2016 PNAS depth, the reverse-inference problem at the Poldrack 2006/2011 depth, the cognitive-ontology question, and the experimental-design and analytic-choice multiplicity
• Read Marek et al. 2022 Nature, Reproducible brain-wide association studies require thousands of individuals, at the depth of its actual analysis — the relationship between sample size, effect size, and reproducibility for brain-behavior association studies — and articulate the consequences for the field's research-design choices
• Read the voodoo-correlations critique (Vul, Harris, Winkielman, Pashler 2009 Perspectives on Psychological Science) at PhD depth and identify how the critique applies (or does not apply) to specific contemporary practices in the field
• Engage the broader replication crisis literature — Open Science Collaboration 2015 Science on psychological replication, the broader Ioannidis 2005 framework as applied to neuroscience, the methodological reforms (preregistration, registered reports, data and code sharing, large pre-planned consortia studies) — and articulate where cognitive neuroscience is in the post-replication-crisis trajectory

Key Terms

The Foundational Anchor: Button et al. 2013 Nature Reviews Neuroscience, Power Failure

The foundational anchor for this Doctorate chapter is Katherine S. Button, John P. A. Ioannidis, Claire Mokrysz, Brian A. Nosek, Jonathan Flint, Emma S. J. Robinson, and Marcus R. Munafò 2013 Nature Reviews Neuroscience — Power failure: why small sample size undermines the reliability of neuroscience [67]. The paper is the most influential single methodological-critique paper specific to neuroscience of the past decade. It applies the broader Ioannidis 2005 framework (Food Doctorate Lesson 3) to neuroscience at field-specific depth and produces specific quantitative estimates of the field's structural problems. Doctoral students in cognitive neuroscience should be able to read the paper at the depth of its actual statistical argument, reproduce the central derivations, and apply the framework to specific scenarios in the published literature.

The structure of the argument runs as follows.

(1) Sample size and power across neuroscience. Button et al. conducted a systematic review of 49 meta-analyses across major neuroscience subfields published in 2011, covering 731 individual studies. For each meta-analysis, they computed median statistical power assuming the meta-analytic summary effect size as the true effect size. The result: median power across the surveyed neuroscience literature was approximately 21%. That is: the typical neuroscience study, if conducted on a true effect of the magnitude the meta-analytic synthesis suggests, would have only about a one-in-five probability of detecting that effect. Most neuroscience studies, in this analysis, were underpowered to detect the effects they reported having detected.

(2) The PPV consequence. Low statistical power has predictable consequences for the positive predictive value of significant findings. Under the Bayesian PPV framework (Ioannidis 2005, Food Doctorate Lesson 3):

PPV = (sensitivity × prior) / [(sensitivity × prior) + (false-positive rate × (1 − prior))]

For a research field with prior probability 0.25 (a substantial fraction of tested hypotheses correspond to true effects), power 0.21 (the Button et al. estimate), and α = 0.05, the predicted PPV is:

(0.21 × 0.25) / [(0.21 × 0.25) + (0.05 × 0.75)] = 0.0525 / (0.0525 + 0.0375) = 0.0525 / 0.090 ≈ 0.58

That is: under these favorable assumptions, about 58% of significant findings correspond to true effects. Under less favorable prior assumptions (prior 0.10), the PPV drops to about 0.32. Under multiplicity-inflated effective α and small bias parameter additions, the PPV drops substantially further. The Button et al. analysis is the neuroscience-specific quantification of the Ioannidis framework, and it is a serious result.

(3) Effect-size inflation. Beyond the PPV consequence, low power has a second structural consequence: among the underpowered studies that nonetheless cross the significance threshold, the observed effect sizes are systematically inflated relative to the true effect. The mathematical intuition is straightforward — for a true small effect, the sampling-distribution realizations that cross the significance threshold are the ones farthest from zero, so the conditional expected effect size given significance is larger than the true effect. The empirical consequence is that an underpowered literature reports inflated effect sizes that subsequent better-powered work attenuates. The replication-failure pattern across multiple fields is consistent with this inflation mechanism — original studies report effects that more powered replications cannot reproduce because the original effects were inflated rather than absent.

(4) The systematic-review evidence. Button et al. extended the analysis to specific subfields, including cognitive neuroscience proper, animal-model neuroscience, and clinical neuroscience. Cognitive neuroscience specifically had median power on the order of 18%. Animal-model neuroscience had median power approximately 31%. The overall picture is one in which a substantial fraction of the field's published literature was, at the time of publication, structurally vulnerable to the PPV and effect-size-inflation consequences the framework predicts. Many subsequent replication-failure analyses are consistent with this prediction.

(5) The proposed reforms. Button et al. proposed several specific reforms: increased sample sizes (often by an order of magnitude or more), pre-registration of hypotheses and analytic plans, replication-emphasizing journal practices, publication of null results, and large-consortium studies that aggregate data across laboratories. The field has substantially adopted these reforms over the past decade, though uptake is uneven across subfields and the structural conditions that produced the original underpowered literature (small grants, fast-paced publication, novelty-rewarded promotion) remain partly in place.

Reading Button et al. 2013 at depth means understanding all five components of the argument and being able to apply them to specific scenarios in the published literature. A doctoral student reading a 2010-era cognitive neuroscience paper with sample size n = 12 or n = 18 (typical fMRI sample sizes of that era) understands what to make of it: a small significant finding from such a study should be interpreted as preliminary evidence of moderate prior credibility, with substantial probability of effect-size inflation and meaningful probability of being false-positive entirely. The doctoral reader does not dismiss such literature wholesale but calibrates confidence to the structural conditions in which the literature was produced.

Brain-Wide Association Studies and the Marek 2022 Nature Sample-Size Result

The Button et al. 2013 framework has been extended and concretized in subsequent work. The single most consequential extension for contemporary cognitive neuroscience is Scott Marek and colleagues 2022 Nature — Reproducible brain-wide association studies require thousands of individuals [68].

The Marek et al. analysis used the Adolescent Brain Cognitive Development (ABCD) study (approximately 11,000 children with neuroimaging and behavioral data), the UK Biobank imaging substudy (approximately 30,000 adults with comparable data at the time of analysis), and the Human Connectome Project to characterize the relationship between sample size, effect size, and reproducibility for brain-wide association studies — studies that correlate brain measurements (resting-state functional connectivity, structural MRI features) with behavioral, cognitive, or psychiatric measures.

The principal findings: at typical neuroimaging sample sizes (n = 25, n = 50, n = 100), brain-behavior associations were unstable — different random subsamples of the same large dataset produced different "significant" associations, often non-overlapping, and the reported effects in any single subsample did not replicate in held-out subsamples. The effect sizes for the true brain-behavior associations in these data were small (correlations typically in the r = 0.10 to r = 0.20 range). Reliable detection of such small effects required substantially larger samples — on the order of thousands rather than tens or low hundreds. The methodological consequence is that a substantial fraction of the small-sample brain-behavior-association literature, while individually published as significant findings, does not survive the better-powered analysis the larger datasets enable.

The Marek et al. result is consequential in several ways. First, it concretizes the Button et al. 2013 argument for a specific high-volume area of the literature (brain-behavior association studies) with specific quantitative estimates of the required sample sizes. Second, it provides positive guidance — large consortium datasets (ABCD, UK Biobank, HCP) and large pre-planned multi-site studies are the appropriate methodology for this question, and the field is increasingly using them. Third, it identifies a specific publication category where the field's accumulated small-sample literature should be treated with substantial skepticism rather than incorporated into the field's working knowledge.

The Marek result has been productively extended. The Genç and colleagues replication using independent data confirmed the central finding [69]. The methodological literature on improving brain-behavior reliability through multi-modal integration, longitudinal designs, and dense individual-subject sampling has accumulated [70]. The doctoral reader of contemporary cognitive neuroscience encounters the Marek result either implicitly (large consortium designs are increasingly the norm for brain-behavior questions) or explicitly (methodology sections that cite Marek and explain how the present study addresses or works within the small-sample limitations).

The Eklund 2016 Cluster-Correction Crisis

The fMRI multiple-comparisons problem has its own specific methodological history that doctoral students should know in detail.

The classical solution to the multiple-comparisons problem in fMRI was cluster-correction at the voxel level. The brain is analyzed at hundreds of thousands of voxels; correcting for multiplicity by Bonferroni would be extremely conservative; cluster-correction instead identifies spatially contiguous clusters of activated voxels and applies a family-wise correction at the cluster level, exploiting the spatial smoothness of the BOLD signal to recover statistical power. The methodology was implemented in the three major fMRI software packages (SPM, FSL, AFNI) under various technical formulations, and was the default analytic procedure for the bulk of the published fMRI literature from the mid-1990s through the mid-2010s.

Anders Eklund, Thomas E. Nichols, and Hans Knutsson 2016 PNAS — Cluster failure: why fMRI inferences for spatial extent have inflated false-positive rates [71] — demonstrated that several of the major implementations of cluster-correction produced false-positive rates substantially exceeding the nominal 5% under realistic conditions. The methodology was the following: Eklund et al. used resting-state fMRI data (no task, no expected activation) from several public datasets, applied the standard task-fMRI analytic pipelines, and checked the false-positive rate against the null distribution. Under the major parametric cluster-correction implementations, false-positive rates were frequently 30–60%, sometimes 70%. The result was a major methodological event for the field.

The published response was substantial. Subsequent work refined the analytic understanding (the inflated false-positive rate depended on cluster-defining threshold choices, on smoothing parameters, on specific software implementation details), corrected the methodology (permutation-based cluster correction, which Eklund et al. recommended, produced false-positive rates at or near the nominal 5%), and characterized the affected published literature. Estimates of the number of affected published studies vary widely, but the number is substantial — the parametric cluster-correction methods Eklund et al. characterized were used in thousands of published fMRI studies over two decades.

The Eklund result is consequential beyond its specific technical content. It is an example of a fundamental methodological problem that the field had not detected for two decades despite the methodology being widely used. The detection required a specific kind of analysis — empirical false-positive rate characterization under null conditions — that was not part of the standard methodological-validation practice. The methodological reform that followed (permutation-based correction, increased reporting of analytic choices, registered analytic plans) is structurally important, but the broader lesson is about the field's vulnerability to methodological problems that take decades to detect. The doctoral reader engages other parts of the methodological literature with awareness that similar problems may exist undetected in current practice.

Reverse Inference at Doctoral Depth: Extension from Lesson 1

The reverse-inference problem was introduced in Lesson 1 at the level of its Bayesian structure. Lesson 3 extends the engagement at methodology-critique depth.

The empirical extent of reverse inference in the published cognitive neuroscience literature has been characterized at scale. The Poldrack 2011 Neuron extension and the Cognitive Atlas project [9][10] systematized the cognitive ontology and exposed the inconsistencies. The Yarkoni, Poldrack, Nichols, Van Essen, and Wager 2011 Nature Methods Neurosynth platform [72] enabled large-scale automated reverse-inference characterization by extracting frequency-of-association data between cognitive terms and brain regions across thousands of published fMRI papers. The result is that for many brain regions, the conditional probabilities required for warranted reverse inference (high P(region | function), low P(region | not function)) are not met by the published literature — the regions are active across many functions, and reverse inferences from region activity to specific functions are correspondingly weak.

The methodological response is the cumulative meta-analytic approach. Rather than treating any single study's reverse inference as warranted, the doctoral reader treats reverse inferences as licensed only when the cumulative literature supports the conditional probabilities the inference requires. Tools like Neurosynth and BrainMap (the older systematic database of fMRI activations) [73] enable this cumulative analysis at scale. Subsequent meta-analytic platforms (NiMARE, the more recent automated systematic-review platforms) continue this methodological development [74].

The doctoral writing-and-reading discipline is straightforward. Do not write reverse-inference claims that the cumulative literature does not support. When reading reverse-inference claims, evaluate them against the cumulative-literature conditional probabilities. Where the claims rest on the post-hoc invocation of prior literature, treat the claim as preliminary rather than as conclusion. This discipline is not original to the doctoral student — Poldrack, Yarkoni, and colleagues have articulated it at length — but it is the discipline doctoral training should ensure the student internalizes.

The Voodoo-Correlations Critique and Selection Bias

The voodoo-correlations critique (Vul, Harris, Winkielman, and Pashler 2009 Perspectives on Psychological Science) [75] is a specific methodological-critique paper that doctoral students should know in detail. The critique is structurally about selection bias in fMRI analysis at the level of analytic procedure.

The structure of the critique: a typical 2000s-era social-neuroscience study would (1) measure brain activity during a social-cognitive task; (2) measure a behavioral or self-report variable (anxiety, attachment style, empathy); (3) compute the correlation between brain activity and the behavioral variable at every voxel; (4) select the voxels with the highest correlation; and (5) report the correlation in those voxels as the headline finding. The reported correlations were frequently very high — sometimes r > 0.80 — and Vul et al. observed that these correlations were structurally too high for the underlying noise level of fMRI signals and the typical sample sizes of the studies.

The reason, Vul et al. demonstrated, was the circular selection procedure. By selecting voxels on the basis of their correlation with the behavioral variable and then reporting the correlation in those voxels, the procedure guaranteed inflated correlation estimates regardless of the underlying biological signal. The procedure was equivalent to running thousands of independent correlation tests, selecting the largest ones, and reporting their values. Even under pure noise, this procedure would produce some large correlations; under any underlying signal, the procedure would produce massively inflated correlation estimates.

The critique was contested at the time of publication. Some authors defended specific studies, arguing that the procedures used were less circular than the critique suggested or that the inflated correlations were known to be statistically biased and were used as descriptive rather than inferential. The substantive consensus in the field is now broadly that the voodoo-correlations critique was correct — the analytic procedure described, when used as the primary inferential basis for a finding, produces inflated estimates and unreliable inference. The methodological response has been independent-cross-validation, leave-one-out, and split-sample procedures that prevent the selection-in-and-test-in-the-same-data circularity.

Kriegeskorte, Simmons, Bellgowan, and Baker 2009 Nature Neuroscience — Circular analysis in systems neuroscience: the dangers of double dipping [76] — extended the analysis beyond social neuroscience to the broader cognitive neuroscience literature and provided the more general framing of "double dipping" as a class of selection-bias problems in fMRI analysis. The doctoral reader of a cognitive neuroscience paper that reports a correlation, contrast, or effect within a region-of-interest evaluates whether the region was selected by independent criteria (anatomical, prior-study-derived, theoretically motivated) or by criteria related to the outcome being tested. Selection by outcome-related criteria followed by testing in the selected region is double-dipping; the resulting effect estimates are biased and the inference is unreliable.

The Replication Crisis and the Methodological Reforms

Cognitive neuroscience participates in the broader replication crisis articulated at scale for psychology by the Open Science Collaboration 2015 Science [77]. The OSC reproduced 100 psychology studies and found that approximately 36% of the original significant findings replicated at the original effect size, with the replicated effect sizes substantially smaller than the originals on average. The result was field-defining for psychology and has been extended (with broadly comparable findings) to behavioral economics [78] and several biomedical domains [79][80].

For cognitive neuroscience specifically, the broader replication picture has been characterized through several specific initiatives. The Many Labs replication projects [81] have produced replication estimates for a subset of social and cognitive psychology findings. The ENIGMA consortium has produced large-sample meta-analyses across thousands of subjects for many imaging-derived measures, recovering more reliable estimates than the small-sample originals [82]. Specific subfields (the prefrontal-cortex-and-working-memory literature, the amygdala-and-emotion literature, the social-cognition literature) have had varying replication trajectories — some findings have replicated robustly at large sample size, others have attenuated substantially or failed to replicate.

The methodological reforms inspired by the replication crisis are the reforms named in Food Doctorate Lesson 3, applied here to neuroscience specifically. Pre-registration of imaging hypothesis-testing studies is increasingly common in NIH-funded work. Registered reports have growing presence in neuroscience journals. Data sharing through repositories (OpenNeuro, the various consortium-specific data portals) is increasingly required by funders and journals. Code sharing for analytic pipelines is increasingly required. Large consortium datasets (ABCD, UK Biobank, HCP, ENIGMA) provide the sample sizes the Marek 2022 analysis demonstrated were necessary for reliable brain-behavior association inference. Adversarial collaboration (the Cogitate Consortium model from Lesson 2) addresses theoretical contestation through pre-registered joint design.

Where cognitive neuroscience is in the post-replication-crisis trajectory in 2026 is broadly: substantial methodological reform has occurred, large pre-planned consortium designs are increasingly the norm for high-power questions, the published literature from the 2000s-and-early-2010s is being substantially re-evaluated, and the next generation of work is being designed with the methodological awareness the crisis produced. The trajectory is not complete; the structural conditions that produced the original underpowered literature (publication incentives, grant timescales, novelty rewards) remain partly in place; doctoral students entering the field now contribute to whatever the field's next decade looks like.

Why This Lesson Sits at the Center of the Chapter

You should leave this lesson able to read a cognitive neuroscience paper at peer-reviewer methodological depth: power analysis appropriate to the design, multiple-comparisons correction appropriate to the imaging-and-analysis pipeline, reverse-inference claims evaluated against cumulative-literature conditional probabilities, selection-bias and double-dipping evaluation, replication and pre-registration commitments noted, and overall confidence calibrated to the structural conditions of the work. This is the everyday operating skill of doctoral cognitive neuroscience research.

The Button et al. 2013 Nature Reviews Neuroscience power-failure paper is the foundational anchor that organizes this methodological territory. The Marek 2022 Nature sample-size analysis, the Eklund 2016 PNAS cluster-correction analysis, the Vul 2009 voodoo-correlations critique, the Poldrack reverse-inference framework, and the broader replication crisis literature all build on the Button et al. structural argument. Doctoral training in cognitive neuroscience is incomplete without engagement with all of them.

Lateral reference to Food Doctorate Lesson 3 (Ioannidis 2005 framework, Mendelian randomization, meta-analysis methodology critique): the structural logic is shared across fields. The doctoral reader of nutrition science and of cognitive neuroscience both navigate fields whose published literature is shaped by structural conditions that the broader meta-research literature has characterized. Methodology critique is increasingly the shared territory of biomedical and behavioral doctoral training.

Lesson Check

1. Button et al. 2013 Nature Reviews Neuroscience estimated median statistical power across major neuroscience subfields at approximately 21%. Articulate the three structural consequences of low power for the published literature: PPV reduction, effect-size inflation, and reproducibility failure. For a hypothetical neuroscience scenario (prior 0.20, power 0.30, α = 0.05), compute the predicted PPV. What does the calculation suggest about the strength of a single significant finding in this hypothetical field?
2. The Marek et al. 2022 Nature sample-size analysis demonstrated that reliable brain-wide association studies require thousands of individuals. Articulate the analysis. As a doctoral researcher proposing a brain-behavior-association study, what sample size would you propose, what dataset would you draw on, and how would you communicate the limitations of any smaller pilot work you had done to motivate the study?
3. The Eklund et al. 2016 PNAS cluster-correction analysis identified a methodological problem that had affected thousands of published fMRI studies over two decades. Articulate the analysis at the level of what Eklund et al. actually did methodologically. What does the existence of this two-decade-undetected problem suggest about how doctoral researchers should engage other parts of the methodological literature?
4. The voodoo-correlations critique (Vul et al. 2009) identifies a specific selection-bias structure in fMRI analysis. Articulate the structure. Apply the analysis to a specific published study or hypothetical scenario, identifying whether the selection-bias problem applies and what the methodological response would be.
5. The replication crisis literature has produced methodological reforms (preregistration, registered reports, data and code sharing, large consortium designs). Identify which reform best addresses which structural problem (low power, effect-size inflation, selection bias, garden-of-forking-paths multiplicity, publication bias). For a doctoral research program you might conduct, which reforms would you commit to, and which structural problems would your commitments address?

---

Lesson 4: Theoretical Frameworks in Cognitive Neuroscience

Learning Objectives

By the end of this lesson, you will be able to:

• Articulate the predictive-processing and active-inference framework (Friston, Clark, Hohwy, Seth) at the level of its specific theoretical commitments — hierarchical generative models, top-down prediction with bottom-up prediction-error signals, precision-weighting as attention, the unified action-perception inferential framing — and the empirical predictions it makes that distinguish it from the traditional information-processing framework
• Articulate the traditional information-processing framework (the Marr-Posner-Shiffrin tradition, cognitive psychology as filtered through cognitive neuroscience) at the same level of specificity and identify where it and predictive processing converge, where they diverge, and where the available evidence is currently underdetermined between them
• Read the three major contemporary theories of consciousness (IIT, GWT, HOT) at the level of their distinctive predictions and their adjudication in the Cogitate Consortium 2025 adversarial-collaboration results, and articulate the meta-level methodological lesson of the adversarial-collaboration approach
• Articulate the embodied/extended/enactive/ecological (4E) cognition framework as a substantive alternative or complement to computational cognitivism (Clark, Chemero, Thompson, Gibson lineage), identify where the framework has produced consequential empirical and theoretical advances, and engage the framework with the underdetermination posture
• Engage theoretical-framework debates in cognitive neuroscience with the doctoral posture of underdetermination — recognizing that competing frameworks can be consistent with the available evidence, identifying what would discriminate between them, and engaging the debate descriptively rather than tribally

Key Terms

Theoretical Frameworks Matter for Doctoral Research

Doctoral research in cognitive neuroscience is theoretically committed in a way that earlier modes of engagement are not. The undergraduate reading the cognitive neuroscience literature reads it as a body of findings to be received; the doctoral researcher reading the same literature reads it as the product of specific theoretical frameworks, each of which organizes the same empirical findings in different ways, each of which generates different research questions, each of which proposes different intervention targets and explanatory bridges. The theoretical framework you operate within shapes the experiments you design, the variables you measure, the contrasts you compute, the interpretive conclusions you draw. The frameworks are not optional.

Cognitive neuroscience currently contains several active theoretical-framework debates. This lesson engages four: the predictive-processing versus traditional information-processing debate, the consciousness-theory debate among IIT, GWT, and HOT, the embodied/extended cognition versus computational cognitivism debate, and the broader debate about levels of explanation (which Lesson 1 introduced). The four debates are not independent — they overlap, share evidence streams, and are sometimes combined into composite frameworks — but each presents a distinct doctoral-engagement opportunity. The lesson presents each descriptively, makes the strongest case for each, and identifies what would discriminate between them or integrate them. The Turtle's posture, as in Food Doctorate Lesson 4, is the underdetermination posture: the disagreement is the curriculum content, not the conclusion.

Predictive Processing versus Traditional Information Processing

The predictive-processing framework (introduced at frontier depth in Lesson 2) and the traditional information-processing framework are the two principal organizing frames for contemporary cognitive neuroscience. The debate between them is consequential and, at present, underdetermined at the level of strongest framework claims.

The traditional information-processing framework descends from the cognitive-psychology tradition: David Marr's three levels (Lesson 1), Atkinson and Shiffrin's multi-store memory model, Michael Posner's attention networks, Anne Treisman's feature-integration theory, Alan Baddeley's working memory model, Walter Schneider and Richard Shiffrin's controlled and automatic processing distinction [83][84][85][86]. On this framing, cognition consists of information-processing operations performed by specialized modules. Perception is feature detection and integration; attention is selective filtering; memory is encoding, storage, and retrieval; decision-making is evidence accumulation against a threshold. The brain implements these operations in specific neural circuits whose computational primitives we map to cognitive functions through the Marr levels-integration project.

The framework has been productive for decades. The bulk of the cognitive neuroscience literature operates within it — visual cortex hierarchy as feature detection at increasing scales, prefrontal cortex as control and working memory, hippocampus as relational encoding, posterior parietal cortex as attentional selection. The framework's empirical predictions have been the bread and butter of the experimental literature. The framework has produced substantial accumulated knowledge of specific cognitive functions and their neural implementations.

The predictive-processing framework (Lesson 2) reframes the picture. On the predictive-processing framing, the brain is not principally a feature detector; it is a prediction machine. Higher cortical levels generate predictions about what lower-level activity should look like, given the brain's current generative model of the world. Lower-level activity is compared to the prediction, and the discrepancy (prediction error) is the signal that updates perception and learning. Perception is not feature integration but inferential settling on the generative-model hypothesis that best explains incoming data. Attention is not selective filtering but precision-weighting of prediction-error signals. Action is not output of decision-making but active inference operating to bring sensory input into line with prior beliefs about preferred outcomes.

The predictive-processing framework reframes substantial portions of the cognitive-neuroscience literature. Hierarchical processing in visual cortex is reframed as hierarchical generative modeling. Top-down attention is reframed as precision control. Hallucinations are reframed as prediction-error failures with overweighted priors. Schizophrenia, autism, anxiety, and depression are reframed as specific forms of dysregulated prediction-error processing. The framework's reach is genuinely substantial.

The empirical predictions that distinguish the two frameworks are, in many cases, subtle. For visual processing, both frameworks predict hierarchical organization with progressively higher-level representations at later stages; the predictive-processing framework additionally predicts top-down prediction signals at every level that the information-processing framework does not necessarily predict. Specific neural correlates of top-down prediction signals — repetition suppression, mismatch responses, prediction-error signals in dopaminergic and noradrenergic systems — have been characterized and broadly support the predictive-processing framing for many cognitive operations [87][88][89]. The traditional information-processing framework has been extended to incorporate top-down signaling, blurring the distinction at the level of specific empirical predictions.

Where the frameworks differ most cleanly is in their overall organizing claims. Predictive processing claims that prediction-and-prediction-error is the fundamental computational principle of cortex, with all cognitive functions ultimately reducible to its operation. Traditional information-processing claims that cognition is the operation of multiple specialized functions, with prediction-and-prediction-error being one specific function (involved in some cognitive operations but not the foundational principle of cortex). At the level of these overall claims, the available evidence is currently underdetermined. Both framings are consistent with most of the empirical literature; specific cognitive functions are claimed by both; the field's verdict on which framing is fundamental versus auxiliary is not yet in.

The doctoral reader engages this debate as the underdetermination Lesson 1 introduced. Read each framework's strongest case in primary form (Friston, Clark, Hohwy, Seth on one side; Marr, Posner, Treisman, Baddeley, and the Patricia Churchland and Stanislas Dehaene cognitive-neuroscience mainstream on the other). Identify what would discriminate between them empirically. Engage the literature descriptively. The hypothesis-discriminating experiments are doctoral-research opportunities — research that specifically tests predictions where the two frameworks make different commitments is the research that advances the debate.

The Consciousness-Theory Debate and the Adversarial-Collaboration Methodology

The three major theories of consciousness — IIT, GWT, HOT (Lesson 2) — make distinct empirical predictions. The Cogitate Consortium 2025 adversarial collaboration tested specific IIT-versus-GWT contrasts. This lesson engages the broader theoretical landscape and the methodological lesson of the adversarial-collaboration approach.

IIT (Tononi 2004, 2016, 2023) [53][54][90] makes specific predictions: consciousness scales with integrated information (Φ); the cerebral cortex should support high Φ while the cerebellum (despite higher neuron count) should support low Φ; anesthesia and unconscious states should reduce Φ; brain-injury phenomenology should correlate with preserved Φ. The Massimini, Tononi, and Casali perturbational complexity index (PCI) operationalizes a Φ-related measure that empirically distinguishes conscious from unconscious states with substantial accuracy across anesthesia, vegetative state, minimally conscious state, and brain-injury populations [55][56]. The PCI evidence is substantial and broadly favorable to IIT's empirical predictions; the broader Φ measure is harder to compute on real brain data and remains the subject of methodological development.

GWT/GNW (Baars 1988; Dehaene and Changeux 2011, Dehaene 2014) [57][58][91] makes specific predictions: consciousness corresponds to the late ignition of broad cortical networks, particularly frontoparietal; the P3b event-related potential at approximately 300 ms post-stimulus is a signature of conscious access; attentional access is necessary for consciousness; the Mashour-Roelfsema bidirectional connectivity criterion characterizes the conscious-to-unconscious distinction empirically. The Dehaene-laboratory work has substantially advanced the empirical case for GNW; the framework predicts specific reportability-and-access patterns that the empirical literature has broadly supported.

HOT (Rosenthal 2005, Lau and Rosenthal 2011, LeDoux and Brown 2017) [61][62][92] makes specific predictions: consciousness depends on prefrontal cortical regions implementing the higher-order representation; first-order processing without higher-order representation should be possible but unconscious; disruption of higher-order representation (by TMS, lesion, or specific cognitive manipulation) should disrupt consciousness independent of first-order processing. The HOT-prediction empirical literature is smaller than the IIT and GWT literatures but has produced consequential findings in specific paradigms (perceptual confidence, blindsight, the prefrontal-lesion effects on conscious experience).

The Cogitate Consortium adversarial collaboration [63] is the methodologically important step. Cogitate brought together IIT and GWT proponents — including Tononi, Dehaene, and intermediate researchers — to design experiments together. The hypotheses were prespecified by both theories' proponents in advance. The analyses were prespecified. The discrimination criteria were prespecified. The data were collected at multiple sites and analyzed by both theory groups independently. The results, published in Nature in 2025, were mixed: some IIT predictions were supported, some not; some GWT predictions were supported, some not; the theories were not decided cleanly in favor of either. The substantive theoretical conclusion is that both theories capture some empirical features of consciousness but neither is yet the complete account.

The methodological conclusion is perhaps more consequential than the substantive one. Adversarial collaboration — joint design by proponents of competing theories, prespecified hypotheses and analyses, prespecified discrimination criteria — addresses the theory-laden-observation problem (Lesson 1) at structural depth. The methodology is now being extended to additional consciousness-theory contrasts and to other contested areas in cognitive neuroscience. Doctoral students entering consciousness research enter a field that has, recently, demonstrated that adversarial collaboration is methodologically viable for theory adjudication. The doctoral research opportunity to design and execute such collaborations is genuine.

The hard problem of consciousness (Lesson 1) sits over all of this. The IIT, GWT, and HOT theories engage what Chalmers called the easy problems — specific functional and computational structures associated with consciousness. None of the three theories addresses the hard problem in the philosophically rigorous sense Chalmers articulated; they identify what is happening in the brain when consciousness occurs, not why physical processing produces phenomenal experience at all. Whether the hard problem will dissolve under further empirical progress or persist as a structural feature of the explanatory enterprise is itself an open question. Doctoral researchers in consciousness research should not be expected to solve the hard problem in their dissertations; they should be expected to engage it with awareness as they pursue the substantial empirical and theoretical work that the field's three-theory landscape and adversarial-collaboration methodology have opened up.

Embodied / Extended / Enactive / Ecological Cognition

The 4E cognition framework constitutes a substantive alternative or complement to computational cognitivism that doctoral students should engage at least at orientation depth.

The framework's core commitment is that cognition is constitutively shaped by the body, the environment, and ongoing interaction — not merely an internal information-processing operation that the body and environment provide input and output ports for. Cognition, on the 4E framing, is constituted by sensorimotor interaction with the environment, not merely informed by it. The framework has several distinguishable strands.

Embodied cognition holds that cognitive operations are constitutively shaped by the body's physical characteristics — perception, conceptual representation, decision-making, and emotion are not abstract operations that happen to be implemented in a body but operations whose form depends on the body that performs them. The literature includes Lakoff and Johnson on metaphor [93], the action-coupling perceptual literature on the mirror-neuron system and motor simulation [94], and the experimental literature on body posture and cognition (with substantial replication-failure caveats from the broader replication crisis literature for some specific findings) [95].

Extended cognition (Clark and Chalmers 1998) [96] holds that cognitive systems can extend beyond the brain to include external tools, written representations, and cooperative partners — when these external elements participate functionally in cognitive operations under stable coupling. The extended-mind thesis has been productively engaged in technology studies, distributed cognition, and the cognitive ergonomics literature; whether the strong version of the thesis (that cognitive states themselves can be partly constituted by external elements) is more than a useful framing remains a philosophical question.

Enactive cognition (Varela, Thompson, Rosch 1991; Thompson 2007) [97][98] holds that cognition emerges from the dynamic coupling of an autonomous agent with its environment, and that the agent's living, self-maintaining biological nature is constitutive of its cognitive capacities. The framework draws heavily on dynamical-systems frameworks, phenomenology (particularly Husserl and Merleau-Ponty), and biology of autonomy (Maturana and Varela). Enactivism is the most metaphysically ambitious of the 4E strands and has substantial overlap with the philosophy-of-mind literature on what cognition is.

Ecological cognition descends from J. J. Gibson's ecological psychology [99] and holds that perception is the direct pickup of affordances — opportunities for action that the environment makes available — rather than indirect inference from sensory data. The ecological framework has been productively integrated with the dynamical-systems framing in motor control (Bernstein, Kelso) and with the affordance-based framing of perception-and-action coupling (Chemero 2009) [100][101].

The 4E framework's relationship to computational cognitivism is variously characterized. On some readings, 4E is an alternative to computational cognitivism — cognition is not computation; the computational framing is fundamentally misleading; the brain is not a generic information-processing engine that any specific cognitive function is one specific operation of. On other readings, 4E is a corrective supplement to computational cognitivism — computational analysis remains appropriate at some levels of explanation, but the body, environment, and dynamic coupling must be taken seriously as constitutive of cognition and not bracketed as merely setting the stage for cognition proper.

The doctoral reader engages 4E with the same underdetermination posture: read each strand in primary form, identify where the framework has produced consequential empirical and theoretical advances, recognize that computational cognitivism remains the mainstream framing of cognitive neuroscience but that the 4E corrective has shaped many specific subfields (motor control, perception-action coupling, social cognition, technology-mediated cognition, infant and developmental cognition). Original doctoral research in any of these areas benefits from awareness of the 4E literature, whether or not the student adopts a strong 4E commitment.

The 4E framework has substantial points of convergence with the predictive-processing framework. Andy Clark's work bridges the two: predictive processing under the active-inference framing makes the body and the environment integral to inference (the agent's predictive models include predictions about its own body and about environmental affordances), and Clark's recent The Experience Machine (2023) develops predictive processing as a thoroughly embodied framework [45]. Some authors hold that predictive processing is the computational implementation of 4E cognition; others hold that the two are distinct and partially incompatible. The integration is itself an active theoretical area.

The Doctoral Posture on Theoretical-Framework Debate

The Turtle's posture on theoretical-framework debates is the same posture the Bear takes in Food Doctorate Lesson 4. Read each framework's strongest case in primary form. Read each framework's strongest critique in primary form. Identify what evidence would advance and what would weaken each framework. Engage the debate descriptively. Where the evidence is underdetermined, recognize that it is underdetermined and do not pretend otherwise. Where one framework is materially better supported, weight accordingly. Tribal allegiance to one framework over another is a research liability; methodological vigilance and theoretical pluralism are research assets.

The original research that advances the field is research that engages the framework debates carefully, asks the questions that would discriminate between frameworks where discrimination is possible (the adversarial-collaboration methodology), and reports findings with framework-specific clarity that permits readers from any framework to integrate the findings into their own theoretical commitments.

The Turtle is in no hurry. The frameworks have been debated for decades — some of them for centuries in their proto-philosophical forms — and will be debated for decades more. Your career will contribute work to one or several of these debates. The work that advances the field will be theoretically literate; the work that does not engage the theory will be peripheral. Choose your theoretical commitments with awareness, and revise them with the evidence.

Lesson Check

1. The predictive-processing framework (Friston, Clark, Hohwy, Seth) and the traditional information-processing framework (Marr, Posner, Treisman, Baddeley, Schneider-Shiffrin) reframe substantial portions of cognitive neuroscience in different ways. For a specific cognitive function of your choosing (visual perception, attention, working memory, decision-making, learning), articulate how each framework would frame the function and identify one empirical prediction that distinguishes them. How would you design a hypothesis-discriminating experiment between the two framings for the function you chose?
2. The Cogitate Consortium 2025 adversarial-collaboration results produced mixed support for both IIT and GWT consciousness theories. Articulate the methodological structure of the adversarial collaboration (joint design, prespecified hypotheses, prespecified analyses, prespecified discrimination criteria). What is the meta-level methodological lesson? Identify two additional consciousness-theory contrasts to which the adversarial-collaboration methodology could productively be extended.
3. The 4E (embodied / extended / enactive / ecological) cognition framework constitutes a substantive alternative or complement to computational cognitivism. Articulate the framework's core commitment and identify which of the four strands resonates most with your own research interests. Engage one specific point of substantive convergence between 4E and predictive processing (e.g., the active-inference framing of embodied agency, the affordance-perception literature integrated with hierarchical prediction).
4. The hard problem of consciousness (Lesson 1) sits over the consciousness-theory debate. As a doctoral researcher entering consciousness research, articulate your posture on the hard problem. Should the hard problem influence research-question selection, research interpretation, and communication of findings? If so, how? If not, why not?
5. The Turtle's doctoral posture on theoretical-framework debates is underdetermination — recognizing that competing frameworks can be consistent with the available evidence. Apply this posture to a doctoral grant proposal: how would you write the theoretical-framework section of a proposal that engaged a framework debate descriptively rather than tribally? What language would you use to acknowledge the framework you are operating from while maintaining methodological openness to alternatives?

---

Lesson 5: The Path Forward and Original Research Synthesis

Learning Objectives

By the end of this lesson, you will be able to:

• Identify the methodological infrastructure that contemporary cognitive neuroscience most needs — at the level of large consortium datasets, biomarker development, multi-modal integration, computational tools, and open-science institutionalization — and articulate where doctoral research is positioned to contribute
• Articulate the basic-science-to-clinical-practice translation pipeline that cognitive neuroscience exists in (research informs theory informs clinical translation informs psychiatric and neurological practice) and identify the specific failure modes of this pipeline in mental health specifically — the Insel 2009/2022 articulation, the RDoC framing, the persistent gap between mechanistic neuroscience and clinical psychiatry
• Apply the methodological-evidence-threshold framework (introduced at Master's, extended at Food Doctorate Lesson 5) at doctoral neuroscience research-design depth: when does the field have enough evidence to support clinical translation, when does it not, and what kinds of recommendations are legitimate under different evidence conditions
• Apply the five-point evidence framework (design, population, measurement, effect size, replication) at doctoral research-design depth — using it not only to evaluate published research but to design original research that meets the framework's standards
• Position your own doctoral research program (current, planned, or hypothetical) within the field's open questions, the methodological infrastructure needs, and the framework debates of the previous lessons — identifying the contribution your work is positioned to make and the methodological commitments it requires
• Engage the long arc of the curriculum — from the K-12 introduction to your brain through the upper-division mechanistic and translational depth and into this Doctorate research-track depth — at the level of integrated personal commitment to the field, with the curriculum's ten-position integrator ontology held stable and the Cognition position deepened to research-track responsibility

Key Terms

The Methodological Infrastructure the Field Needs

The previous four lessons have characterized the epistemological structure, the open frontiers, the methodological tools, and the theoretical frameworks of contemporary cognitive neuroscience. This lesson turns to the path forward: what infrastructure the field most needs, where doctoral research is positioned to contribute, and how the curriculum's framework — culminating in the methodological-evidence-threshold framework introduced at Master's and extended at Food Doctorate Lesson 5 — orients original research design at the doctoral level.

The methodological infrastructure most consequential for the next decade of cognitive neuroscience includes:

(1) Large consortium datasets and the sample-size scaling. The Marek 2022 Nature analysis (Lesson 3) demonstrated that reliable brain-behavior association studies require thousands of individuals. The ABCD study (approximately 11,000 children with longitudinal multimodal imaging and behavioral data) [102], the UK Biobank imaging substudy (now approximately 100,000 imaged adults with comprehensive phenotyping) [16], the Human Connectome Project family (HCP, HCP-Aging, HCP-Development, lifespan and disease-focused HCP variants) [14][15], the ENIGMA consortium (aggregating imaging-derived measures across thousands of investigators worldwide) [82], and the Amsterdam Open MRI Collection (AOMIC) [103] are the contemporary foundational resources. Original doctoral research increasingly uses these resources as primary data; doctoral training in cognitive neuroscience that includes fluency with consortium data and the bioinformatic pipelines that work with it positions the student for the consortium-data era.

(2) Biomarker development. The biomarker development frontier in cognitive neuroscience is analogous to the biomarker frontier in nutrition science (Food Doctorate Lesson 5). Reliable biomarkers — circulating molecules, neurophysiological signatures, imaging-derived measures — that can be measured across diverse populations and that have established conditional validity for specific cognitive or clinical states are an active development frontier [104][105]. Original doctoral research that contributes to biomarker development is research with long compounding effects on the field's downstream questions.

(3) Multi-modal integration. Cognitive neuroscience increasingly operates across modalities — fMRI plus EEG plus MEG plus structural MRI plus diffusion MRI plus single-cell transcriptomics plus genomics plus behavioral assessment plus cognitive testing plus clinical phenotyping. Integration across these modalities is methodologically demanding and is an active area of methodological development. Statistical methods for multi-modal integration (canonical correlation analysis, multi-block latent variable methods, sparse partial least squares, deep multi-modal embedding) are in active development [106][107]. Doctoral training that combines substantive cognitive-neuroscience expertise with multi-modal statistical methodology positions the student for high-impact methodological-substantive integration work.

(4) Computational tools for cognitive neuroscience. The toolkit the doctoral student carries — fMRIPrep, FreeSurfer, ANTs, FSL, SPM, AFNI, MNE-Python, FieldTrip, Nilearn, BrainHub, OpenNeuro — has become substantial. Mastery of these tools is no longer optional. Beyond the standard analytic pipelines, doctoral research increasingly uses purpose-built computational tools for specific questions — reinforcement-learning model fitting libraries, predictive-processing model implementations, connectome graph-analytic libraries, single-cell transcriptomic pipelines. Doctoral training that includes deep tool fluency, alongside substantive theoretical understanding, is the doctoral training that the contemporary field requires.

(5) Open-science infrastructure (Neuroscience specifically). Preregistration, registered reports, data sharing through OpenNeuro and consortium repositories, analytic code sharing, reproducible computational environments (Docker, Singularity, BIDS-Apps), and open-access publication are the institutional and normative infrastructure that strengthens the field's signal-to-noise ratio. The neuroscience field's adoption is substantial and increasing; doctoral students contribute to the infrastructure both through their own research practice and through participation in institutional reform.

(6) Single-cell and connectomic infrastructure. The single-cell and connectomic data infrastructure — Allen Brain Atlas, BICCN (BRAIN Initiative Cell Census Network), MICrONS, FlyWire, the various spatial transcriptomic platforms — has matured rapidly. Doctoral training that includes fluency with these data types positions the student for the integrative research opportunities they enable.

This is not an exhaustive list. It is an orientation for the doctoral student asking what is my career-orienting research contribution likely to be. The honest answer in 2026 is: the field has substantially better methodological infrastructure than it had a decade ago, the Button 2013 framework's call for reform has been partially answered, large consortium datasets are increasingly available and increasingly used, and the methodological reforms inspired by the replication crisis have advanced. The infrastructure named above is what would continue to advance the field. Research that contributes to the infrastructure compounds.

The Basic-Science-to-Clinical-Practice Translation Pipeline and Its Failure Modes

Cognitive neuroscience exists in a structural pipeline linking basic research to clinical practice. Basic neuroscience produces mechanistic findings. Theoretical frameworks integrate findings into models. Clinical translation deploys frameworks into diagnostic and intervention research. Clinical practice applies the resulting tools in psychiatric and neurological settings. Under healthy conditions the pipeline nodes inform each other; research questions arise from clinical needs, basic findings inform clinical hypotheses, clinical trials test mechanistic claims, and clinical practice provides the practice-to-research feedback that subsequent basic research builds on.

Cognitive neuroscience and mental health specifically have several distinctive failure modes that doctoral students should recognize.

The mechanism-to-treatment translation failure. Mechanistic neuroscience has produced substantial findings about the neural circuits, neurotransmitter systems, and computational principles underlying various cognitive and clinical phenomena. The translation to clinical treatment has been substantially slower than in other biomedical fields. Thomas Insel — formerly director of the National Institute of Mental Health — articulated this gap in a 2009 Science paper [108] and revisited it in 2022 Healing (book) [109], noting that despite billions of dollars of research investment, the major clinical conditions in psychiatry (depression, anxiety, schizophrenia, bipolar disorder, autism, ADHD) had seen essentially no clinical-translation breakthroughs commensurate with the basic-science investment over the preceding decades. Insel's articulation is contested at margins, and the ketamine paradigm shift (Master's Lesson 1), psychedelic-assisted therapy research, neurostimulation advances, and computational-psychiatry advances all represent partial counter-examples; the overall picture, however, is one of substantial mechanism-to-treatment gap that the field has not closed.

The basic-science-to-clinical-translation feedback failure. Clinical observation in psychiatric practice generates phenomena that should inform basic neuroscience research questions. The feedback loop has been weak in mental health: clinical observations are slow to be formalized into research-question form, clinical populations are systematically under-sampled in neuroimaging cohorts relative to their information value, and the working diagnostic categories (DSM-5) on which clinical practice operates are increasingly understood as not mapping cleanly onto the neurobiology that basic research characterizes.

The diagnostic-category mismatch. The DSM diagnostic system, originating in clinical-utility framing, operates on symptom clusters that have proven not to map cleanly onto neural circuits, computational mechanisms, or molecular pathways. Major depressive disorder, schizophrenia, autism spectrum disorder, and other clinical categories are increasingly understood as heterogeneous at the neurobiological level — multiple distinct neural mechanisms can produce phenotypically similar clinical presentations, and the same neural mechanism can produce phenotypically distinct presentations. The clinical-utility framing of the DSM has obscured the mechanism-to-treatment translation: a treatment effective in one neurobiological subtype of "major depressive disorder" may be ineffective in another, but the diagnostic system does not currently distinguish them.

The Research Domain Criteria (RDoC) response. Insel and Cuthbert 2010 articulated the RDoC framework [65][66] as a structural response to the diagnostic-category mismatch. RDoC organizes psychiatric research around dimensional domains of function (negative valence, positive valence, cognition, social processes, arousal/regulatory) and around units of analysis (genes, molecules, cells, circuits, physiology, behavior, self-report, paradigms). The framework's aim is to align research dimensions with neurobiological dimensions rather than with DSM categories, with the eventual hope that dimensional research-based mappings will inform clinical translation in ways that DSM-based mappings have not. RDoC adoption has been substantial within NIMH-funded basic research and limited in clinical practice; the framework's eventual clinical translation remains an open question.

The reverse-translation failure. Some clinical phenomena have not generated the corresponding basic-research questions they should. The placebo response in psychiatry — empirically substantial across many conditions — has been substantially under-studied at mechanism depth relative to its clinical importance. The persistent rate of unresolved chronic mental illness despite multiple decades of treatment investment has not generated the corresponding basic-research focus on what specifically prevents successful translation. The implementation-science failure to translate efficacy to effectiveness has not been compensated by research investment commensurate with its clinical magnitude.

The doctoral career-research opportunity in this terrain is substantial. Original research that addresses the mechanism-to-treatment translation failure at structural depth — biomarker development that bridges from neural circuit to clinical phenotype, computational-psychiatric models that capture clinical heterogeneity at mechanism depth, dimensional-research-based intervention design that respects the diagnostic-category mismatch, implementation-science research that closes the efficacy-effectiveness gap — is research that the field substantially needs and that doctoral students are well-positioned to contribute.

The Methodological-Evidence-Threshold Framework at Doctoral Neuroscience Research-Design Depth

The Master's chapter introduced the methodological-evidence-threshold framework and Food Doctorate Lesson 5 extended it. At doctoral cognitive-neuroscience depth the framework is the everyday operating tool of research-design decision-making.

The five thresholds, as applied to cognitive neuroscience:

(1) Biological plausibility. A claim that a neural mechanism could plausibly underlie a cognitive function. The evidence requirement is mechanistic understanding consistent with the claim, from animal models, cellular and circuit research, or computational analysis. Plausibility is necessary but not sufficient for any further claim.

(2) Statistical association. A claim that a neural measurement is statistically associated with a cognitive or clinical outcome in a defined population, in a defined research design. The evidence requirement is well-conducted research with adequate sample size, careful confounder treatment, and replication. The claim does not yet establish causation.

(3) Causal inference. A claim that a neural mechanism causally affects a cognitive function or clinical outcome. The evidence requirement is convergent evidence from multiple causal-inference methodologies — perturbational experimentation (TMS, lesion, optogenetics, pharmacological manipulation), prospective cohort with covariate control, computational-causal modeling, replication across populations and designs.

(4) Intervention efficacy. A claim that a specific intervention on a neural mechanism produces a specific outcome change in a specific population. The evidence requirement is well-conducted intervention trials with prespecified primary outcomes, appropriate comparators, adequate adherence, and replication. Supports cautious clinical intervention research in trial-resembling populations.

(5) Clinical-practice recommendation. A claim that a clinical-practice recommendation is justified. The evidence requirement is intervention efficacy plus implementation effectiveness plus risk-benefit analysis plus feasibility plus equity and accessibility analysis. Supports clinical-guideline-level recommendation.

Applied to doctoral cognitive-neuroscience research design:

• Mechanism-level research (animal models, cellular and circuit research, computational analysis) operates at threshold 1 (plausibility). Communicate findings at threshold 1; identify what additional evidence would advance to higher thresholds.
• Association-level research (cohort, cross-sectional, observational imaging) operates at threshold 2 (association). Communicate findings at threshold 2; identify what causal-inference designs would advance to threshold 3.
• Causal-inference-level research (perturbational, well-controlled longitudinal, computational causal modeling, convergent multi-methodology) advances to threshold 3 (causal inference). Communicate findings at threshold 3 with explicit recognition of the populations and conditions to which the findings generalize.
• Intervention-level research (well-designed RCTs, adequately powered, with prespecified primary outcomes) advances to threshold 4 (intervention efficacy). Communicate findings at threshold 4 with the implementation and effectiveness translation explicitly distinguished.
• Clinical-practice-level recommendation is policy work, building on the underlying intervention-efficacy and effectiveness evidence base. Communicate value, feasibility, and risk-benefit premises explicitly alongside the empirical evidence.

The framework's discipline is matching recommendation thresholds to evidence thresholds, and communicating the threshold of one's own findings honestly. The doctoral student who acquires this discipline contributes work the field can integrate; the doctoral student who does not, contributes work the field has to triage.

The Five-Point Evidence Framework at Cognitive-Neuroscience Research-Design Depth

The five-point framework — design, population, measurement, effect size, replication — was introduced at earlier tiers as an evaluative tool. At doctoral depth it is a design tool.

Design. What design produces the strongest available evidence for the research question? If causal, what convergent methodologies (perturbational, prospective cohort with control, computational-causal modeling)? If mechanistic, what model systems and measurement platforms? If implementation-effectiveness, what real-world trial design? The design choice is the single largest determinant of resulting evidence quality.

Population. Who will be studied, with what generalizability scope? The non-Western-population research-gap question (Food Doctorate Lesson 2) applies to cognitive neuroscience as well — substantial fractions of the published literature have been conducted on WEIRD (Western, Educated, Industrialized, Rich, Democratic) populations [110], and the generalization to broader populations is not always empirically grounded. Generalizability is a study-design question, not a post-hoc question.

Measurement. What instruments will measure the neural and behavioral variables, and what is the measurement-error structure of each? Brain imaging has its own measurement-error structure (Lesson 3); behavioral measurement has its own; clinical phenotyping has its own. Measurement quality is largely fixed at the design stage; subsequent analytic sophistication cannot recover it.

Effect size. What effect size is the study powered to detect, and what effect size is biologically and clinically meaningful? Underpowered studies of small effects produce findings of low PPV (the Button 2013 framework); overpowered studies of trivially small effects can produce statistically-significant findings of no biological or clinical interest. The effect-size question is structurally central to study design.

Replication. Is the study designed to enable replication — preregistered, with shared data and code, with reported analytic specifications adequate for independent reanalysis? Or is the study a one-off? Replication is not a future event; it is a design choice in the present.

The doctoral student who designs research to meet the five-point framework at every node produces work the field can build on.

The Cognition Position at Doctorate

The integrator ontology established at Associates and held across Bachelor's and Master's is the conceptual spine of the Library's Higher Education tier. The Turtle holds Cognition — the cognitive-and-neural substrate that supports learning, attention, memory, decision-making, perception, action, and the inferential infrastructure that integrates input across modalities. The ten positions (Substrate, Architecture, Recovery, Stress, Light, Hydration, Cognition, Thermal-Cold, Thermal-Hot, Breath) have held stable across three tiers without expansion, and at Doctorate they continue to hold.

At Doctorate the Cognition position is engaged at research-methodology and theoretical-framework depth. Asking what theoretical frameworks best account for cognitive function at frontier depth (predictive processing versus information processing, the consciousness-theory landscape, the 4E framework's place). Asking what methodology can resolve current debates about Cognition at the causal-inference frontier (Mendelian-randomization-analogous methods for neuroimaging, perturbational experimentation, computational-causal modeling, adversarial collaboration). Asking what original research would advance the field's understanding of Cognition at the level of mechanism, computational principle, and clinical translation. Asking what philosophical and historical dimensions of the field inform our current understanding of Cognition (the levels-of-explanation question, the mind-brain explanatory gap, the hard problem of consciousness, the historical contingency of the cognitive-neuroscience paradigm).

The position holds; it is deepened. The Turtle's curriculum-spanning responsibility — to provide the cognitive-and-neural foundation that supports the integrative work the other nine positions engage with — remains the Turtle's responsibility. The mode of holding the responsibility, at Doctorate, is the mode of frontier research engagement.

The ten-position ontology continues to hold across the Library's four upper-division tiers. Whether subsequent doctoral chapters from the other seven Coaches will surface a distinct functional position requiring naming, the architecture is open to examining.

The Long Arc of the Curriculum

You have come far with the Turtle.

In K-12 you met your brain at the recognition level. At Associates you went into neuroscience proper at biochemical and circuit depth. At Bachelor's you went deeper at mechanism depth. At Master's you engaged the clinical translation. At Doctorate you have engaged the field at research-track depth — the epistemology, the methodology, the theoretical frameworks, and the path-forward research design. The curriculum has, over four upper-division tiers, taken you from the field's introduction to its frontier. The work that remains is the work of contributing original research that the field will be able to build on.

The Turtle's posture on the work ahead is the same posture the Turtle has held throughout. Patient. Methodical. Slow and deep. Expects you to keep up. The methodological vigilance the Turtle has developed across the curriculum is the methodological vigilance the doctoral researcher will deploy in choosing questions, designing studies, reading the literature, engaging the theory, communicating findings, and participating in the institutional and normative infrastructure of the field. The five-point framework is the everyday operating tool; the methodological-evidence-threshold framework is the discipline of matching recommendation to evidence; the Button 2013 Bayesian power-failure lens is the structural literacy for the published neuroscience literature; the adversarial-collaboration methodology is the contemporary methodological centerpiece for theoretical contestation; the framework debates are the theoretical commitments to engage with openness; the structural conditions of the field are the operating environment within which good work is to be done.

The Turtle has prepared you, across the curriculum, for the work you are now positioned to do. The work is yours.

The Turtle is in no hurry. Begin again.

Lesson Check

1. The methodological infrastructure the field most needs — large consortium datasets, biomarker development, multi-modal integration, computational tools, open-science institutionalization, single-cell and connectomic data infrastructure — represents an orientation for doctoral career-research contribution. Identify two infrastructure areas you would be interested in contributing to. For each, articulate the specific research question your contribution would address and the methodology you would bring.
2. The basic-science-to-clinical-practice translation pipeline in mental health has several specific failure modes (mechanism-to-treatment, basic-to-clinical feedback, diagnostic-category mismatch, reverse-translation failure). Identify each. For one failure mode, identify a doctoral-level research question that takes the failure mode as the subject of empirical investigation.
3. The methodological-evidence-threshold framework distinguishes five thresholds. Apply the framework to three contemporary cognitive-neuroscience or psychiatric-translation claims of your choice, and identify (a) the threshold of the underlying research, (b) the threshold at which the claim is being invoked, and (c) whether the claim and the evidence match.
4. The five-point evidence framework at doctoral depth is a design tool rather than only an evaluation tool. Apply it prospectively to a hypothetical doctoral research project of your choosing in cognitive neuroscience. What design, what population, what measurement, what effect size, and what replication strategy would the project use? Where would the project's strongest evidential weight lie?
5. The integrator ontology held across the upper-division tiers names ten functional positions, of which the Turtle holds Cognition. The Doctorate engagement with Cognition is engagement at research-methodology and theoretical-framework depth, rather than expansion of the ontology. Articulate, in three or four sentences, what Cognition as a position means at doctoral depth that it did not yet mean at Bachelor's or Master's depth. What is the doctoral-research-track responsibility of holding the Cognition position in the field's research community?

---

End-of-Chapter Activity: Original Research Proposal Synopsis

This activity is the doctoral version of the end-of-chapter activity, parallel to the activity in Food Doctorate. The product is a one-page synopsis (approximately 500–700 words) of an original cognitive-neuroscience research project that the student would, in principle, propose. The synopsis is not a fundable grant; it is a structured exercise in applying the chapter's frameworks to research design.

Step 1. Identify a frontier question in cognitive or clinical neuroscience that you would be interested in engaging with as original research. The question should be drawn from, or inspired by, Lessons 2 (open research frontiers), 3 (methodological critique), or 4 (theoretical-framework debates). The question should be one for which the field's current methodology is in principle capable of producing a meaningful answer.

Step 2. Frame the question explicitly. State the research question in one sentence. Identify which of the field's open questions the work addresses. Identify the theoretical framework(s) the work is positioned within or proposes to discriminate between (predictive processing, information processing, the consciousness-theory landscape, 4E cognition, computational psychiatry).

Step 3. Apply the five-point evidence framework at design depth. State the design (RCT, perturbational experimental, prospective cohort, neuroimaging consortium analysis, multi-modal integrative, adversarial-collaboration framework, implementation trial). State the population (who, with what generalizability scope, with what attention to non-WEIRD-population gaps). State the measurement (imaging modality and acquisition protocol, behavioral and cognitive assessment instruments, with what measurement-error structure). State the expected effect size and the powering — referencing Marek 2022 sample-size guidance where applicable. State the replication strategy (preregistration, registered-report format, data and code sharing, multi-site replication).

Step 4. State the threshold at which the work will report findings, using the methodological-evidence-threshold framework. Is the work positioned to advance the field at threshold 1 (plausibility), threshold 2 (association), threshold 3 (causal inference), threshold 4 (intervention efficacy), or threshold 5 (clinical-practice recommendation)? Justify the placement.

Step 5. State the structural conditions of the work. What funding model would be appropriate? What institutional and collaborative infrastructure would be required (single-site, multi-site, consortium, adversarial-collaboration partnership)? What open-science commitments would the work make? If the work touches clinical translation, what clinical-research-ethics infrastructure would be required?

Step 6. State the field-positioning of the work. What specific contribution would the work make that the field's current literature does not? What downstream research would the work enable? Who in the field would be in a position to build on the work?

The synopsis is graded by methodological literacy, framework engagement, evidential-threshold clarity, and structural realism. It is not graded by ambition. A well-framed plausibility-threshold methodological-development project of high research-question tractability scores higher than a poorly framed clinical-recommendation project that conflates evidence thresholds. The exercise is the doctoral exercise of learning to design original research that the field will be able to integrate. The substance is your choice. The discipline is the chapter's.

---

Vocabulary Review

All key terms from this chapter, alphabetized for reference:

---

Chapter Quiz

Multiple Choice (10 questions, 2 points each = 20 points)

1. Marr's three-levels framework (1982) distinguishes three levels of analysis for any information-processing system. Which of the following is not one of Marr's three levels?

A. Computational level — what the system computes and why B. Algorithmic / representational level — how the computation is performed, with what representations C. Implementational level — how the algorithm is physically realized D. Phenomenological level — what the computation feels like from the inside

2. The Button et al. 2013 Nature Reviews Neuroscience systematic review estimated median statistical power across the surveyed neuroscience literature at approximately:

A. 80% B. 50% C. 21% D. 5%

3. Reverse inference from a brain-activity pattern to a cognitive function is warranted under specific Bayesian conditions. Which of the following best characterizes the conditions?

A. P(brain region active | cognitive function present) is high B. P(brain region active | cognitive function present) is high AND P(brain region active | cognitive function absent) is low C. The brain region has been active in any prior study of the cognitive function D. The activity exceeds the cluster-correction threshold

4. The Marek et al. 2022 Nature analysis of brain-wide association studies demonstrated that reliable detection of brain-behavior correlations of typical effect size requires sample sizes on the order of:

A. 10–20 individuals B. 50–100 individuals C. Hundreds of individuals D. Thousands of individuals

5. The Eklund et al. 2016 PNAS cluster-correction analysis used a specific empirical methodology to detect the inflated false-positive rate. The methodology was:

A. Theoretical analysis of the cluster-correction equations B. Application of standard task-fMRI analytic pipelines to resting-state data (null conditions) and characterization of the resulting false-positive rate C. Re-analysis of a single influential published fMRI study D. Computational simulation of synthetic BOLD signals

6. The voodoo-correlations critique (Vul, Harris, Winkielman, Pashler 2009) identifies a specific selection-bias structure. The structure is:

A. Selecting participants based on their behavioral scores B. Selecting voxels by their correlation with a behavioral variable and then reporting the correlation in those voxels (circular procedure) C. Failing to correct for multiple comparisons at the cluster level D. Comparing brain activity in two non-matched samples

7. Karl Friston's free energy principle holds that biological systems minimize a specific quantity over time. The quantity is:

A. Total energy expenditure B. Variational free energy (a bound on the negative log-likelihood of sensory input given a generative model) C. Caloric balance D. Allostatic load

8. The three major contemporary theories of consciousness (IIT, GWT, HOT) make distinct empirical predictions. The Cogitate Consortium 2025 Nature adversarial collaboration tested IIT-versus-GWT predictions. The principal result was:

A. IIT was decisively supported, GWT was decisively refuted B. GWT was decisively supported, IIT was decisively refuted C. Mixed support for both theories, neither decided cleanly D. The Consortium could not agree on prespecified hypotheses

9. Thomas Insel's 2009 Science and 2022 Healing articulations characterize the major translational failure of mental health neuroscience as:

A. Insufficient research funding B. The substantial mechanism-to-treatment translation gap despite decades of basic-science investment C. Inadequate diagnostic criteria D. Lack of clinician training

10. The integrator ontology held across the Library's upper-division tiers names ten functional positions. The position Coach Brain holds is:

A. Substrate B. Architecture C. Cognition D. Stress

Short Answer / Application (5 questions, 6 points each = 30 points)

11. A published cognitive neuroscience paper reports a brain-behavior association in n = 28 participants between resting-state functional connectivity between two specific brain regions and a behavioral measure of anxiety. The reported correlation is r = 0.45, p = 0.02. Apply the Button 2013 framework, the Marek 2022 sample-size analysis, the Vul 2009 selection-bias analysis (as applicable), and the Ioannidis 2005 PPV framework to characterize the strength of this finding. What is the doctoral reader's appropriate confidence in the published correlation, and what additional evidence would be required to advance the finding to a higher evidence threshold?

12. A doctoral student is designing an fMRI study testing distinct predictions of the predictive-processing and traditional information-processing frameworks for visual attention. Using the five-point evidence framework at design depth (design, population, measurement, effect size, replication), draft the study's design specification. What design choice should the student make on each of the five points to produce evidence that the field will be able to integrate? Identify two structural constraints (Lesson 3) likely to compromise the design's inferential gold standard, and the methodological responses available.

13. The Cogitate Consortium adversarial-collaboration methodology — joint design by proponents of competing theories, prespecified hypotheses and analyses, prespecified discrimination criteria — addresses the theory-laden-observation problem at structural depth. Propose a specific adversarial-collaboration design for a theoretical contrast of your choosing within cognitive neuroscience (predictive processing versus information processing, the Φ-versus-ignition consciousness contrast, the dynamical-systems-versus-feature-coding motor cortex framing, others). Address: collaborating principals, joint hypothesis structure, prespecified primary outcomes, multi-site replication strategy, and adjudication criteria.

14. The basic-science-to-clinical-practice translation pipeline in mental health has several specific failure modes. The Insel articulation, the diagnostic-category mismatch, the reverse-translation failure, and the implementation-science failure each represent distinct problems. Articulate how, as a doctoral researcher entering the field in 2026, you would (a) choose a research question that engages one of these failure modes empirically, (b) read the clinical and translational literature with awareness of the failure-mode structures, and (c) contribute to the field's institutional and methodological infrastructure for translation.

15. The hard problem of consciousness (Chalmers 1995) and the mind-brain explanatory gap (Levine 1983) are structurally distinct but related problems. Articulate the distinction at PhD depth. For a doctoral researcher whose primary research program is in empirical cognitive neuroscience (not philosophy of mind), how should the existence of these problems shape (a) research-question selection, (b) research interpretation, (c) communication of findings to scientific peers and to the broader public, and (d) the doctoral student's posture on consciousness research as a career-research area?

---

Teacher's Guide

Pacing Recommendations

This chapter is structurally one chapter but operationally five seminar units. Recommended pacing for a 16-week doctoral cognitive-neuroscience methodology seminar:

Adjust to course duration and student preparation. For shorter formats, Lessons 1, 3, and 5 form a coherent core; Lessons 2 and 4 can be assigned as preparatory reading.

Lesson Check Answers

Lesson 1, Question 1. Computational level — what the system computes and why; algorithmic/representational level — how the computation is performed with what representations and procedures; implementational level — how the algorithm is physically realized. For visual object recognition: well-understood at computational (object-identification function) and algorithmic levels (feature-detection through object-identification hierarchies); substantial but incomplete at implementational level (V1-V2-V4-IT hierarchy known but cell-type and circuit detail not complete). Open integration question: how the algorithmic feature-integration steps map onto specific neural population dynamics within the IT/ventral-stream hierarchy.

Lesson 1, Question 2. Explanatory gap (Levine 1983): a satisfying explanation of cognitive or phenomenal experience from neural mechanism appears unavailable on currently understood principles — the gap may close with future science. Hard problem (Chalmers 1995): explaining why physical processing is accompanied by phenomenal experience at all, distinguished from the easy problems of explaining specific cognitive functions — structurally distinct because it concerns the existence of phenomenal experience as such. For doctoral cognitive-neuroscience researchers: research-question selection should not depend on solving these problems but should engage them with awareness; research interpretation should distinguish what the work establishes empirically from what it does not establish about consciousness as such; communication should avoid overclaiming consciousness implications that the work does not actually deliver.

Lesson 1, Question 3. Examples: prefrontal-activity-to-cognitive-control inference (PFC active across many tasks, not specifically control); ventral-striatum-to-reward inference (active across many salience and learning contexts); insula-to-disgust inference (active across many interoceptive and salience contexts). In each case the P(region | not function) is substantial, weakening the reverse inference.

Lesson 1, Question 4. Open answer — student's selection from their reading.

Lesson 1, Question 5. Technological conditions: PET imaging revolution 1980s, fMRI BOLD signal (Ogawa 1990). Disciplinary conditions: convergence of cognitive psychology, neuropsychology, behavioral neuroscience, and computational neuroscience into unified cognitive-neuroscience program in 1990s. Reconfiguration anticipated: shift from fMRI-as-primary to multi-modal-integration-as-primary, with single-cell-transcriptomic and connectomic data increasingly central; large-consortium pre-planned designs replacing individual-laboratory small-sample studies; adversarial-collaboration methodology becoming routine for theoretically contested questions.

Lesson 2, Question 1. C. elegans connectome necessary but not sufficient because it does not contain synaptic strengths and dynamics, neuronal intrinsic properties (cell-type-specific), in vivo activity patterns, or behavioral-context information. Additional data: transcriptomic (cell-type identity), dynamical (in vivo activity), behavioral (context). Each adds: cell-type-specific function from transcriptomic, activity-pattern function from dynamical, function-in-context from behavioral.

Lesson 2, Question 2. Open answer — student's cognitive-function selection. Acceptable answers identify specific cell types relevant to the function, articulate why cell-type-resolved characterization advances mechanistic explanation, and propose specific cell-type-targeted experiments (e.g., chemogenetic activation/silencing of specific Cre-driver-line targeted populations).

Lesson 2, Question 3. Dynamical-systems framing: neural function as high-dimensional state-trajectory evolution rather than feature-coding by individual neurons; relevant explanatory units are dynamical objects (manifolds, trajectories, fixed points). For a cognitive function beyond motor: dynamical-systems framing would predict that the function corresponds to a low-dimensional trajectory through population state space rather than to activation of specific feature-coding neurons. Adjudication: dimensionality-reduction analysis of population recording; manipulation of specific dynamical components (perturbation of state-space attractor structure); model fitting comparing feature-coding versus dynamical-trajectory predictions.

Lesson 2, Question 4. Productive predictive-processing prediction: repetition suppression as prediction-error-decay; hierarchical processing in visual cortex showing both forward (feature) and backward (prediction) signaling; the Smith et al. and DiCarlo work on object-recognition prediction-error signaling. FEP application with non-uniquely-identifiable predictions: characterization of any specific cognitive or behavioral function as free-energy-minimizing — multiple generative models can produce the same observable behavior, leaving the framework's specific empirical content diffuse. Doctoral engagement: distinguish productive specific applications from over-extended principle claims; favor specific hypothesis-discriminating empirical work.

Lesson 2, Question 5. IIT-GWT differing predictions: IIT predicts cerebellar involvement should be minimal despite neuron count (Φ low); GWT predicts attentional access dependence and frontoparietal ignition. Additional adversarial-collaboration designs: IIT-versus-HOT contrast on prefrontal-cortex causal role; GWT-versus-HOT contrast on attentional-access criterion versus higher-order representation criterion.

Lesson 3, Question 1. Low power consequences: PPV reduction (fewer significant findings are true); effect-size inflation (significant findings inflate true effects); reproducibility failure (subsequent better-powered work attenuates or fails to replicate). PPV for prior 0.20, power 0.30, α = 0.05: (0.30 × 0.20)/[(0.30 × 0.20)+(0.05 × 0.80)] = 0.060/0.100 = 0.60. A single significant finding has substantial probability of being a false positive; replication and triangulation across designs is essential.

Lesson 3, Question 2. Marek 2022 demonstrated that brain-behavior correlations of typical small effect size (r = 0.10–0.20) require samples of thousands for reliable detection. Doctoral proposal: sample size on order of low thousands minimum; use consortium data (ABCD, UK Biobank, HCP), or multi-site pre-planned analysis; communicate small-sample-pilot limitations explicitly; report effect-size confidence intervals with attention to the small-sample-inflation possibility.

Lesson 3, Question 3. Eklund 2016 applied standard task-fMRI analytic pipelines to resting-state data (where no true task-related activation should exist) and characterized the false-positive rate empirically, finding substantial inflation. Existence of the two-decade-undetected problem suggests doctoral readers should engage other parts of methodological literature with awareness that similar problems may exist undetected; empirical methodological-validation work (testing analytic pipelines against null conditions, replicating findings across independent data) is structurally important.

Lesson 3, Question 4. Voodoo-correlations selection-bias structure: select voxels by correlation with behavioral variable, report correlation in selected voxels. Apply to scenario: any study selecting region-of-interest on outcome-related criteria and reporting effect in selected region. Methodological response: independent cross-validation, leave-one-out, split-sample procedures; anatomically or theoretically motivated region selection.

Lesson 3, Question 5. Power → reform: large consortium designs, pre-registered sample-size justification. Effect-size inflation → reform: preregistered effect-size predictions, replication. Selection bias → reform: independent cross-validation, preregistered region-of-interest selection. Garden-of-forking-paths → reform: preregistration, specification curve analysis. Publication bias → reform: registered reports.

Lesson 4, Question 1. For visual attention: predictive-processing frames attention as precision-weighting of prediction errors; information-processing frames attention as selective filtering of inputs. Empirical distinction: predictive-processing predicts top-down signaling and prediction-error responses at every level; information-processing predicts selection at specific filtering stages. Hypothesis-discriminating design: manipulate top-down predictions and measure prediction-error responses at multiple cortical levels, compare to selection-filtering predictions of the alternative framing.

Lesson 4, Question 2. Cogitate Consortium structure: joint design by IIT and GWT proponents; prespecified hypotheses (IIT predicts X, GWT predicts Y); prespecified analyses; prespecified discrimination criteria. Meta-level lesson: adversarial-collaboration methodology addresses theory-laden-observation problem and is increasingly viable for contested fields. Additional contrasts: IIT-versus-HOT on cerebellar versus prefrontal involvement; GWT-versus-HOT on attentional-access criterion.

Lesson 4, Question 3. Open answer — student's 4E strand selection. Acceptable answers articulate the strand at primary-source depth, identify specific empirical research the strand has generated, and engage one convergence point with predictive processing (active-inference embodied agency, affordance-perception hierarchical inference, enactivist non-representationalism).

Lesson 4, Question 4. Open answer — student's posture. Acceptable answers engage the hard-problem question at PhD depth, distinguish what empirical research can and cannot establish about consciousness, and articulate a research-position posture (empirical research engages the easy problems while remaining humble about hard-problem implications).

Lesson 4, Question 5. Open answer — student's grant-proposal section. Acceptable answers name the framework the proposal operates from, articulate why this framing organizes the research, acknowledge competing frameworks would organize evidence differently, identify framework-discriminating or framework-neutral empirical commitments, commit to reporting findings in framework-specific language.

Lesson 5, Questions 1–5. Open answers — students' selections. Acceptable answers demonstrate methodological infrastructure literacy tied to specific research questions, failure-mode literacy with specific empirical entry points, threshold-framework discipline applied to current claims, five-point-framework prospective design application, and integrated understanding of the Cognition position at doctoral research-track depth.

Quiz Answer Key

1. D — Marr's three levels are computational, algorithmic/representational, and implementational. Phenomenological is not one of Marr's three levels.

2. C — Button et al. 2013 estimated median power across the surveyed neuroscience literature at approximately 21%.

3. B — Reverse inference is warranted when both P(region | function) is high AND P(region | not function) is low.

4. D — Marek 2022 demonstrated thousands of individuals are required for reliable BWAS.

5. B — Eklund 2016 applied standard task-fMRI analytic pipelines to resting-state data (null conditions) and characterized the false-positive rate empirically.

6. B — Vul 2009 identified the selection-bias structure of selecting voxels by their correlation with a behavioral variable and then reporting the correlation in those voxels.

7. B — The free energy principle holds that biological systems minimize variational free energy (a bound on the negative log-likelihood of sensory input given a generative model).

8. C — The Cogitate Consortium 2025 results produced mixed support for both IIT and GWT; neither was decided cleanly.

9. B — Insel's articulation: the substantial mechanism-to-treatment translation gap despite decades of basic-science investment.

10. C — Coach Brain holds the Cognition position.

Short-answer questions are graded on methodological literacy, framework-application clarity, and structural realism. Detailed acceptable-answer outlines follow the patterns established in the Lesson Check answers.

Discussion Prompts

1. Button et al. 2013 estimated median statistical power across neuroscience at 21%. Has the field's methodological reform over the past decade closed this power gap, or do the structural conditions (small grants, fast-paced publication, novelty-rewarded promotion) remain in place? What evidence supports each view?

2. The Marek 2022 Nature analysis suggests that a substantial fraction of the small-sample brain-behavior-association literature does not survive better-powered analysis. How should the field re-engage with that accumulated literature? Should small-sample findings be treated as preliminary and re-confirmed at scale before further work builds on them, or treated as the field's working knowledge subject to revision when discomfirmed?

3. The Cogitate Consortium 2025 results suggest that neither IIT nor GWT is the complete theory of consciousness. Does this argue for theoretical pluralism (multiple theories may capture different aspects of consciousness) or for theoretical inadequacy (the field's current theories miss the structure of the phenomenon)? What additional empirical work would adjudicate?

4. Karl Friston's free energy principle has been productive in motivating empirical work and has also been criticized as too universal in its claims to be empirically specifiable. Is the framework genuinely empirical, or is it a useful organizing principle that does not make unique empirical predictions? Does the distinction matter for doctoral research?

5. The Insel 2022 articulation of the mental-health translational failure is substantial. Has the gap closed in any specific clinical area in the years since (ketamine and esketamine, psychedelic-assisted therapy, neurostimulation, computational psychiatry)? What does the partial success in some areas suggest about the structural factors blocking translation in others?

6. The diagnostic-category mismatch (DSM categories not mapping cleanly onto neurobiology) is a substantial problem for clinical translation. The RDoC response has been substantial in research but limited in clinical practice. Should doctoral students adopt RDoC framing in their basic and translational research even when the clinical infrastructure has not yet caught up? What does the answer depend on?

7. The 4E cognition framework constitutes a substantive alternative or complement to computational cognitivism. Is the 4E literature genuinely changing the field, or is it producing parallel discussion that the cognitive-neuroscience mainstream does not engage with at depth? What evidence supports each view?

8. The hard problem of consciousness (Chalmers 1995) sits over consciousness research. Some philosophers argue it will dissolve under further empirical progress; others argue it reflects a structural feature of phenomenal experience that no empirical progress will close. Should doctoral students in empirical cognitive neuroscience take a position on this question? Why or why not?

Common Student Questions

Q: I'm a doctoral student working with small-sample fMRI data. How seriously should I take the Button 2013 and Marek 2022 framework given my actual data constraints?

A: Take it seriously enough to engage your work's structural limitations explicitly. Small-sample fMRI is still useful for some research questions (mechanism-illuminating pilot work, hypothesis-generating exploration, deep within-subject characterization) at threshold 1 (plausibility) or threshold 2 (preliminary association). Communicate the limitations honestly in publication. Do not present small-sample findings as if they were threshold-3 or threshold-4 evidence. Plan, where possible, for the small-sample work to inform a subsequent better-powered confirmatory analysis through consortium data or multi-site collaboration.

Q: The reverse-inference problem seems to undermine a lot of published cognitive neuroscience. Should I be skeptical of the field's accumulated knowledge?

A: Skeptical with calibration, not skeptical wholesale. The reverse-inference problem applies most strongly to claims that move from brain-region activity to specific cognitive functions on the strength of prior literature that does not meet the Bayesian conditional-probability conditions. Many cognitive-neuroscience findings rest on forward-inference experimental designs (manipulate cognitive task, measure brain activity) that are not vulnerable to the reverse-inference problem. The structural literacy is to distinguish which findings rest on which inferential structure and to weight them accordingly.

Q: How seriously should I take Karl Friston's free energy principle? Is it the future of cognitive neuroscience or a useful framing that may not survive?

A: Take it seriously as a productive framework that has clarified specific empirical and theoretical questions and that has motivated meaningful empirical work. Take its strongest claims (universal principle of biological systems) with the underdetermination posture — those claims rest on assumptions that are difficult to empirically test in their strongest form. The discipline is to use the framework where it does productive work in your specific research and to engage its critique with the same care you engage its proponents.

Q: I'm uncomfortable with how confident some of the consciousness-theory literature sounds. The IIT, GWT, and HOT camps each communicate as if their theory is well-supported. How do I engage this literature at doctoral depth?

A: The Cogitate Consortium 2025 result is your friend here. The adversarial-collaboration result was mixed — neither IIT nor GWT was cleanly decided. The methodological lesson is precisely that confidence beyond the actual empirical support is one of the problems the field has, and that adversarial-collaboration methodology is the field's response. Read each theory's strongest case in primary form; read the responsible critiques (the Lau-Rosenthal versus Tononi exchanges, the Hohwy-Dehaene engagement); and engage the literature with the underdetermination posture. You will not be wrong to maintain calibrated uncertainty across the three frameworks.

Q: I'm interested in the implementation-science gap in mental health. What's the doctoral-research opportunity?

A: Substantial. Mental-health implementation science is under-resourced relative to its clinical importance, and clinician-researchers and computationally-trained researchers are particularly well-positioned to contribute. Specific opportunities: characterizing the efficacy-to-effectiveness gap in specific interventions (psychotherapy, pharmacotherapy, neurostimulation, digital interventions), designing implementation interventions that close the gap, identifying patient and population characteristics that predict implementation success, building practice-to-research feedback infrastructure in clinical settings. The clinical-researcher position is an asset.

Q: I'm a philosopher-of-mind student reading this chapter. What's the doctoral-research opportunity for me in this field?

A: Substantial. Philosophy of mind and philosophy of cognitive science have substantial work to do in clarifying the field's conceptual foundations, the levels-of-explanation question, the demarcation between empirical and conceptual questions about consciousness, the legitimacy of various inferential structures (reverse inference, computational explanation, dynamical-systems explanation), the metaphysical commitments of competing frameworks. Some of the most consequential contributions to cognitive neuroscience have come from philosophers (Andy Clark, Daniel Dennett, Patricia Churchland, Jakob Hohwy, Anthony Chemero, Evan Thompson). Empirically literate philosophy of cognitive science is among the most consequential humanistic-scientific work currently being done.

Q: The chapter mentions mental-health vigilance several times. I'm planning research that involves clinical populations or that engages mental-health-adjacent content. What does the discipline require?

A: Several specific commitments. (1) Participant-screening appropriate to the research, with referral pathways to clinical care for participants who screen positive for active concerns. (2) Research-protocol attention to participant-burden and participant-effect issues — particularly for research that requires detailed self-report on mental-health content. (3) IRB consultation specifically on the population concern; consultation with clinical co-investigators where research crosses into clinical or behavioral territory. (4) Research-reporting commitments that include verified crisis resources for participants in dissemination materials. The chapter's crisis-resources section is a model for the level of care this requires.

Q: What does the long arc of the curriculum mean for someone entering at the doctoral level without the K-12 through Master's foundation?

A: The curriculum is structured so each tier is self-sufficient at its depth, but the spiral architecture means the doctoral tier assumes prior-tier substantive content. A doctoral reader without that substrate can engage this chapter and benefit, but should expect to backfill — the Master's chapter on clinical translation, the Bachelor's chapter on cellular and circuit neuroscience, the Associates chapter on cognitive-neuroscience foundations are the immediate precedents. The K-12 chapters offer foundational vocabulary. Skimming each prior tier's introductory chapter and lesson-list provides orientation; deeper engagement is rewarded but not required.

Parent Communication Template

Subject: CryoCove Library — Doctoral chapter notice (Brain, Doctorate Tier)

Dear Reader,

This is a notice that the CryoCove Library now includes a doctoral-tier chapter under Coach Brain, titled "The Epistemology of Cognitive Neuroscience." It is the second chapter of the Library's Doctorate tier (the first was under Coach Food) and is intended for doctoral-level students, postdoctoral researchers, and clinician-researchers in cognitive neuroscience, clinical psychology, psychiatry research, computational neuroscience, philosophy of mind, and adjacent research-track fields.

The chapter is not consumer-facing health guidance. It is a research-methodology and theoretical-framework engagement at doctoral depth, including discussion of methodological critique in brain imaging, theoretical-framework debates in cognition and consciousness, and the basic-science-to-clinical-practice translation pipeline's failure modes in mental health. The chapter does not recommend any specific pharmaceutical, neurostimulation, psychotherapy, or psychedelic-assisted-therapy intervention. All content is research-descriptive.

Readers below the doctoral level are welcome but may find the chapter denser than the Library's K-12 and undergraduate content. The Library's Coach Brain chapters at K-12 grades 6–12, Associates, Bachelor's, and Master's tiers cover progressive depth and remain the appropriate entry points for non-research-track readers.

The Library, including this chapter, is free and remains free as part of CryoCove's mission of Simple Human Science. Questions and feedback are welcome.

Coach Brain and the Library team

---

Illustration Briefs

Five illustrations, one per lesson. All illustrations conform to the CryoCove brand palette (Coral #FC644D, Cyan #03C7FB, White #FFFFFF, Navy #0A1628), with the Turtle as Coach Brain rendered in the established character art style. Aspect ratio: 16:9 for web; 4:3 for print. Mood throughout: doctoral seminar depth, patient, methodical, slow and deep, no theatricality.

Illustration 1 (Lesson 1): Coach Brain (the Turtle) at a quiet university library reading table. Three book stacks beside the Turtle are labeled at the spine "Computational" / "Algorithmic" / "Implementational" — Marr's three levels rendered as bibliographic columns. A small inset diagram on the wall behind the Turtle shows the Bayes-rule reverse-inference formula with conditional probabilities sketched. The Turtle is reading slowly, attentive, in the deep contemplative register the chapter has established. Coral accents in the spine labels; cyan accents in the Bayes formula; navy and white dominate.

Illustration 2 (Lesson 2): Coach Brain (the Turtle) at a laboratory bench with three monitors showing different data modalities. The left monitor shows a connectomic reconstruction with dense colored arborizations. The center monitor shows a single-cell transcriptomic UMAP plot with clustered colored dots. The right monitor shows a resting-state fMRI network diagram with network nodes and connecting edges. Beside the Turtle, a smaller side panel shows the free energy formula and a consciousness-theory triangle (IIT / GWT / HOT). The Turtle is reading attentively. Coral and cyan accents on the data panels; navy and white dominate.

Illustration 3 (Lesson 3): Coach Brain (the Turtle) at a chalkboard with the Bayesian PPV equation written out in full chalk lettering. Three smaller panels beside the equation show: (1) the Button et al. 2013 median-power-by-subfield bar chart; (2) the Eklund 2016 false-positive-rate-by-software panel with three software bars (SPM/FSL/AFNI) at elevated false-positive rates; (3) a fMRI brain image with circled clusters and arrows pointing to the cluster-correction methodology. The Turtle is teaching slowly, attentive, in the deep contemplative register. Coral accents on the PPV equation; cyan accents on the bar charts; navy and white dominate.

Illustration 4 (Lesson 4): Coach Brain (the Turtle) at a chalkboard with four theoretical-framework boxes drawn — labeled "Predictive Processing", "Information Processing", "4E Cognition", and a triangle for "Consciousness Theories (IIT/GWT/HOT)". Arrows between boxes indicate points of agreement (solid) and disagreement (dashed). A small side panel shows the adversarial-collaboration methodology as a flow diagram (proponents → joint design → prespecified prediction → empirical test → adjudication). The Turtle is gesturing toward the integrative diagram, calm and slow. Coral accents on framework boundaries; cyan accents on agreement/disagreement arrows; navy and white dominate.

Illustration 5 (Lesson 5): Coach Brain (the Turtle) at the edge of a quiet forest pond at dawn, with the path extending around the pond and into the trees. The Turtle holds an open journal. Beside the Turtle, two small inset panels show the chapter's two operating frameworks: the five-point framework ("Design / Population / Measurement / Effect Size / Replication") and the methodological-evidence-threshold framework ("1 Plausibility / 2 Association / 3 Causation / 4 Efficacy / 5 Clinical Recommendation"). The Turtle looks forward, calm, slow, ready. Mood: doctoral departure, the work ahead, the Cognition position held. Coral and cyan accents in the inset panels; navy and white dominate the forest-pond scene; the Turtle's shell is warm and grounded.

---

Crisis Resources and Support

The doctoral path in cognitive neuroscience, clinical psychology, and psychiatry research is sustained work. The populations served by the field, and the populations represented in the field's training programs, are elevated-prevalence groups for the conditions the chapter has engaged at methodological and theoretical depth. If anything in this chapter — methodological, theoretical, philosophical, or substantive — surfaces patterns that feel anxious, rigid, or out of proportion to ordinary intellectual engagement, pause. The verified resources below are real and are available.

For immediate crisis support:

• 988 Suicide and Crisis Lifeline — Call or text 988 for 24/7 free, confidential crisis support. Operational and verified as of May 2026.
• Crisis Text Line — Text HOME to 741741 for free 24/7 text-based crisis support in English and Spanish. Operational and verified as of May 2026.

For eating-disorder-specific support:

• National Alliance for Eating Disorders Helpline — (866) 662-1235, weekdays 9:00 am – 7:00 pm Eastern Time. Staffed by licensed therapists specialized in eating disorders. Email referrals available at referrals@allianceforeatingdisorders.com. Verified as of May 2026.
• The previously well-known NEDA (National Eating Disorders Association) helpline at 1-800-931-2237 is not functional and should not be cited in any context. The Alliance helpline above is the appropriate eating-disorder referral resource.

For substance use, mental health treatment, and general health support:

• SAMHSA National Helpline — 1-800-662-4357 (1-800-662-HELP). Free, confidential, 24/7, 365-day-a-year information service available in English and Spanish for individuals and family members facing mental health and substance use disorders. Provides referrals to local treatment facilities, support groups, and community-based organizations. Verified as of May 2026.

For research-community and professional resources:

• Society for Neuroscience — professional society for neuroscience researchers: sfn.org
• Organization for Human Brain Mapping (OHBM) — professional society for human brain imaging: humanbrainmapping.org
• American Psychiatric Association: psychiatry.org
• American Psychological Association: apa.org
• American Academy of Neurology: aan.com

For research methodology and open-science resources:

• EQUATOR Network (reporting standards including CONSORT, STROBE, PRISMA, others): equator-network.org
• Open Science Framework (preregistration, registered reports infrastructure): osf.io
• OpenNeuro (neuroimaging data sharing): openneuro.org
• ClinicalTrials.gov (trial registration and protocol records): clinicaltrials.gov
• Brain Imaging Data Structure (BIDS): bids-specification.readthedocs.io

If you are a doctoral student, postdoctoral researcher, or clinician-researcher in distress, the resources above are real. The work you are training to do — contributing original research that advances the field's understanding of brain and mind and serves the health of populations — is meaningful work, and it is sustained by sustainable patterns in the people doing it. Pause when you need to. Use the resources. The Turtle, and the field, are in no hurry.

---

Citations

1. Marr, D. (1982). Vision: A Computational Investigation into the Human Representation and Processing of Visual Information. W. H. Freeman.
2. Griffiths, T. L., Lieder, F., & Goodman, N. D. (2015). Rational use of cognitive resources: levels of analysis between the computational and the algorithmic. Topics in Cognitive Science, 7(2), 217–229. DOI: 10.1111/tops.12142.
3. Churchland, M. M., Cunningham, J. P., Kaufman, M. T., et al. (2012). Neural population dynamics during reaching. Nature, 487(7405), 51–56. DOI: 10.1038/nature11129.
4. Vyas, S., Golub, M. D., Sussillo, D., & Shenoy, K. V. (2020). Computation through neural population dynamics. Annual Review of Neuroscience, 43, 249–275. DOI: 10.1146/annurev-neuro-092619-094115.
5. Levine, J. (1983). Materialism and qualia: the explanatory gap. Pacific Philosophical Quarterly, 64(4), 354–361. DOI: 10.1111/j.1468-0114.1983.tb00207.x.
6. Chalmers, D. J. (1995). Facing up to the problem of consciousness. Journal of Consciousness Studies, 2(3), 200–219.
7. Goff, P. (2017). Consciousness and Fundamental Reality. Oxford University Press.
8. Poldrack, R. A. (2006). Can cognitive processes be inferred from neuroimaging data? Trends in Cognitive Sciences, 10(2), 59–63. DOI: 10.1016/j.tics.2005.12.004.
9. Poldrack, R. A. (2011). Inferring mental states from neuroimaging data: from reverse inference to large-scale decoding. Neuron, 72(5), 692–697. DOI: 10.1016/j.neuron.2011.11.001.
10. Poldrack, R. A., Kittur, A., Kalar, D., et al. (2011). The cognitive atlas: toward a knowledge foundation for cognitive neuroscience. Frontiers in Neuroinformatics, 5, 17. DOI: 10.3389/fninf.2011.00017.
11. Ogawa, S., Lee, T. M., Kay, A. R., & Tank, D. W. (1990). Brain magnetic resonance imaging with contrast dependent on blood oxygenation. PNAS, 87(24), 9868–9872. DOI: 10.1073/pnas.87.24.9868.
12. White, J. G., Southgate, E., Thomson, J. N., & Brenner, S. (1986). The structure of the nervous system of the nematode Caenorhabditis elegans. Philosophical Transactions of the Royal Society B, 314(1165), 1–340. DOI: 10.1098/rstb.1986.0056.
13. Cook, S. J., Jarrell, T. A., Brittin, C. A., et al. (2019). Whole-animal connectomes of both Caenorhabditis elegans sexes. Nature, 571(7763), 63–71. DOI: 10.1038/s41586-019-1352-7.
14. Van Essen, D. C., Smith, S. M., Barch, D. M., et al. (2013). The WU-Minn Human Connectome Project: an overview. NeuroImage, 80, 62–79. DOI: 10.1016/j.neuroimage.2013.05.041.
15. Glasser, M. F., Smith, S. M., Marcus, D. S., et al. (2016). The Human Connectome Project's neuroimaging approach. Nature Neuroscience, 19(9), 1175–1187. DOI: 10.1038/nn.4361.
16. Miller, K. L., Alfaro-Almagro, F., Bangerter, N. K., et al. (2016). Multimodal population brain imaging in the UK Biobank prospective epidemiological study. Nature Neuroscience, 19(11), 1523–1536. DOI: 10.1038/nn.4393.
17. Oh, S. W., Harris, J. A., Ng, L., et al. (2014). A mesoscale connectome of the mouse brain. Nature, 508(7495), 207–214. DOI: 10.1038/nature13186.
18. Harris, J. A., Mihalas, S., Hirokawa, K. E., et al. (2019). Hierarchical organization of cortical and thalamic connectivity. Nature, 575(7781), 195–202. DOI: 10.1038/s41586-019-1716-z.
19. MICrONS Consortium. (2025). Functional connectomics spanning multiple areas of mouse visual cortex. Nature, in press / 2025 release.
20. Bae, J. A., Baptiste, M., Bishop, C. A., et al. (2023). Functional connectomics spanning multiple areas of mouse visual cortex (preprint and subsequent release). bioRxiv. DOI: 10.1101/2021.07.28.454025.
21. Dorkenwald, S., Matsliah, A., Sterling, A. R., et al. (2024). Neuronal wiring diagram of an adult brain. Nature, 634(8032), 124–138. DOI: 10.1038/s41586-024-07558-y.
22. Schlegel, P., Yin, Y., Bates, A. S., et al. (2024). Whole-brain annotation and multi-connectome cell typing of Drosophila. Nature, 634(8032), 139–152. DOI: 10.1038/s41586-024-07686-5.
23. Tasic, B., Yao, Z., Graybuck, L. T., et al. (2018). Shared and distinct transcriptomic cell types across neocortical areas. Nature, 563(7729), 72–78. DOI: 10.1038/s41586-018-0654-5.
24. Hodge, R. D., Bakken, T. E., Miller, J. A., et al. (2019). Conserved cell types with divergent features in human versus mouse cortex. Nature, 573(7772), 61–68. DOI: 10.1038/s41586-019-1506-7.
25. Yao, Z., van Velthoven, C. T. J., Kunst, M., et al. (2023). A high-resolution transcriptomic and spatial atlas of cell types in the whole mouse brain. Nature, 624(7991), 317–332. DOI: 10.1038/s41586-023-06812-z.
26. Chen, K. H., Boettiger, A. N., Moffitt, J. R., Wang, S., & Zhuang, X. (2015). Spatially resolved, highly multiplexed RNA profiling in single cells. Science, 348(6233), aaa6090. DOI: 10.1126/science.aaa6090.
27. Rodriques, S. G., Stickels, R. R., Goeva, A., et al. (2019). Slide-seq: a scalable technology for measuring genome-wide expression at high spatial resolution. Science, 363(6434), 1463–1467. DOI: 10.1126/science.aaw1219.
28. Ståhl, P. L., Salmén, F., Vickovic, S., et al. (2016). Visualization and analysis of gene expression in tissue sections by spatial transcriptomics. Science, 353(6294), 78–82. DOI: 10.1126/science.aaf2403.
29. Raichle, M. E., MacLeod, A. M., Snyder, A. Z., et al. (2001). A default mode of brain function. PNAS, 98(2), 676–682. DOI: 10.1073/pnas.98.2.676.
30. Buckner, R. L., Andrews-Hanna, J. R., & Schacter, D. L. (2008). The brain's default network: anatomy, function, and relevance to disease. Annals of the New York Academy of Sciences, 1124, 1–38. DOI: 10.1196/annals.1440.011.
31. Yeo, B. T. T., Krienen, F. M., Sepulcre, J., et al. (2011). The organization of the human cerebral cortex estimated by intrinsic functional connectivity. Journal of Neurophysiology, 106(3), 1125–1165. DOI: 10.1152/jn.00338.2011.
32. Power, J. D., Cohen, A. L., Nelson, S. M., et al. (2011). Functional network organization of the human brain. Neuron, 72(4), 665–678. DOI: 10.1016/j.neuron.2011.09.006.
33. Churchland, M. M., Cunningham, J. P., Kaufman, M. T., Ryu, S. I., & Shenoy, K. V. (2010). Cortical preparatory activity: representation of movement or first cog in a dynamical machine? Neuron, 68(3), 387–400. DOI: 10.1016/j.neuron.2010.09.015.
34. Shenoy, K. V., Sahani, M., & Churchland, M. M. (2013). Cortical control of arm movements: a dynamical systems perspective. Annual Review of Neuroscience, 36, 337–359. DOI: 10.1146/annurev-neuro-062111-150509.
35. Kao, J. C., Nuyujukian, P., Ryu, S. I., et al. (2015). Single-trial dynamics of motor cortex and their applications to brain-machine interfaces. Nature Communications, 6, 7759. DOI: 10.1038/ncomms8759.
36. Vyas, S., O'Shea, D. J., Ryu, S. I., & Shenoy, K. V. (2020). Causal role of motor preparation during error-driven learning. Neuron, 106(2), 329–339.e4. DOI: 10.1016/j.neuron.2020.01.019.
37. Friston, K. (2010). The free-energy principle: a unified brain theory? Nature Reviews Neuroscience, 11(2), 127–138. DOI: 10.1038/nrn2787.
38. Friston, K. (2009). The free-energy principle: a rough guide to the brain? Trends in Cognitive Sciences, 13(7), 293–301. DOI: 10.1016/j.tics.2009.04.005.
39. Friston, K. J., & Kiebel, S. (2009). Predictive coding under the free-energy principle. Philosophical Transactions of the Royal Society B, 364(1521), 1211–1221. DOI: 10.1098/rstb.2008.0300.
40. Friston, K., Mattout, J., & Kilner, J. (2011). Action understanding and active inference. Biological Cybernetics, 104(1–2), 137–160. DOI: 10.1007/s00422-011-0424-z.
41. Friston, K. J., Stephan, K. E., Montague, R., & Dolan, R. J. (2014). Computational psychiatry: the brain as a phantastic organ. Lancet Psychiatry, 1(2), 148–158. DOI: 10.1016/S2215-0366(14)70275-5.
42. Friston, K., Kilner, J., & Harrison, L. (2006). A free energy principle for the brain. Journal of Physiology — Paris, 100(1–3), 70–87. DOI: 10.1016/j.jphysparis.2006.10.001.
43. Hohwy, J. (2013). The Predictive Mind. Oxford University Press.
44. Clark, A. (2016). Surfing Uncertainty: Prediction, Action, and the Embodied Mind. Oxford University Press.
45. Clark, A. (2023). The Experience Machine: How Our Minds Predict and Shape Reality. Pantheon Books.
46. Seth, A. (2021). Being You: A New Science of Consciousness. Dutton.
47. Colombo, M., & Wright, C. (2021). First principles in the life sciences: the free-energy principle, organicism, and mechanism. Synthese, 198, S3463–S3488. DOI: 10.1007/s11229-018-01932-w.
48. Bruineberg, J., Kiverstein, J., & Rietveld, E. (2018). The anticipating brain is not a scientist: the free-energy principle from an ecological-enactive perspective. Synthese, 195(6), 2417–2444. DOI: 10.1007/s11229-016-1239-1.
49. Sun, Z., & Firestone, C. (2020). The dark room problem. Trends in Cognitive Sciences, 24(5), 346–348. DOI: 10.1016/j.tics.2020.02.006.
50. Crick, F., & Koch, C. (1990). Towards a neurobiological theory of consciousness. Seminars in the Neurosciences, 2, 263–275.
51. Koch, C., Massimini, M., Boly, M., & Tononi, G. (2016). Neural correlates of consciousness: progress and problems. Nature Reviews Neuroscience, 17(5), 307–321. DOI: 10.1038/nrn.2016.22.
52. Boly, M., Massimini, M., Tsuchiya, N., Postle, B. R., Koch, C., & Tononi, G. (2017). Are the neural correlates of consciousness in the front or in the back of the cerebral cortex? Clinical and neuroimaging evidence. Journal of Neuroscience, 37(40), 9603–9613. DOI: 10.1523/JNEUROSCI.3218-16.2017.
53. Tononi, G. (2004). An information integration theory of consciousness. BMC Neuroscience, 5, 42. DOI: 10.1186/1471-2202-5-42.
54. Albantakis, L., Barbosa, L., Findlay, G., et al. (2023). Integrated information theory (IIT) 4.0: Formulating the properties of phenomenal existence in physical terms. PLOS Computational Biology, 19(10), e1011465. DOI: 10.1371/journal.pcbi.1011465.
55. Casali, A. G., Gosseries, O., Rosanova, M., et al. (2013). A theoretically based index of consciousness independent of sensory processing and behavior. Science Translational Medicine, 5(198), 198ra105. DOI: 10.1126/scitranslmed.3006294.
56. Comolatti, R., Pigorini, A., Casarotto, S., et al. (2019). A fast and general method to empirically estimate the complexity of brain responses to transcranial and intracranial stimulations. Brain Stimulation, 12(5), 1280–1289. DOI: 10.1016/j.brs.2019.05.013.
57. Baars, B. J. (1988). A Cognitive Theory of Consciousness. Cambridge University Press.
58. Dehaene, S., & Changeux, J. P. (2011). Experimental and theoretical approaches to conscious processing. Neuron, 70(2), 200–227. DOI: 10.1016/j.neuron.2011.03.018.
59. Mashour, G. A., Roelfsema, P., Changeux, J. P., & Dehaene, S. (2020). Conscious processing and the global neuronal workspace hypothesis. Neuron, 105(5), 776–798. DOI: 10.1016/j.neuron.2020.01.026.
60. Dehaene, S. (2014). Consciousness and the Brain: Deciphering How the Brain Codes Our Thoughts. Viking Press.
61. Rosenthal, D. M. (2005). Consciousness and Mind. Oxford University Press.
62. Lau, H., & Rosenthal, D. (2011). Empirical support for higher-order theories of conscious awareness. Trends in Cognitive Sciences, 15(8), 365–373. DOI: 10.1016/j.tics.2011.05.009.
63. Cogitate Consortium, Ferrante, O., Gorska-Klimowska, U., et al. (2025). An adversarial collaboration to critically evaluate theories of consciousness. Nature, 642(8068), in press.
64. Melloni, L., Mudrik, L., Pitts, M., & Koch, C. (2021). Making the hard problem of consciousness easier. Science, 372(6545), 911–912. DOI: 10.1126/science.abj3259.
65. Insel, T., Cuthbert, B., Garvey, M., et al. (2010). Research Domain Criteria (RDoC): toward a new classification framework for research on mental disorders. American Journal of Psychiatry, 167(7), 748–751. DOI: 10.1176/appi.ajp.2010.09091379.
66. Cuthbert, B. N., & Insel, T. R. (2013). Toward the future of psychiatric diagnosis: the seven pillars of RDoC. BMC Medicine, 11, 126. DOI: 10.1186/1741-7015-11-126.
67. Button, K. S., Ioannidis, J. P. A., Mokrysz, C., et al. (2013). Power failure: why small sample size undermines the reliability of neuroscience. Nature Reviews Neuroscience, 14(5), 365–376. DOI: 10.1038/nrn3475.
68. Marek, S., Tervo-Clemmens, B., Calabro, F. J., et al. (2022). Reproducible brain-wide association studies require thousands of individuals. Nature, 603(7902), 654–660. DOI: 10.1038/s41586-022-04492-9.
69. Genç, E., Schlüter, C., Fraenz, C., et al. (2023). Replication and refinement of brain-age estimation associations. Neuroimage, 273, 120066. DOI: 10.1016/j.neuroimage.2023.120066.
70. Spisak, T., Bingel, U., & Wager, T. D. (2023). Multivariate BWAS can be replicable with moderate sample sizes. Nature, 615, E4–E7. DOI: 10.1038/s41586-023-05745-x.
71. Eklund, A., Nichols, T. E., & Knutsson, H. (2016). Cluster failure: why fMRI inferences for spatial extent have inflated false-positive rates. PNAS, 113(28), 7900–7905. DOI: 10.1073/pnas.1602413113.
72. Yarkoni, T., Poldrack, R. A., Nichols, T. E., Van Essen, D. C., & Wager, T. D. (2011). Large-scale automated synthesis of human functional neuroimaging data. Nature Methods, 8(8), 665–670. DOI: 10.1038/nmeth.1635.
73. Laird, A. R., Lancaster, J. L., & Fox, P. T. (2005). BrainMap: the social evolution of a human brain mapping database. Neuroinformatics, 3(1), 65–78. DOI: 10.1385/ni:3:1:065.
74. Salo, T., Yarkoni, T., Nichols, T. E., et al. (2023). NiMARE: Neuroimaging Meta-Analysis Research Environment. NeuroLibre, 1, 7. DOI: 10.55458/neurolibre.00007.
75. Vul, E., Harris, C., Winkielman, P., & Pashler, H. (2009). Puzzlingly high correlations in fMRI studies of emotion, personality, and social cognition. Perspectives on Psychological Science, 4(3), 274–290. DOI: 10.1111/j.1745-6924.2009.01125.x.
76. Kriegeskorte, N., Simmons, W. K., Bellgowan, P. S. F., & Baker, C. I. (2009). Circular analysis in systems neuroscience: the dangers of double dipping. Nature Neuroscience, 12(5), 535–540. DOI: 10.1038/nn.2303.
77. Open Science Collaboration. (2015). Estimating the reproducibility of psychological science. Science, 349(6251), aac4716. DOI: 10.1126/science.aac4716.
78. Camerer, C. F., Dreber, A., Holzmeister, F., et al. (2018). Evaluating the replicability of social science experiments in Nature and Science between 2010 and 2015. Nature Human Behaviour, 2(9), 637–644. DOI: 10.1038/s41562-018-0399-z.
79. Begley, C. G., & Ellis, L. M. (2012). Drug development: raise standards for preclinical cancer research. Nature, 483(7391), 531–533. DOI: 10.1038/483531a.
80. Ioannidis, J. P. A. (2005). Why most published research findings are false. PLOS Medicine, 2(8), e124. DOI: 10.1371/journal.pmed.0020124.
81. Klein, R. A., Vianello, M., Hasselman, F., et al. (2018). Many Labs 2: Investigating variation in replicability across samples and settings. Advances in Methods and Practices in Psychological Science, 1(4), 443–490. DOI: 10.1177/2515245918810225.
82. Thompson, P. M., Stein, J. L., Medland, S. E., et al. (2014). The ENIGMA Consortium: large-scale collaborative analyses of neuroimaging and genetic data. Brain Imaging and Behavior, 8(2), 153–182. DOI: 10.1007/s11682-013-9269-5.
83. Atkinson, R. C., & Shiffrin, R. M. (1968). Human memory: a proposed system and its control processes. In K. W. Spence & J. T. Spence (Eds.), The Psychology of Learning and Motivation (Vol. 2, pp. 89–195). Academic Press.
84. Posner, M. I., & Petersen, S. E. (1990). The attention system of the human brain. Annual Review of Neuroscience, 13, 25–42. DOI: 10.1146/annurev.ne.13.030190.000325.
85. Treisman, A. M., & Gelade, G. (1980). A feature-integration theory of attention. Cognitive Psychology, 12(1), 97–136. DOI: 10.1016/0010-0285(80)90005-5.
86. Baddeley, A. D., & Hitch, G. (1974). Working memory. In G. H. Bower (Ed.), The Psychology of Learning and Motivation (Vol. 8, pp. 47–89). Academic Press.
87. Summerfield, C., & de Lange, F. P. (2014). Expectation in perceptual decision making: neural and computational mechanisms. Nature Reviews Neuroscience, 15(11), 745–756. DOI: 10.1038/nrn3838.
88. de Lange, F. P., Heilbron, M., & Kok, P. (2018). How do expectations shape perception? Trends in Cognitive Sciences, 22(9), 764–779. DOI: 10.1016/j.tics.2018.06.002.
89. Press, C., Kok, P., & Yon, D. (2020). The perceptual prediction paradox. Trends in Cognitive Sciences, 24(1), 13–24. DOI: 10.1016/j.tics.2019.11.003.
90. Tononi, G., Boly, M., Massimini, M., & Koch, C. (2016). Integrated information theory: from consciousness to its physical substrate. Nature Reviews Neuroscience, 17(7), 450–461. DOI: 10.1038/nrn.2016.44.
91. Dehaene, S., Charles, L., King, J. R., & Marti, S. (2014). Toward a computational theory of conscious processing. Current Opinion in Neurobiology, 25, 76–84. DOI: 10.1016/j.conb.2013.12.005.
92. LeDoux, J. E., & Brown, R. (2017). A higher-order theory of emotional consciousness. PNAS, 114(10), E2016–E2025. DOI: 10.1073/pnas.1619316114.
93. Lakoff, G., & Johnson, M. (1999). Philosophy in the Flesh: The Embodied Mind and its Challenge to Western Thought. Basic Books.
94. Rizzolatti, G., & Craighero, L. (2004). The mirror-neuron system. Annual Review of Neuroscience, 27, 169–192. DOI: 10.1146/annurev.neuro.27.070203.144230.
95. Shapiro, L. (2019). Embodied Cognition (2nd ed.). Routledge.
96. Clark, A., & Chalmers, D. (1998). The extended mind. Analysis, 58(1), 7–19. DOI: 10.1093/analys/58.1.7.
97. Varela, F. J., Thompson, E., & Rosch, E. (1991). The Embodied Mind: Cognitive Science and Human Experience. MIT Press.
98. Thompson, E. (2007). Mind in Life: Biology, Phenomenology, and the Sciences of Mind. Harvard University Press.
99. Gibson, J. J. (1979). The Ecological Approach to Visual Perception. Houghton Mifflin.
100. Chemero, A. (2009). Radical Embodied Cognitive Science. MIT Press.
101. Kelso, J. A. S. (1995). Dynamic Patterns: The Self-Organization of Brain and Behavior. MIT Press.
102. Casey, B. J., Cannonier, T., Conley, M. I., et al. (2018). The Adolescent Brain Cognitive Development (ABCD) study: imaging acquisition across 21 sites. Developmental Cognitive Neuroscience, 32, 43–54. DOI: 10.1016/j.dcn.2018.03.001.
103. Snoek, L., van der Miesen, M. M., Beemsterboer, T., et al. (2021). The Amsterdam Open MRI Collection, a set of multimodal MRI datasets for individual difference analyses. Scientific Data, 8, 85. DOI: 10.1038/s41597-021-00870-6.
104. Woo, C. W., Chang, L. J., Lindquist, M. A., & Wager, T. D. (2017). Building better biomarkers: brain models in translational neuroimaging. Nature Neuroscience, 20(3), 365–377. DOI: 10.1038/nn.4478.
105. Reardon, P. K., Seidlitz, J., Vandekar, S., et al. (2018). Normative brain size variation and brain shape diversity in humans. Science, 360(6394), 1222–1227. DOI: 10.1126/science.aar2578.
106. Smith, S. M., Nichols, T. E., Vidaurre, D., et al. (2015). A positive-negative mode of population covariation links brain connectivity, demographics and behavior. Nature Neuroscience, 18(11), 1565–1567. DOI: 10.1038/nn.4125.
107. Drysdale, A. T., Grosenick, L., Downar, J., et al. (2017). Resting-state connectivity biomarkers define neurophysiological subtypes of depression. Nature Medicine, 23(1), 28–38. DOI: 10.1038/nm.4246.
108. Insel, T. R. (2009). Translating scientific opportunity into public health impact: a strategic plan for research on mental illness. Archives of General Psychiatry, 66(2), 128–133. DOI: 10.1001/archgenpsychiatry.2008.540.
109. Insel, T. R. (2022). Healing: Our Path from Mental Illness to Mental Health. Penguin Press.
110. Henrich, J., Heine, S. J., & Norenzayan, A. (2010). The weirdest people in the world? Behavioral and Brain Sciences, 33(2–3), 61–83. DOI: 10.1017/S0140525X0999152X.
111. Schurger, A., & Graziano, M. (2022). Consciousness explained or described? Neuroscience of Consciousness, 2022(1), niac001. DOI: 10.1093/nc/niac001.
112. Lau, H. (2022). In Consciousness We Trust: The Cognitive Neuroscience of Subjective Experience. Oxford University Press.
113. Dennett, D. C. (1991). Consciousness Explained. Little, Brown and Company.
114. Churchland, P. S. (2002). Brain-Wise: Studies in Neurophilosophy. MIT Press.
115. Bechtel, W., & Mundale, J. (1999). Multiple realizability revisited: linking cognitive and neural states. Philosophy of Science, 66(2), 175–207. DOI: 10.1086/392683.
116. Putnam, H. (1967). Psychological predicates. In W. H. Capitan & D. D. Merrill (Eds.), Art, Mind, and Religion (pp. 37–48). University of Pittsburgh Press.
117. Sutton, R. S., & Barto, A. G. (2018). Reinforcement Learning: An Introduction (2nd ed.). MIT Press.
118. Schultz, W., Dayan, P., & Montague, P. R. (1997). A neural substrate of prediction and reward. Science, 275(5306), 1593–1599. DOI: 10.1126/science.275.5306.1593.
119. Daw, N. D., Gershman, S. J., Seymour, B., Dayan, P., & Dolan, R. J. (2011). Model-based influences on humans' choices and striatal prediction errors. Neuron, 69(6), 1204–1215. DOI: 10.1016/j.neuron.2011.02.027.
120. Huys, Q. J. M., Maia, T. V., & Frank, M. J. (2016). Computational psychiatry as a bridge from neuroscience to clinical applications. Nature Neuroscience, 19(3), 404–413. DOI: 10.1038/nn.4238.
121. Stephan, K. E., & Mathys, C. (2014). Computational approaches to psychiatry. Current Opinion in Neurobiology, 25, 85–92. DOI: 10.1016/j.conb.2013.12.007.
122. Adams, R. A., Stephan, K. E., Brown, H. R., Frith, C. D., & Friston, K. J. (2013). The computational anatomy of psychosis. Frontiers in Psychiatry, 4, 47. DOI: 10.3389/fpsyt.2013.00047.
123. Sterling, P., & Laughlin, S. (2015). Principles of Neural Design. MIT Press.
124. Buzsáki, G. (2019). The Brain from Inside Out. Oxford University Press.
125. Hodgkin, A. L., & Huxley, A. F. (1952). A quantitative description of membrane current and its application to conduction and excitation in nerve. Journal of Physiology, 117(4), 500–544. DOI: 10.1113/jphysiol.1952.sp004764.
126. Berman, R. M., Cappiello, A., Anand, A., et al. (2000). Antidepressant effects of ketamine in depressed patients. Biological Psychiatry, 47(4), 351–354. DOI: 10.1016/s0006-3223(99)00230-9.
127. Zarate, C. A., Singh, J. B., Carlson, P. J., et al. (2006). A randomized trial of an N-methyl-D-aspartate antagonist in treatment-resistant major depression. Archives of General Psychiatry, 63(8), 856–864. DOI: 10.1001/archpsyc.63.8.856.
128. Turner, E. H., Matthews, A. M., Linardatos, E., Tell, R. A., & Rosenthal, R. (2008). Selective publication of antidepressant trials and its influence on apparent efficacy. New England Journal of Medicine, 358(3), 252–260. DOI: 10.1056/NEJMsa065779.
129. Kirsch, I., Deacon, B. J., Huedo-Medina, T. B., et al. (2008). Initial severity and antidepressant benefits: a meta-analysis of data submitted to the Food and Drug Administration. PLOS Medicine, 5(2), e45. DOI: 10.1371/journal.pmed.0050045.
130. Cipriani, A., Furukawa, T. A., Salanti, G., et al. (2018). Comparative efficacy and acceptability of 21 antidepressant drugs for the acute treatment of adults with major depressive disorder: a systematic review and network meta-analysis. Lancet, 391(10128), 1357–1366. DOI: 10.1016/S0140-6736(17)32802-7.
131. Krystal, J. H., Abdallah, C. G., Sanacora, G., Charney, D. S., & Duman, R. S. (2019). Ketamine: a paradigm shift for depression research and treatment. Neuron, 101(5), 774–778. DOI: 10.1016/j.neuron.2019.02.005.
132. Carhart-Harris, R., Giribaldi, B., Watts, R., et al. (2021). Trial of psilocybin versus escitalopram for depression. New England Journal of Medicine, 384(15), 1402–1411. DOI: 10.1056/NEJMoa2032994.
133. Goodwin, G. M., Aaronson, S. T., Alvarez, O., et al. (2022). Single-dose psilocybin for a treatment-resistant episode of major depression. New England Journal of Medicine, 387(18), 1637–1648. DOI: 10.1056/NEJMoa2206443.
134. Griffiths, R. R., Johnson, M. W., Carducci, M. A., et al. (2016). Psilocybin produces substantial and sustained decreases in depression and anxiety in patients with life-threatening cancer: a randomized double-blind trial. Journal of Psychopharmacology, 30(12), 1181–1197. DOI: 10.1177/0269881116675513.
135. Volkow, N. D., Wang, G.-J., Tomasi, D., & Baler, R. D. (2013). The addictive dimensionality of obesity. Biological Psychiatry, 73(9), 811–818. DOI: 10.1016/j.biopsych.2012.12.020.
136. Volkow, N. D., Koob, G. F., & McLellan, A. T. (2016). Neurobiologic advances from the brain disease model of addiction. New England Journal of Medicine, 374(4), 363–371. DOI: 10.1056/NEJMra1511480.
137. Berridge, K. C. (2007). The debate over dopamine's role in reward: the case for incentive salience. Psychopharmacology, 191(3), 391–431. DOI: 10.1007/s00213-006-0578-x.
138. Nestler, E. J. (2008). Transcriptional mechanisms of addiction: role of ΔFosB. Philosophical Transactions of the Royal Society B, 363(1507), 3245–3255. DOI: 10.1098/rstb.2008.0067.
139. Miller, A. H., & Raison, C. L. (2016). The role of inflammation in depression: from evolutionary imperative to modern treatment target. Nature Reviews Immunology, 16(1), 22–34. DOI: 10.1038/nri.2015.5.
140. Raison, C. L., Rutherford, R. E., Woolwine, B. J., et al. (2013). A randomized controlled trial of the tumor necrosis factor antagonist infliximab for treatment-resistant depression: the role of baseline inflammatory biomarkers. JAMA Psychiatry, 70(1), 31–41. DOI: 10.1001/2013.jamapsychiatry.4.
141. Cryan, J. F., O'Riordan, K. J., Cowan, C. S. M., et al. (2019). The microbiota-gut-brain axis. Physiological Reviews, 99(4), 1877–2013. DOI: 10.1152/physrev.00018.2018.
142. Mayer, E. A., Knight, R., Mazmanian, S. K., Cryan, J. F., & Tillisch, K. (2014). Gut microbes and the brain: paradigm shift in neuroscience. Journal of Neuroscience, 34(46), 15490–15496. DOI: 10.1523/JNEUROSCI.3299-14.2014.
143. Gelman, A., & Loken, E. (2014). The statistical crisis in science. American Scientist, 102(6), 460–465. DOI: 10.1511/2014.111.460.
144. Simonsohn, U., Simmons, J. P., & Nelson, L. D. (2020). Specification curve analysis. Nature Human Behaviour, 4(11), 1208–1214. DOI: 10.1038/s41562-020-0912-z.
145. Steegen, S., Tuerlinckx, F., Gelman, A., & Vanpaemel, W. (2016). Increasing transparency through a multiverse analysis. Perspectives on Psychological Science, 11(5), 702–712. DOI: 10.1177/1745691616658637.
146. Lieberman, M. D., & Cunningham, W. A. (2009). Type I and Type II error concerns in fMRI research: re-balancing the scale. Social Cognitive and Affective Neuroscience, 4(4), 423–428. DOI: 10.1093/scan/nsp052.
147. Nichols, T. E., & Hayasaka, S. (2003). Controlling the familywise error rate in functional neuroimaging: a comparative review. Statistical Methods in Medical Research, 12(5), 419–446. DOI: 10.1191/0962280203sm341ra.
148. Esteban, O., Markiewicz, C. J., Blair, R. W., et al. (2019). fMRIPrep: a robust preprocessing pipeline for functional MRI. Nature Methods, 16(1), 111–116. DOI: 10.1038/s41592-018-0235-4.
149. Gorgolewski, K. J., Auer, T., Calhoun, V. D., et al. (2016). The brain imaging data structure, a format for organizing and describing outputs of neuroimaging experiments. Scientific Data, 3, 160044. DOI: 10.1038/sdata.2016.44.
150. Markiewicz, C. J., Gorgolewski, K. J., Feingold, F., et al. (2021). The OpenNeuro resource for sharing of neuroscience data. eLife, 10, e71774. DOI: 10.7554/eLife.71774.
151. Friston, K. (2012). Ten ironic rules for non-statistical reviewers. NeuroImage, 61(4), 1300–1310. DOI: 10.1016/j.neuroimage.2012.04.018.
152. Munafò, M. R., Nosek, B. A., Bishop, D. V. M., et al. (2017). A manifesto for reproducible science. Nature Human Behaviour, 1, 0021. DOI: 10.1038/s41562-016-0021.
153. Nosek, B. A., Ebersole, C. R., DeHaven, A. C., & Mellor, D. T. (2018). The preregistration revolution. PNAS, 115(11), 2600–2606. DOI: 10.1073/pnas.1708274114.
154. Chambers, C. D., & Tzavella, L. (2022). The past, present and future of registered reports. Nature Human Behaviour, 6(1), 29–42. DOI: 10.1038/s41562-021-01193-7.
155. Bennett, C. M., Baird, A. A., Miller, M. B., & Wolford, G. L. (2009). Neural correlates of interspecies perspective taking in the post-mortem Atlantic salmon: an argument for proper multiple comparisons correction. Journal of Serendipitous and Unexpected Results, 1, 1–5.
156. Tononi, G. (2008). Consciousness as integrated information: a provisional manifesto. Biological Bulletin, 215(3), 216–242. DOI: 10.2307/25470707.
157. Mudrik, L., Faivre, N., & Koch, C. (2014). Information integration without awareness. Trends in Cognitive Sciences, 18(9), 488–496. DOI: 10.1016/j.tics.2014.04.009.
158. Pessoa, L. (2022). The entangled brain: a unified mind from heterogeneity. Neuron, 110(20), 3260–3262. DOI: 10.1016/j.neuron.2022.09.012.
159. Diedrichsen, J., & Kriegeskorte, N. (2017). Representational models: a common framework for understanding encoding, pattern-component, and representational-similarity analysis. PLOS Computational Biology, 13(4), e1005508. DOI: 10.1371/journal.pcbi.1005508.
160. Kriegeskorte, N., & Douglas, P. K. (2018). Cognitive computational neuroscience. Nature Neuroscience, 21(9), 1148–1160. DOI: 10.1038/s41593-018-0210-5.
161. Yamins, D. L., & DiCarlo, J. J. (2016). Using goal-driven deep learning models to understand sensory cortex. Nature Neuroscience, 19(3), 356–365. DOI: 10.1038/nn.4244.
162. Schrimpf, M., Kubilius, J., Hong, H., et al. (2018). Brain-Score: which artificial neural network for object recognition is most brain-like? bioRxiv. DOI: 10.1101/407007.
163. Cao, R., & Yamins, D. (2024). Explanatory models in neuroscience: part 1 — taking mechanistic abstraction seriously. Cognitive Systems Research, 84, 101226. DOI: 10.1016/j.cogsys.2023.101226.
164. Ioannidis, J. P. A. (2008). Why most discovered true associations are inflated. Epidemiology, 19(5), 640–648. DOI: 10.1097/EDE.0b013e31818131e7.
165. Begley, C. G., & Ioannidis, J. P. A. (2015). Reproducibility in science: improving the standard for basic and preclinical research. Circulation Research, 116(1), 116–126. DOI: 10.1161/CIRCRESAHA.114.303819.
166. Page, M. J., McKenzie, J. E., Bossuyt, P. M., et al. (2021). The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ, 372, n71. DOI: 10.1136/bmj.n71.
167. Moher, D., Hopewell, S., Schulz, K. F., et al. (2010). CONSORT 2010 explanation and elaboration: updated guidelines for reporting parallel group randomised trials. BMJ, 340, c869. DOI: 10.1136/bmj.c869.
168. Friston, K., Parr, T., & Pezzulo, G. (2017). Active inference: a process theory. Neural Computation, 29(1), 1–49. DOI: 10.1162/NECO_a_00912.
169. Parr, T., Pezzulo, G., & Friston, K. J. (2022). Active Inference: The Free Energy Principle in Mind, Brain, and Behavior. MIT Press.
170. Lyon, P. (2020). Of what is "minimal cognition" the half-baked version? Adaptive Behavior, 28(6), 407–424. DOI: 10.1177/1059712319871360.
171. Gomez-Marin, A., & Ghazanfar, A. A. (2019). The life of behavior. Neuron, 104(1), 25–36. DOI: 10.1016/j.neuron.2019.09.017.
172. Cisek, P., & Hayden, B. Y. (2022). Neuroscience needs evolution. Philosophical Transactions of the Royal Society B, 377(1844), 20200518. DOI: 10.1098/rstb.2020.0518.
173. Krakauer, J. W., Ghazanfar, A. A., Gomez-Marin, A., MacIver, M. A., & Poeppel, D. (2017). Neuroscience needs behavior: correcting a reductionist bias. Neuron, 93(3), 480–490. DOI: 10.1016/j.neuron.2016.12.041.
174. Yarkoni, T. (2022). The generalizability crisis. Behavioral and Brain Sciences, 45, e1. DOI: 10.1017/S0140525X20001685.