An editor asks AI to summarize a stack of reviewer reports before reading them herself. Days later, reviewing her own decision, she finds herself wondering whether the judgment she reached was fully her own. That small moment points to a much larger question now facing scholarly publishing.

Generative AI has transformed scholarly publishing in under three years. Yet despite thousands of articles, webinars, conferences, and policy discussions, there remains little clarity about which kinds of cognition AI can perform and which must remain human. Discussions about AI often become polarized: some treat it as capable of replacing an ever-expanding range of cognitive work, while others emphasize its limitations and risks. Both perspectives can obscure a more fundamental question: what kinds of cognition can AI actually perform, and where must human responsibility remain?

AI is now a permanent feature of the scholarly landscape. The question is no longer whether it should be used, but how it should be understood and governed.

This article begins from the premise that not all cognition is the same. Some cognitive tasks involve pattern recognition, retrieval, and mechanical processing. Others require discovery, discernment, and original insight. These forms of cognition are not simply different in degree; they are different in kind. AI can retrieve, recombine, and generate from existing knowledge in increasingly sophisticated ways, but it cannot originate the inquiry, judgment, or insight that gives those capabilities their value. For authors, reviewers, editors, and publishers, understanding that distinction is one of the most important governance challenges of the AI era.

The Cognitive Responsibility Framework proposed here is a conceptual framework for governing human–AI collaboration in scholarly publishing. In a previous post in The Scholarly Kitchen, Before the Guardrails: Why AI Governance in Research Must Start with Purpose, I introduced an early distinction between mechanical cognitive work, which AI can increasingly perform, and judgment-based cognitive work, where human responsibility must remain central. This article builds on that distinction. Part 1 develops the conceptual hierarchy and examines where the boundary lies. Part 2 explores what happens when that boundary is repeatedly ignored: what this framework calls cognitive debt, the gradual erosion of the cognitive capacities that give scholarship its meaning.

The framework is not intended to classify people or rank scholarly activities. Rather, it offers a practical vocabulary for deciding where AI can appropriately extend human cognition and where human responsibility must remain primary. Readers who disagree about the precise nature of intelligence or discovery may still find the framework useful, because its purpose is governance, not cognitive science.

A note on what this framework is and is not: this is an opinion piece, not a scientific claim. The distinctions it draws, between thought and intelligence, between mechanical and non-mechanical cognition, are conceptual heuristics, and the boundary between them will rarely be as clean in practice as a hierarchy diagram suggests. Cognitive debt, introduced in Part 2, is likewise offered as a risk to watch for, not a measured or validated metric. The purpose of the framework is not to settle these questions empirically but to give editors, reviewers, and publishers a shared vocabulary for asking them.

A Conceptual Hierarchy of Human Cognition

Why does this matter for publishers? Every editorial workflow ultimately involves deciding which forms of cognition can appropriately be delegated to AI and which cannot. The framework that follows provides a practical way of making those distinctions.

For the purposes of this framework, intelligence refers to the governing faculty that directs and evaluates thought, rather than to psychometric ability or general reasoning as measured by IQ.

Within this framework, AI functions as artificial thought. Thought, whether human or artificial, is the acting faculty of cognition. It retrieves, compares, combines, analyzes, and generates possibilities from what is already known, but it cannot determine which possibilities deserve trust or action. Intelligence provides that direction, evaluating thought and recognizing when existing patterns must give way to discovery, discernment, and new understanding. The distinction is offered as an operational model for cognitive governance rather than as a comprehensive theory of mind.

Intelligence draws upon every form of cognition, operating through discovery and discernment without being reducible to any of them. It enables a researcher to recognize when a theory no longer fits the evidence despite appearing internally consistent, or when an accepted consensus is preventing important questions from being asked. The other forms of cognition do not accumulate into intelligence; they are the means through which it operates.

The framework should not be understood as a collection of independent cognitive faculties operating in sequence. Rather, it describes different functions within a single integrated cognitive system. Scholarly activities move fluidly across these functions, making the distinction between mechanical and non-mechanical cognition a useful guide rather than a rigid boundary.

This ordering resembles David Bohm’s distinction between machines and living organisms. In a machine, the parts are assembled to produce a functioning whole. In an organism, the whole continuously organizes the activity of its parts. Likewise, knowledge, thought, discovery, discernment, and intelligence remain continuously interconnected: higher forms of cognition organize and direct lower ones, while the system as a whole operates through continuous feedback.

Knowledge and thought together constitute mechanical cognition: the capacity to store, retrieve, organize, and manipulate information through recognizable patterns. It is repetitive, trainable, and content-based, operating on knowledge that can be accumulated, transferred, and scaled. Non-mechanical cognition refers to the forms of cognition grounded in discovery, discernment, and judgment through direct engagement with evidence. These forms are context-sensitive, qualitative, and awareness-based; they depend on direct observation, reflection, and tacit understanding. Thought constructs possibilities from what knowledge provides. Discovery and discernment provide the context, meaning, and purpose that direct thought, determining which possibilities correspond to reality and deserve trust.

Learning illustrates the boundary between these two domains. One form extends what is already known through repetition and refinement, and AI increasingly performs this kind of learning by strengthening pattern recognition and prediction. The other is discovery: the universal human capacity to let experience revise what prior knowledge suggests. A child encountering a bird for the first time exercises the same essential cognitive process as a scientist who watches unexpected evidence overturn an established theory. Discovery is not a specialist skill but the willingness to let observation reshape understanding. As perception changes, discernment begins.

Discovery provides new encounters with evidence; discernment determines their significance, evaluating not simply what information exists, but whether it is true, significant, and well-founded. Consider a journal editor reading a manuscript on a contested topic. Discernment allows the editor to recognize that the data are sound while the conclusions overreach, or that the methodology is genuinely novel despite a cautious framing.

Understanding is the fruit of discovery and discernment: the coherent integration of knowledge that emerges when experience has been tested against evidence. Applied repeatedly to decisions under uncertainty, discernment gives rise to judgment. Throughout this process, intelligence remains the governing faculty. It determines how thought responds to what discovery and discernment reveal, recognizing when existing patterns remain sufficient and when they must give way to new insight. Knowledge accumulates through thought; meaning, purpose, and direction flow through discovery, discernment, and intelligence.

Mechanical Cognition: What AI Does Well

Mechanical cognition is what knowledge and thought become when applied in patterned, repeatable, and rule-governed ways. Current AI systems excel here. Within our framework, AI is artificial thought. Like human thought at its most systematic, it retrieves, compares, analyzes, summarizes, predicts, and generates; the difference is one of speed, scale, and lack of fatigue. Formatting citations, summarizing manuscripts, screening submissions, extracting structured information, and identifying recurring patterns are all forms of mechanical cognition that AI performs efficiently.

None of this makes mechanical cognition a lesser form of scholarship. Replication studies, systematic reviews, technical editing, and compliance checks derive their value precisely from consistency and repeatability. Where the objective is accuracy and standardization, AI can substantially strengthen human capability.

The same principle applies to scholarly writing. AI can improve clarity, strengthen structure, refine language, and make ideas more accessible. What it should not replace is the original insight that arises from direct engagement with evidence. Expression should remain faithful to insight rather than substitute for it. The value of scholarship lies not only in how clearly ideas are communicated, but in the discovery, discernment, and judgment from which they emerge.

Discovery and Discernment: Learning Beyond Pattern Recognition

AI systems can surface unexpected patterns, identify anomalies, and reveal relationships that researchers later recognize as significant. Within this framework, however, AI recognizes patterns by relating new inputs to accumulated knowledge. It may identify relationships that appear novel to human observers, but within this framework, recognition is treated as distinct from discovery. Discovery begins with direct human engagement with evidence. Discernment then evaluates the significance of that encounter, providing the context, meaning, and purpose that direct thought toward understanding and, ultimately, judgment.

Richard Feynman recalled his father saying that knowing the name of a bird in every language tells you nothing about the bird itself. Within this framework, the name is knowledge. Using that knowledge to identify and describe the bird is thought. Observing the bird directly and letting experience revise prior understanding is discovery. Recognizing that the name is not the bird itself is discernment.

This distinction has immediate implications for scholarly publishing. Take an editor who asks AI to summarize reviewer reports before reading them. The summary may be accurate, but it becomes the editor’s first encounter with the manuscript: an interpretation has already intervened before direct observation begins. Judgment may still be careful, but it is no longer entirely independent, because the initial relationship with the evidence has been mediated. Subtleties that might have emerged through direct reading, a reviewer’s hesitation, a methodological concern, a tension buried in a footnote, may never surface, because AI has already determined what appears most important.

The principle is even more important in peer review. A reviewer who asks AI to evaluate a manuscript before reading it places an intermediary between themselves and the evidence. The resulting review may identify genuine strengths and weaknesses, but its conclusions are no longer rooted entirely in the reviewer’s own understanding. The value of peer review lies not simply in reaching sound conclusions, but in arriving at them through careful reading, discovery, reflection, and judgment.

The distinction between mechanical and non-mechanical cognition is therefore more than theoretical. Every time authors, reviewers, or editors decide whether to engage directly with evidence or allow AI to mediate that engagement, they make a choice about cognitive responsibility. Repeated across thousands of scholarly decisions, those choices shape not only individual expertise but also how scholarly institutions learn.