AI Is Bloom’s Taxonomy in Reverse

We have built tools that let people start at the top of human cognition without climbing the levels beneath it. The consequences of that are only beginning to become clear.

In 1956, a committee of American educational psychologists led by Benjamin Bloom published a framework for classifying educational objectives that would go on to influence teaching and learning across the English-speaking world for the better part of seven decades. Bloom’s Taxonomy, as it became known, arranged cognitive skills in a hierarchy from the simplest to the most demanding:

Remember. Understand. Apply. Analyse. Evaluate. Create.

The framework was revised in 2001 by Lorin Anderson and David Krathwohl, who adjusted the language slightly and confirmed the hierarchy’s direction: genuine cognitive development moves upward through these levels. You cannot meaningfully evaluate something you have not first understood. You cannot create something of real substance without the analytical capacity that comes from having first remembered, understood, applied, and evaluated. Creation sits at the top because it requires everything beneath it.

The hierarchy is not arbitrary. It reflects something true about how human beings develop the capacity to think. The lower levels are not lesser. They are foundational. A student who has not developed the lower levels cannot sustain the higher ones. The climb matters as much as the destination.

We have now built tools that allow people to begin at Create and work downward. Or, more precisely, to produce outputs that look like the product of the top level without having climbed through the levels that were supposed to produce the capacity for it.

This is not a minor pedagogical inconvenience. It is a structural problem with consequences we are only beginning to understand.

The Process Is Not a Route to the Outcome. The Process Is the Outcome.

Bloom’s Taxonomy is frequently misread as a simple ranking of task difficulty. It is not. The taxonomy’s deeper claim is that the process of working through the lower levels is what builds the cognitive capacity for the higher ones. Writing an essay is not valuable primarily because it produces an essay. It is valuable because the process of producing it: selecting relevant material, organising an argument, evaluating evidence, expressing a position in language precise enough to be examined. That process develops capacities that cannot be developed any other way. The essay is the output. The cognitive development is the point.

This is why Bloom’s framework has been so durable in educational theory. It is not a rubric for marking work. It is a model of what learning is for. The lower levels are not stepping stones to be crossed as quickly as possible on the way to the interesting stuff. They are the mechanism by which the interesting stuff becomes possible at all.

The GPS made the point first, and it runs through much of Well… How Did We Get Here? Nobody decided to become worse at navigating. The capability atrophied because the technology made it unnecessary to exercise it. The spatial memory that built navigational skill was never consciously abandoned. It simply was not used, and unused cognitive capacities, like unused muscles, weaken over time.

Researchers have introduced the concept of AI-chatbot-induced cognitive atrophy: the potential deterioration of essential cognitive abilities resulting from excessive dependence on AI tools. The GPS is the prototype for this. The LLM essay generator is its educational equivalent, operating at a level of cognitive significance several orders of magnitude greater.

Cognitive Atrophy from AI Use Is Now Empirically Measurable

The academic literature on AI and cognitive development is moving fast, and the findings are, taken together, sobering.

A study by Michael Gerlich at SBS Swiss Business School, examining 666 participants, found that increased reliance on AI tools is linked to diminished critical thinking abilities, with cognitive offloading identified as the primary driver of the decline. Cognitive offloading, the transfer of mental processes to an external system, is not inherently harmful. Writing things down is a form of cognitive offloading. So is using a calculator. The question is always which cognitive capacities are being offloaded, and whether their offloading atrophies the skills that matter.

A 2025 analysis in the American Journal of Education and Information Technology drew a distinction that matters: using an AI tool to generate a draft that is then critically analysed and rewritten by the student is an augmentation practice. Using AI to generate a final draft that is submitted with minimal engagement is an atrophy practice. The tool is the same. The pedagogy around its use determines whether the student is climbing the taxonomy or bypassing it.

Research in interactive learning environments found that as tasks became more complex and uncertain, students using chatbots relied more heavily on lower-level thinking, remembering and understanding, rather than developing the higher-order skills the tasks were designed to build. The AI was not elevating the students to higher cognitive levels. It was providing a route around the cognitive effort those levels require.

The AI-Generated Essay Has the Output Without the Development That Produced It

Consider what happens when a student uses an AI tool to produce an essay on, say, the causes of the First World War.

The traditional process of reading primary and secondary sources, identifying conflicting interpretations, constructing a thesis, marshalling evidence, writing and revising a draft, moves through Bloom’s levels in sequence. The student remembers the facts, works to understand the historiographical debates, applies that understanding to their specific argument, analyses the evidence, evaluates the competing interpretations, and finally creates an original synthesis. The essay that emerges is evidence that the process happened. The cognitive development it represents is the actual educational outcome.

The AI-assisted process can produce a plausible essay on the same subject in minutes. The essay may be well-structured, accurately sourced, and competently argued. What it cannot do is produce the cognitive development that the traditional process was designed to generate. The student who submits it has the output without the process. They have the appearance of creation without the climb that genuine creation requires.

Wayne Holmes, Professor of Critical Studies of Artificial Intelligence and Education at University College London and UNESCO Chair in the Ethics of Artificial Intelligence and Education, has noted that emergent technologies replacing learning is not a new idea, but the scale and accessibility of current AI tools makes the question more urgent than it has ever been.

The urgency is real because the reversal operates at scale. When a single student bypasses the taxonomy once, the effect is limited. When an entire generation routinely bypasses it across most of their educational experience, the long-term effects on society’s cognitive infrastructure are genuinely unknown. That uncertainty is itself a reason for caution.

The Boundary Between AI-Appropriate and Human-Only Cognition Keeps Moving

The standard response is to teach students to use AI tools well. Use them for the lower levels, where AI genuinely excels, and reserve the higher levels, analysis, evaluation, creation, for human cognitive effort. This is the approach most educational institutions are currently attempting, and it is better than nothing.

The problem is that the boundary is not stable. AI tools are improving rapidly, and the cognitive levels at which they can produce plausible outputs are moving upward. A tool that could only reliably handle Remember and Understand in 2022 can now produce outputs that, on the surface, resemble Analyse and Evaluate. The categories are not fixed against a moving technology.

The “use AI for the lower levels” approach has a deeper problem. The lower levels of Bloom’s Taxonomy are not merely stepping stones. They are, as the framework argues, the foundation for the higher levels. A student who never does the work of remembering and understanding, because AI handles that, does not simply have more time for higher-order thinking. They lack the material from which higher-order thinking is built. Analysis requires something to analyse. Evaluation requires a body of knowledge against which to evaluate. You cannot shortcut the foundations and then build on them.

The Online Learning Consortium’s analysis of AI and Bloom’s Taxonomy noted that true understanding involves judgement, reflection, and the creation of meaning: capacities that draw on experience, values, subject knowledge, and empathy. These are human capacities that AI cannot possess, because AI systems extract and apply information by identifying statistical patterns rather than developing genuine conceptual understanding. The AI can produce the outputs that look like these capacities. It cannot develop them in the student who uses it.

The Fear Beneath the Fear

There is a fear in the AI conversation that most commentary cannot quite name.

The public AI conversation runs primarily on two fears. Fear of replacement: AI will do what I do, more efficiently, and I will become unnecessary. And fear of deception: AI-generated content will be mistaken for human thought, and we will not be able to tell the difference.

The deepest fear, the one the Bloom’s reversal points to, is the fear of hollowing. Not that AI will replace human thinking from the outside, but that it will hollow it out from the inside, gradually and invisibly, by removing the friction that builds cognitive capacity. The concern is not that a machine will think instead of us. It is that, by letting the machine do our cognitive work for us, we will become progressively less capable of doing it ourselves. That atrophy would happen gradually, masked by the fluency of our AI-generated outputs, invisible until the capacity is significantly diminished.

This is the GPS problem scaled to the entirety of intellectual life.

The fear is subconscious, I think, because it is difficult to articulate. Most people who use AI tools are not consciously aware of what they might be trading. The outputs look fine. The process feels efficient. The atrophy, if it is happening, is invisible in the short term. It only becomes visible at the level of populations over time, which is exactly the level at which it becomes very difficult to reverse.

Creation Is an Achievement. AI Produces Its Outputs Without Producing the Mind Behind Them.

Bloom’s Taxonomy places Create at the top rather than at the beginning for a reason. Creation, in the framework’s logic, is not a starting point. It is an achievement. It is what becomes possible after a person has developed the cognitive infrastructure that genuine creation requires.

A child who is handed a canvas and a set of paints and told to create something may produce something colourful. A painter who has spent years studying technique, observing the world, understanding composition, and developing the capacity to make and evaluate aesthetic judgements produces something different: not just technically more proficient but cognitively richer, because it emerges from a developed mind rather than an undeveloped one.

The AI tool can produce the canvas and the paints. It can produce something that looks like the painter’s work. What it cannot do is produce the developed mind. And what we risk, in making it too easy to produce the outputs of the developed mind without developing one, is a world full of plausible-looking canvases produced by people who have never learned to see.

Bloom’s hierarchy is not a bureaucratic framework for marking essays. It is a description of what human cognitive development is for. We should be careful about building tools that invert it, and more careful still about deploying those tools at scale before we understand what the inversion costs.

Sources

Anderson, L.W. and Krathwohl, D.R. (eds.) (2001) A Taxonomy for Learning, Teaching, and Assessing: A Revision of Bloom’s Taxonomy of Educational Objectives. New York: Addison Wesley Longman.

Bloom, B.S., Engelhart, M.D., Furst, E.J., Hill, W.H. and Krathwohl, D.R. (1956) Taxonomy of Educational Objectives: The Classification of Educational Goals. Handbook 1: Cognitive Domain. New York: Longman.

Dergaa, I., et al. (2024) ‘AI-chatbot-induced cognitive atrophy (AICICA): A conceptual framework’, cited in: Dergaa, I., et al. (2026) ‘AI-overdependence and human cognitive decline: Hazards, evidence, and mitigation strategies’, ScienceDirect. Available at: https://www.sciencedirect.com/science/article/pii/S2451958826001764 [Accessed August 2026].

Education Week (2025) ‘Teachers Worry AI Will Impede Students’ Critical Thinking Skills’, 24 October. Available at: https://www.edweek.org/technology/teachers-worry-ai-will-impede-students-critical-thinking-skills [Accessed August 2026].

Gerlich, M. (2025) ‘AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking’, Societies, 15(1), article 6. doi:10.3390/soc15010006.

Holmes, W. (2026) Comment in: UNESCO Courier (2026) ‘Learning to think in the AI era’, April. Available at: https://courier.unesco.org/en/articles/learning-think-ai-era [Accessed August 2026].

Online Learning Consortium (2025) ‘Using Bloom’s Taxonomy to Understand AI Adoption in Higher Education’, 3 December. Available at: https://onlinelearningconsortium.org/olc-insights/2025/10/blooms-for-ai-adoption/ [Accessed August 2026].

Pham, T.T.H., et al. (2025) ‘The Impact of AI on Students’ Reading, Critical Thinking, and Problem-Solving Skills’, American Journal of Education and Information Technology, 9(2). doi:10.11648/j.ajeit.20250902.12.

Tao, J., et al. (2024) ‘Promoting cognitive skills in AI-supported learning environments: the integration of Bloom’s taxonomy’, Educational Studies. doi:10.1080/03004279.2024.2332469.

National Education Union (2026) NEU Survey on AI and Critical Thinking in English State Schools. London: NEU, April 2026. Reported in: Reuters (2026) ‘UK teachers warn AI hurting students’ critical thinking‘, 2 April. Available at: https://www.anews.com.tr/world/2026/04/02/uk-teachers-warn-ai-hurting-students-critical-thinking-union-survey-finds [Accessed August 2026].