AI and the TOK Essay: Knowledge Questions Every IB Student Should Explore in 2026

TOK is about how we know what we claim to know — and AI is generating new instances of that question faster than almost any other field. When DeepMind’s AlphaFold 2 predicted over 200 million protein structures with accuracy comparable to experimental methods, it did not simply produce data. It raised the question of whether a machine can produce scientific knowledge, or whether it produces something else that scientists then verify as knowledge. That is a TOK question.

When a US law firm submitted court documents citing legal cases that had been fabricated by ChatGPT — and ChatGPT confirmed those cases existed — the problem was not just technical. It was epistemological: what happens when the form of knowledge (authoritative language, structured citation, confident delivery) is perfectly imitated without any underlying truth? That is also a TOK question.

AI does not make your essay easier. What it does is provide contemporary, well-documented, cross-disciplinary examples that can ground arguments in any area of knowledge. The key is using AI as a lens for exploring knowledge questions — not as the topic itself.

The IB does not ban AI. Its official position states: “The IB is not going to ban the use of such software but will work with schools to help them support their students on how to use these tools ethically in line with our principles of academic integrity” (IBO). However, the policy draws a clear line: work produced by AI tools — even only in part — is not considered the student’s own and must be credited in the body of the text and referenced in the bibliography.

There are two separate questions here that students often conflate:

1. Can I use AI to help write my TOK essay? — Limited. You may use AI to explore ideas, generate counterexamples, or check your understanding. But the reasoning, structure, and argumentation must be yours. The TOK essay is assessed on your independent critical thinking. An essay that is predominantly AI-generated will not receive marks for analytical depth, because that depth must be your own.
2. Can AI be the subject of my TOK essay? — Yes. AI is a rich source of knowledge examples and raises genuine epistemological questions relevant to any prescribed title. Writing about AI as a knowledge phenomenon is entirely different from using AI to write.

This article focuses on the second question: AI as a topic and source of examples — not AI as a writing tool.

AI is not confined to one area of knowledge. It raises distinct epistemological questions in each of the five AOKs — which means it can serve as a connecting thread no matter which title you choose.

Natural Sciences

AlphaFold 2 predicted protein structures with accuracy comparable to experimental methods — raising the question of whether a machine can produce scientific knowledge, or whether it produces outputs that scientists then verify as knowledge. The researchers behind AlphaFold received the 2024 Nobel Prize in Chemistry (Nobel Prize). Knowledge question: Does AI produce scientific knowledge, or does it produce predictions that scientists then verify as knowledge? Is there a difference?

Human Sciences

AI systems trained on human data can reproduce and amplify societal biases — raising questions about whether AI-generated insights about human behaviour constitute knowledge or artefacts of the training data. Knowledge question: When an AI produces a finding about human behaviour by analysing patterns in data, is that a discovery or a reflection?

The Arts

Jason Allen’s AI-generated artwork won the 2022 Colorado State Fair digital arts competition and was subsequently denied copyright protection — opening questions about authorship, creativity, and what “artistic knowledge” requires (Smithsonian Magazine). Knowledge question: If a work produces an aesthetic response but was not created with human intention, does it constitute artistic expression — and does the answer change what we think artistic knowledge is?

Mathematics

Large language models can produce correct solutions to many mathematical problems, yet also produce confidently wrong answers in others — sometimes failing basic arithmetic while succeeding at more complex proofs. A growing body of research on large language model reasoning has documented that LLMs can appear to “reason” through mathematical steps without having any underlying understanding of the concepts involved. Knowledge question: Does producing the correct answer to a mathematical problem constitute mathematical knowledge — or does knowledge require understanding the why?

History

AI trained on historical data can generate historically plausible narratives — raising questions about how we distinguish historical knowledge from historically-shaped speculation, and whether knowing who produced a historical account matters. Knowledge question: If an AI can produce a historically plausible account, does studying the “historian” (the algorithm and its training data) before studying the work change how we evaluate it?

Each example below is a documented, citable event that raises a genuine knowledge question. They are presented as prompts — what you do with them in your essay is the work that earns marks.

1. AlphaFold 2 and the protein folding problem

In 2020, DeepMind’s AlphaFold 2 solved a 50-year-old challenge in biology — predicting the 3D structure of proteins with atomic-level accuracy. The result was so significant that Demis Hassabis and John Jumper were awarded the Nobel Prize in Chemistry in 2024 (Nobel Prize). Over 200 million protein structures are now freely available via the AlphaFold Protein Structure Database, used by more than 3 million researchers in 190 countries (DeepMind).

The knowledge question it raises: Did AlphaFold produce scientific knowledge, or did it produce predictions that scientists then verified as knowledge? Is there a difference — and does it matter?

2. Mata v. Avianca and AI hallucination

In 2023, US lawyers representing Roberto Mata filed court documents citing multiple legal cases that did not exist — they had been fabricated by ChatGPT. When questioned, ChatGPT confirmed the cases “indeed exist” and could be found in reputable databases. The court sanctioned the lawyers $5,000 (Mata v. Avianca, 678 F. Supp. 3d 443, S.D.N.Y. 2023).

The knowledge question it raises: If an AI produces a confident, detailed, authoritative-sounding claim that is entirely fabricated, what does this tell us about the relationship between the form of knowledge and its truth? Is the allure of an AI-generated idea different in kind from the allure of a human idea — or just in degree?

3. Théâtre D’opéra Spatial and AI authorship

In August 2022, Jason Allen won first place in the digital arts category at the Colorado State Fair with an image generated using Midjourney AI after entering at least 624 text prompts and refining the output in Photoshop. The judges were unaware it was AI-generated but said they would have awarded it first place regardless. The US Copyright Office later ruled the work could not be copyrighted because of the extent of AI involvement (Smithsonian Magazine).

The knowledge question it raises: If the work produces an aesthetic response and the judges judged it as art, does it matter how it was made? What does this tell us about whether artistic knowledge is in the object, the process, or the response?

4. Large language models and mathematical “reasoning”

Large language models can produce correct solutions to many mathematical problems, yet also produce confidently wrong answers in others — sometimes failing basic arithmetic while succeeding at complex proofs. Researchers have documented that LLMs can appear to “reason” through mathematical steps without having any underlying understanding of the concepts involved. This is distinct from AlphaFold, which operates on pattern prediction in structural data.

The knowledge question it raises: Is producing the correct answer to a mathematical problem sufficient for claiming mathematical knowledge — or does knowledge require something the answer alone cannot show?

Several of the November 2026 prescribed titles open naturally to AI-grounded arguments. Below are four connections — each framed as a question for you to explore, not an answer to adopt. The full operative word analysis and recommended AOK pairings are in our complete title breakdown.

Title 2 — Failure and the production of knowledge

“To what extent do you agree that failure is an essential part of the production of knowledge?”

AI systems learn through failure — error minimisation in machine learning is a form of institutionalised failure. The question this opens: If failure is essential to knowledge production, what counts as failure when the “knower” is not human?

Title 3 — Ideas more alluring than facts

“In the production of knowledge, why is it that ideas are so often more alluring than facts?”

AI-generated content produces fluent, confident, alluring outputs regardless of factual accuracy. The Mata v. Avianca case — where ChatGPT invented court cases that looked authoritative — is a live example. The question this opens: Does AI change the relationship between the allure of ideas and the reliability of facts — and does that matter for how knowledge is produced in the human sciences?

Title 4 — Ethics in art and natural science

“To what extent do you agree that the artist and the natural scientist should be equally concerned with ethical questions?”

When Jason Allen’s AI-generated artwork won a state fair competition and was later denied copyright, it raised questions about artistic authorship and ethical responsibility. The question this opens: If the artist is a human who directed an AI, and the natural scientist is using AI to generate hypotheses, are their ethical responsibilities equal — or does AI shift where responsibility lies?

Title 5 — Sharing knowledge

“Does the need to share knowledge pose challenges in the production of knowledge?”

AI dramatically accelerates the dissemination of knowledge — but also the dissemination of misinformation. The question this opens: When knowledge can be produced and shared by AI at scale, does the “need to share” still pose a challenge — or does it now pose a fundamentally different kind of challenge?

For Title 1 (the historian and their work) and Title 6 (intuition in mathematics), AI also opens productive angles — the AOK connections in the section above map directly. The full title-by-title breakdown, including operative word analysis and recommended AOK pairings, is in our November 2026 Prescribed Titles guide.

The mistake is using AI as the subject of the essay rather than as an example within the essay. The TOK essay does not ask you to write about AI. It asks you to explore a knowledge question — about failure, intuition, ethics, ideas, sharing, or the producer of knowledge. AI is a lens and a source of examples. It is not a title.

A student who writes 600 words about how AI works and then 200 words about the knowledge question has written the wrong essay. The question drives the essay. AI serves the argument.

The test: At any point in your essay, ask yourself: am I currently making a claim about knowledge — or am I describing technology? If you are describing technology for more than two sentences, you have drifted from TOK into a different kind of writing.

A related pitfall: using AI examples without a knowledge claim. Saying “AlphaFold solved protein folding” without asking what that means for how we understand scientific knowledge is a missed opportunity. Examples must serve arguments. Arguments must address the title.

Where to Go From Here

AI opens genuine knowledge questions across every area of knowledge — and several of the November 2026 prescribed titles respond directly to those questions. But the essay that earns top marks is not the one with the most impressive example. It is the one that uses examples to make sustained, critical arguments about knowledge itself.

Start with the title. Identify the operative words. Then ask: does an AI example help me answer this particular question about knowledge? If it does, use it. If it doesn’t, use something else. The title drives the essay. Everything else — including AI — serves it. If you want structured support from an experienced TOK educator, we can help you move from ideas to a polished draft.