Rules on the Use of AI

Preambule

Artificial intelligence has become an ordinary working tool and a part of everyday life. We want it to be a benefit to FYKOS as well — and we want FYKOS to remain what it is. That is why we are introducing this Code, which lays down binding rules for the use of AI.

FYKOS is an educational program: we set problems so that solving them moves you forward, not so that we find out the answer. FYKOS is also a competition: the rankings, the camps, and other rewards mean something only if what we are comparing is the work that you have done. This Code says what, in solving a problem, has to remain yours.

This Code rests on mutual trust. We believe that you come to FYKOS with the right motivation — not merely to submit the correct result, but to genuinely understand a problem and solve it.

The Purpose of Solving FYKOS Problems

Solving a problem involves both the physics work itself and a number of activities that are not physics. AI plays a different role in each of them.

The physics parts. In some problems, the core of a correct solution is deciding what matters and what can be neglected, and choosing a physical approximation that still holds. In others, the core is describing the given situation in mathematical terms, carrying out the calculation, and recognizing whether the result makes sense. Both rest on physical — and sometimes mathematical — thinking, which you can develop only by taking these steps yourself, including by getting something wrong and working out why.

The non-physics parts. Solving a problem often also involves background research, writing up your reasoning, and drawing figures. Writing up a complete solution is not a side matter: explaining your own reasoning clearly is a skill you will use far beyond physics, and FYKOS lets you practice it with personal feedback. That feedback is worth something only when the grader is commenting on your own thinking.

Principles for the Use of AI

For solving problems to fulfill the purpose described above, every submitted solution must satisfy the following four principles.

P1 — The participant's own physics work. The physics solution to a problem must be the participant's own work. AI must not be used to solve a problem, or to carry out key steps of its solution in place of the participant, not even where the participant afterward puts the reasoning obtained this way into their own words. The same limits apply to AI as to consulting another person.

P2 — The participant's own text. The participant formulates the text of the solution in their own words; whole sentences and paragraphs must not be generated, not even from the participant's own notes. Language proofreading, advice on the style of text that has already been written, and translation of a finished solution into a competition language are admissible.

P3 — Verification and sources. Whatever the participant takes from AI must be verified against the original source or by their own derivation, and that source must be cited. AI is neither a source nor a co-author; it is only a tool. The participant bears full responsibility for the submitted solution, including any errors introduced by the tool.

P4 — Declaration of use. If AI was used in solving a problem, the participant includes in the solution a brief note on what it was used for and to what extent. A few sentences are enough — not a transcript of prompts. Declared use that is in accordance with this Code is not a reason to reduce the score; undeclared use is a breach of the Rules.

Where the Line Is

The principles are stated in general terms. To get a more concrete grip on them, ask yourself three questions about your solution — if the answer to any of them is no, that is a reason to stop and look at it again.

Defense.

Could you explain your solution at the board without notes, including where each assumption came from?

Authorship.

Did you work out for yourself what matters in the problem?

Verification.

Have you checked everything you took from elsewhere?

An example. The open problem the boat is sailing (Year 36, Series 4): Discuss what physical phenomena affect the cruising speed of a ship and submarine. What resistive forces act on them? What is the highest cruising speed that a ship or submarine can sail? The table below gives examples of prompts to AI that are fine, and prompts that are already over the line.

The difference is not in how much AI writes, but in who does the essential work. The right-hand column also contains questions that look like background research — except that the answer to them is exactly what the problem is meant to test.

Background research. Looking for literature is part of solving a problem, and AI will help you with it, as long as the research itself is not what the problem is meant to test. The open problem bread (Year 38, Series 3) reads: Determine the inner surface area of all the cavities in a sourdough loaf. It is fine to ask for studies that measure the porosity of baked goods and then take from them what you need. It is over the line to let AI estimate the inner surface from the available studies: in doing so, you hand it both the choice of method and the reasoning about how to get from the quantities that are actually measured — typically the distribution of pore sizes and their volume, not their surface area — to the result.

Mathematics. Solving a complicated equation is usually not what a problem is meant to test, and you do not have to be able to evaluate every integral or differential equation you run into. Getting help with them is fine, as long as you declare it in accordance with P4 and do not lose control of your solution: asking whether an equation can be solved is a different matter from leaving the discussion of the result, or the choice of the physically meaningful solution, to AI. The line also runs through how much you put in front of AI — it will sometimes finish the problem for you even if you describe the situation in mathematically unsuitable terms or fail to make the necessary simplifications. In that case, though, the essential work was not done by you. After consulting AI, you should also know how to approach a similar equation next time.

Always over the line: letting AI choose the model and the assumptions; having it generate a derivation, an interpretation of the result, or the text of the solution; taking over citations you have not verified; letting AI process your measured data from start to finish; inventing or altering measured data.

Usually fine: having a concept you do not know explained to you; looking for literature that you then read and cite; translating a foreign-language source; LaTeX, TikZ, language proofreading; help with syntax and with parts of the code for processing your own data or for a simulation.

It is also worth saying that if you cannot work out how to solve a problem, it is fine not to submit it, or to submit only part of it. Presenting someone else's answer as your own is not fine.

Grading

To close, here is a summary of how using AI affects the grading.

We grade what is written. A solution whose reasoning is not apparent, or which fails to get its point across, will receive few points regardless of how it came about.

Declaring your use of AI works in your favor — it lets the grader write you a more useful comment. By itself, it never reduces your score. Where there is a reasonable suspicion that this Code has been breached, we proceed instead according to the Breach of Rules section of the Rules.

What do we promise in return? The grading of your solution and the comment on it are always written by a human. We write the problem statements and the solutions we publish by the same standards we expect of you.

Why This Matters to Us

The rankings are only one part of FYKOS; the more important part is what you take away from solving the problems. Besides the competition itself, we also run camps and other events where people who enjoy physics meet — people who, together with you, make up a wider community. By following the FYKOS AI Code of Conduct, you help build the culture of that community.

This Code is inspired by the AI usage rules of the KSICHT correspondence seminar, and we will update it as these tools and their practical applications evolve.

Comments and questions about this Code can be sent to fykos@fykos.org.

Approved by the organizers of FYKOS on September 1, 2026