AI Philosophy Competition: $11,000 in prizes.
AI is now exceptionally capable in mathematics and coding, but how good is it at philosophy? We are organising the first AI Philosophy Competition to find out.
Entrants may submit up to three original essays on any philosophical topic. Essays must be primarily AI-generated, though some human guidance is allowed. Entrants must also submit a methodology report explaining how each essay was produced.
The essays will be judged by a committee of philosophers including David Chalmers, John Hawthorne, Matthew Mandelkern, Rachel Sterken, Herman Cappelen, Branden Fitelson, Kenny Easwaran, Hilary Greaves, Daniel Greco, Cian Dorr, Jake Nebel, and Christian Tarsney, with more to be announced.
The prizes are $3,000 for first place, $2,000 for second, and $1,000 for third. There is also a prize pool of $5,000 for creative methodologies, to be distributed as the judge committee sees fit.
The submission deadline is 31 October 2026. Results will be announced at the end of January. The competition is organised by Zachary Goodsell and Elliott Thornley. It is funded by Forethought Research. Full details are available on the competition website.
should be a lot of fun!
Is already published work eligible? Looks like not, or I’d submit this: https://link.springer.com/article/10.1007/s13347-025-00975-5
“DM suggested the topic and wrote the original outline. Literature reviews were conducted with the help of OpenAI Deep Research/GPT 4.5, and Elicit. The outline was critiqued by Claude 3.7. The metaphor of the hall of mirrors and the comparison to Rorty’s critique was suggested by GPT 4o. The generation of the ideas and the initial draft was iteratively done by DM, Claude Opus 3.7, Grok 3, and GPT 4.5. Most sections were drafted by ChatGPT 4o, except for the technical grounding, which was written by Grok 3, with revisions by DM and ChatGPT4o, …”
Unfortunately not. We’re looking for work that hasn’t been published or submitted elsewhere. Note also that we want the ideas/arguments in the essays to come from AIs themselves. Here are some examples of methods that are/aren’t eligible for prizes, from the rules page:
Unfortunately it is very easy to cheat by giving the LLM one’s own ideas and then letting it do the write-up, or a rewrite. This is undetectable because Pangram can only check whether the final text was written by a human, not whether he or she provided the ideas to the LLM.
No, that won’t work since they require the full LLM transcripts.
I think this actually raises interesting question about how to interpret AI outputs. A similar issue seems to also have come up in the OpenAI Navier-Stokes kerfuffle. The initial statement from Tristan Buckmaster mentions this:
Even with access to transcripts of a given interaction with a model, I think it is still note entirely clear that we can fully understand how AI was used or have all the information needed to interpret the results.
I think this has been a consistent trend as model have gotten more capable. In the days of classic ML you could have in theory collected new data that was guaranteed to be held-out and run the model on that. “Generative” AI and training on large datasets from the internet brought of issues of “contamination” and raised more questions about whether labs were quietly training on benchmarks. Now “agentic” AI is creating new layers of complexity in interpreting model outputs.
To make clear the application to something like the competition mentioned here, what if you have a professional philosopher write some novel essay, then ask an AI to attempt to find prompts that appear innocuous but that illicit a response very similar to the expert-created essay when feed to an LLM? You might not be able to detect such a case simply by reviewing the transcript of the session that produced the essay.
I’m not particularly worried that a professional philosopher would hand over an essay for an AI to take credit, nor that you could write a prompt that was anything like innocuous enough to get it to recapitulate the details, but I agree that for very high-stakes areas, this is far more worrying; this is partly because we don’t have any level of assurance about almost anything about these systems!
I agree, I meant my example above as an extreme hypothetical to demonstrate the issue, but I agree this is unlikely to actually happen for this competition.
I’m there with you on that. I think its worrisome that as systems have gotten more advanced it seems like the increased complexity of AI systems has implicitly been accepted as an excuse for lack of insight by model developers into how these systems work.
That’s not so clear:
So methodology reports are only used for “unclear” cases, and chat logs are not mandatory, just recommended.
I’m tempted to enter, but I’m also worried about capabilities externalities. I would love to hear people’s thoughts.
are you worried about (i) the mere sin of LLM usage, or (ii) that some innovative prompting technique may come to light?
for (i), it is perhaps mitigated by the 400 trillion tokens per day currently processed? for (ii), perhaps better revealed now when the models are at the least capable they ever will be again?
I’m very skeptical of overhang arguments. Instead of burning through the overhang, we seem to just build a new base upon which further advances are then developed.