Can a machine write philosophy? The first contest opens, with Chalmers judging

Artificial intelligence is now extraordinarily good at mathematics and at writing code. And at philosophy?

Two philosophers at the National University of Singapore — Elliott Thornley and Zachary Goodsell — have decided to settle the question experimentally.

The AI Philosophy Competition, announced on 1 September, is the first of its kind. Entrants submit up to three original papers, on any philosophical subject, produced predominantly by artificial intelligence. Human steering is permitted up to a point; but each paper must arrive with a methods report explaining how the text was produced.

The closing date is 31 October 2026. Results are due at the end of January.

The panel shows how seriously this is meant#

What makes the competition hard to wave away is its panel.

The judges are David Chalmers, John Hawthorne, Herman Cappelen, Cian Dorr, Branden Fitelson, Kenny Easwaran, Matthew Mandelkern, Rachel Sterken, Daniel Greco, Jake Nebel and Christian Tarsney. The list will grow.

These are names from the present mainstream of philosophy of mind, metaphysics, philosophy of language and decision theory. The competition is not a marginal curiosity; it is being run from the centre of the field.

The prizes: $3,000 for first place, $2,000 for second, $1,000 for third. There is a further pot of $5,000 for creative method, left to the judges' discretion. The funding comes from Forethought Research.

Why the methods report matters#

The shrewdest design decision here is that the prize rewards not only the text but the method.

For the sentence "an AI wrote a paper" now means almost nothing. The difference lies elsewhere. Was a single prompt given, or were hundreds of drafts generated and sifted? Was the model set to argue against its own case? Were the sources checked?

Requiring a methods report makes that difference visible. And it concedes, by implication, that the text which emerges is also the product of a process a human being built.

The editor who ran the story and disagreed with it#

Justin Weinberg, the Daily Nous editor who carried the news, did something unusual: he published the announcement along with his own case against it.

Weinberg allows that some uses of artificial intelligence at the research stage may be acceptable — and may in time be advisable. But philosophy written predominantly by artificial intelligence he takes to be misdirected.

His reason for running the item anyway is worth noting. He thinks it valuable to know what the technology is capable of. And he adds that withholding the news would not slow the arrival of a future he does not want.

As journalism the posture is exemplary: report a development without endorsing it, and declare your own position plainly as you report it.

Two serious methodological objections#

The first comments beneath the announcement found the competition's blind spot at once.

The first objection. Every judge knows that the texts under assessment are the work of artificial intelligence, and that knowledge is bound to colour the verdict. What is more, unlike mathematics or code, philosophy has no objective and widely accepted criteria of what counts as good. The competition therefore does not measure where AI philosophy stands in relation to human philosophy; it only ranks entries against each other.

The second objection. The competition's website notes that artificial intelligence may be used for the initial sift if entries are numerous. So in a competition designed to measure the philosophical capacity of artificial intelligence, artificial intelligence may decide which texts reach the judges.

The second is the stronger. That a measuring instrument should be the very thing it measures is a familiar difficulty in the philosophy of science.

The real question lies elsewhere#

In this column we recently took up the arrival of artificial intelligence in the philosophy classroom and the ban on AI authorship at Philosophy & Public Affairs. This competition is the third act of the same argument.

And the argument's real subject is not whether a machine can do philosophy.

The real subject is this: what makes a philosophical text a good one?

If the answer is "the validity and originality of the argument", then who wrote the text is in principle beside the point — and the machine will clear the bar sooner or later.

If the answer is that a text must bear the trace of a problem someone genuinely struggled with, then authorship is no technical detail. It is part of what the text means.

The history of philosophy carries the marks of the second answer. Descartes's Meditations is not a chain of arguments but the record of a process of thinking. Wittgenstein's Philosophical Investigations sets down not its conclusions but the hesitations that lead to them.

The results to be announced in January will not answer that question directly. But which texts the judges choose — and on what grounds — will show what the field believes its own criterion to be.

The competition's real product, then, will not be the winning papers. It will be the judges' reasons.

---

Details: the conditions of entry and application information are at zacharygoodsell.com.