Will AI End Philosophy? A Physicist's Claim Meets a Philosopher's Reply

Louis Vervoort holds a PhD in physics from Aix-Marseille and a second PhD in philosophy of science from the Université de Montréal, with a postdoc at the École Normale Supérieure in Paris and a current post at the Higher School of Economics in Moscow. His 253-page open-access Springer book Problem Solving in Philosophy (Springer, 2026) — the subtitle signals how to do philosophy in the age of ultra-intelligent AI, released 22–23 May 2026, turns that biography into method: philosophy can no longer live on its ancient toolkit alone, it should proceed like natural science — theoretical physics in particular — solving problems through theory-synthesis as ultra-intelligent AI arrives.

At the core is the maximally coherent theory — the MCT. For well-posed questions, the best theory is the one that coherently answers the largest number of them, contradicts the least accepted knowledge and accommodates counterexamples most gracefully. The author proposes a semi-quantitative measure of solidity to be maximised, claiming philosophy can now approach scientific standards of certainty. The lineage runs through Quine’s web of belief and Ladyman and Ross’s naturalised metaphysics; the MCT is a synthetic relative.

A one-prompt ‘experiment’ on causality

The first case study is causality. Vervoort asked GPT-4 at the time which notion of causality is best; the model answered ‘counterfactual causality’, which matched his own preference, and he took that as confirmation. Sabine argues this is not an experiment in the physics sense, just tallying, and that calling a single chat reply a solution begs the question — it assumes the counting criterion that was meant to be justified.

The same template is applied to Gettier, the interpretation of probability, free will and induction. On this view free will clearly does not exist, induction is a pseudoproblem, while ‘big questions’ such as the meaning of life or the existence of God are held over — in Sabine’s quip, until a future ChatGPT build happens to agree. The reader sees the limit of the claim: not everything is declared solved, the deferred zones are explicitly marked.

Why induction is the litmus test

The paper that preceded the book already drew a reply — ‘Can we automate philosophy through AI? And should we want to?’ in Springer’s AI and Ethics. The reply draws a clean line: if you conceive philosophy as a set of propositions (PP), automation is in principle possible; if you conceive it as an activity of thinking through problems yourself (PA), it is not. Sabine jokes that this stretches the obvious to 15 pages, yet the distinction matters because it marks the philosophy-science boundary — two models trading theses can generate text, not a practice of inquiry.

Sabine’s deeper objection is internal: how to count the ‘best’ explanation is itself a philosophical problem, so the project of sciencifying philosophy contradicts its own ground. Philosophy is where we discuss questions science cannot yet answer; some migrate to science — consciousness is doing so now, free will may be next — but asking why induction works remains philosophical today. Vervoort’s move to declare induction an axiom that needs no justification confuses induction being an axiom for science with explaining why that axiom holds; the first does not dissolve Hume’s circularity.

Finally Sabine adds the point both sides miss: humans are lazy, most philosophy papers are already unread word piles and AI will take over that chore — it already is. That pressure might raise quality. She gives the book 6/10 on her bullshit meter: the quantitative coherence idea is worthwhile, but it does not make philosophy redundant; if AI did end philosophy, philosophers would spend fifty years debating whether it happened.