Language Log

« previous post | next post »

An interesting recent paper (Adithya Bhaskar, Xi Ye, & Danqi Chen, “Language Models that think, chat better”, arXiv.org 09/24/2025) starts like this:

THINKING through the consequences of one’s actions—and revising them when needed—is a defining feature of human intelligence (often called “system 2 thinking”, Kahneman (2011)). It has also become a central aspiration for large language models (LLMs).¹

The footnote:

¹Language models think, therefore, language models are?

There are no other references to René Descartes in the paper, or to ontological uncertainty. So let's follow up a bit.

The famous "Je pense, donc je suis" quote comes from the fourth chapter of Discours de la Méthode (1637), "Preuves de l'existence de Dieu et de l'âme humaine ou fondements de la métaphysique" ("Proofs of the existence of God and of the human soul or foundations of metaphysics"). It starts this way:

Je ne sais si je dois vous entretenir des premières méditations que j'y ai faites; car elles sont si métaphysiques et si peu communes, qu'elles ne seront peut-être pas au goût de tout le monde: et toutefois, afin qu'on puisse juger si les fondements que j'ai pris sont assez fermes, je me trouve en quelque façon contraint d'en parler.

I don't know whether I should discuss with you my first meditations on this topic; for they are so metaphysical and so unusual that they will perhaps not be to everyone's taste: and yet, in order that one may judge whether the foundations I have taken are sufficiently firm, I find myself somehow constrained to speak of them.

To make sense of this, we need to revisit the full title of the work, and the context of this chapter — "Discourse on the method for conducting one's reason properly, and seeking truth in the sciences. Plus Optics, Meteorology, and Geometry. Which are the tests of this method."

In other words, this is science and mathematics, not theology; or more exactly, this chapter explores the theological and metaphysical foundations for science and mathematics, which the author is a bit ashamed of discussing.

The whole first paragraph of section four (in the modern spelling found in the cited edition):

Je ne sais si je dois vous entretenir des premières méditations que j'y ai faites; car elles sont si métaphysiques et si peu communes, qu'elles ne seront peut-être pas au goût de tout le monde: et toutefois, afin qu'on puisse juger si les fondements que j'ai pris sont assez fermes, je me trouve en quelque façon contraint d'en parler. J'avais dès longtemps remarqué que pour les moeurs il est besoin quelquefois de suivre des opinions qu'on sait être fort incertaines, tout de même que si elles étaient indubitables, ainsi qu'il a été dit ci-dessus: mais pour ce qu'alors je désirais vaquer seulement à la recherche de la vérité, je pensai qu'il fallait que je fisse tout le contraire, et que je rejetasse comme absolument faux tout ce en quoi je pourrais imaginer le moindre doute, afin de voir s'il ne
resterait point après cela quelque chose en ma créance qui fût entièrement indubitable. Ainsi, à cause que nos sens nous trompent quelquefois, je voulus supposer qu'il n'y avait aucune chose qui fût telle qu'ils nous la font imaginer; et parce qu'il y a des hommes qui se méprennent en raisonnant, même touchant les plus simples matières de géométrie, et y font des paralogismes, jugeant que j'étais sujet à faillir autant qu'aucun autre, je rejetai comme fausses toutes les raisons que j'avais prises auparavant pour démonstrations; et enfin, considérant que toutes les mêmes pensées que nous avons étant éveillés nous peuvent aussi venir quand nous dormons, sans qu'il y en ait aucune pour lors qui soit vraie, je me résolus de feindre que toutes les choses qui m'étaient jamais entrées en l'esprit n'étaient non plus vraies que les illusions de mes songes. Mais aussitôt après je pris garde que, pendant que je voulais ainsi penser que tout était faux, il fallait nécessairement que moi qui le pensais fusse quelque chose; et remarquant que cette vérité: Je pense, donc je suis, était si ferme et si assurée, que toutes les plus extravagantes suppositions des sceptiques n'étaient pas capables de l'ébranler, je jugeai que je pouvais la recevoir sans scrupule pour le premier principe de la philosophie que je cherchais.

John Veitch's translation:

I am in doubt as to the propriety of making my first meditations in the place above mentioned matter of discourse; for these are so metaphysical, and so uncommon, as not, perhaps, to be acceptable to every one. And yet, that it may be determined whether the foundations that I have laid are sufficiently secure, I find myself in a measure constrained to advert to them. I had long before remarked that, in relation to practice, it is sometimes necessary to adopt, as if above doubt, opinions which we discern to be highly uncertain, as has been already said; but as I then desired to give my attention solely to the search after truth, I thought that a procedure exactly the opposite was called for, and that I ought to reject as absolutely false all opinions in regard to which I could suppose the least ground for doubt, in order to ascertain whether after that there remained aught in my belief that was wholly indubitable. Accordingly, seeing that our senses sometimes deceive us, I was willing to suppose that there existed nothing really such as they presented to us; and because some men err in reasoning, and fall into paralogisms, even on the simplest matters of geometry, I, convinced that I was as open to error as any other, rejected as false all the reasonings I had hitherto taken for demonstrations; and finally, when I considered that the very same thoughts (presentations) which we experience when awake may also be experienced when we are asleep, while there is at that time not one of them true, I supposed that all the objects (presentations) that had ever entered into my mind when awake, had in them no more truth than the illusions of my dreams. But immediately upon this I observed that, whilst I thus wished to think that all was false, it was absolutely necessary that I, who thus thought, should be somewhat; and as I observed that this truth, I think, therefore I am (COGITO ERGO SUM), was so certain and of such evidence that no ground of doubt, however extravagant, could be alleged by the sceptics capable of shaking it, I concluded that I might, without scruple, accept it as the first principle of the philosophy of which I was in search.

(The Latin version "Cogito ergo sum" comes from his 1644 Principia Philosophiae.)

So Bhaskar et al.'s footnote (jokingly?) hints that current AI chain-of-thought reasoning might eventually lead to a foundational premise on the basis of which non-hallucinatory conclusions can reliably be derived.

That's the first time I've seen this idea suggested. I'm skeptical that things will work out that way, unless maybe the systems are primed to parrot La Méthode. But we'll see…

October 3, 2025 @ 6:22 am · Filed by Mark Liberman under Artificial intelligence

Permalink