# Today ## 1. draft of generating philosophy section 1 One challenge to the claim that LLMs can do philosophy begins from the thought that philosophy needs a philosopher. Even if philosophy often reaches us in writing, one might still hold that a text counts as philosophy only when it stands in the right relation to a philosopher's activity. An imperfect comparison would be with art. Someone may grant that an image is visually striking and still deny that it is an artwork unless it issues from artistic agency%%not how i write%%; likewise, one may grant that a piece of writing contains recognisable arguments, distinctions, and objections and still deny that it is philosophy unless it is the work of a philosopher. On some conceptions of philosophy, this first challenge is decisive from the outset. If philosophy is a practice of self-transformation, as Hadot (1995) holds, or a form of therapy, as it is for the later Wittgenstein (1953), then philosophy is not merely something that leaves a trace in texts. It is an activity lived through by a philosopher. Nietzsche's picture, as Sorgner (2017) reads it, moves in a similar direction: the philosopher's work expresses drives, valuation, and a psychophysiology bound up with human embodiment. If views of this sort are right, the question whether an LLM can do philosophy is settled before we reach the page.[^pr] Still, that is not how contemporary analytic philosophy ordinarily conducts its business. Papers are written, anonymised, sent to referees, and judged for the distinctions they draw, the objections they answer, and the theories they make available for public scrutiny. What matters under these conditions is what a text enables a competent reader to see. This is why the text-oriented conception has genuine force. Dellsén et al. (2024) argue that philosophical progress consists in putting people in a position to understand phenomena better by grasping the dependence relations in which they stand, or fail to stand, to others. That picture fits the discipline's ordinary practice rather well: the argument is made public, the reader works through it, and the philosophical success or failure of the performance is assessed from what the text itself makes available. Still, the challenge from authorship remains even once we grant all that. One may accept blind review, accept that philosophical texts are assessed through what they put before readers, and still insist that such assessment presupposes a further relation to a philosopher. On this view, the anonymous paper counts as philosophy only because it is the expression, continuation, or residue of philosophical activity, even when the referee brackets the author's name. A language model could then reproduce philosophy's textual shape while missing the relation that gives such writing its standing as philosophy. The question, then, is whether recognisable cases of philosophical success support that thought. Putnam's Twin Earth case is a good place to begin. --- --- - One might endorse a text based approach to philosophy yet still deny that LLMs are capable of producing philosophical texts. - In particular, it could be argued that philosophical texts need to be *authored* - that whatever words an LLM spits out cannot be philosophy, because philosophy needs a person, or at least a *mind* of some sort, behind its creation. - An imperfect comparison would be with art: regardless of how aesthetically pleasing an image (or a song, text, poem etc.) produced by an AI system is, it does not count as art. - a lot of people, perhaps a majority, would argue that generative AI systems, in virtue of their not being people, cannot make art. This would not be to say that AI outputs cannot be aesthetically pleasing, only that they are not artworks: on this view, art is the product of the right sort of mental activity on the part of its maker — imaginative, expressive — and a system without mental states has no such activity to draw on. In this section we address what we might call the challenge from authorship: the idea that philosophical texts must be created by people, or at least some sort of mind. The introduction set aside practitioner-based conceptions of philosophy in favour of a text-based approach. The challenge from authorship grants this text-based approach and makes a different claim: that a text's standing as philosophy depends on who produced it. A philosopher who assesses philosophical quality through published texts, and who relies on blind review as the discipline's standard practice, might still hold that a text produced without understanding does not count as philosophy, however good it is. An imperfect comparison would be with art: a lot of people, perhaps a majority, would argue that generative AI systems, in virtue of their not being people, cannot make art. This would not be to say that AI outputs cannot be aesthetically pleasing, only that they are not artworks: on this view, art is the product of the right sort of mental activity on the part of its maker — imaginative, expressive — and a system without mental states has no such activity to draw on. # Actions ---