Philosophy of AI
What current AI systems are, described without inflation or dismissal.
Philosophy of AI, as I practise it, is conceptual work on the systems we already have. The question is not whether machines could one day think, but what the systems now deployed are doing when they produce sentences that look like assertions, and what we are entitled to make of them.
That question has to be answered without inflation or dismissal. Probing studies find internal structure in language models that correlates with the truth of what they represent, and interventions along that structure can make outputs more truthful. That is genuine evidence that something relevant exists. What it does not yet show is that the structure governs production when no one is intervening — a lever that moves the output when pulled is not thereby what steers it when left alone. My claim is not that no machine could be a source of testimonial knowledge, but that these machines, as they are now built and as we now understand them, are not; and the argument states precisely what would have to be shown for the verdict to change.
A second strand, in a paper under review at Philosophy & Technology, concerns the legitimacy of foundation models.