Events

Jessie Hall, University of Toronto

PIZZA WILL BE SERVED!

Large language models (LLMs) like ChatGPT, Claude, Grok, Copilot, Gemini, etc. have taken the popular imagination–and the headlines– by storm. The linguistic capabilities of LLMs and the broad swath of applications to which they have been so far applied has provoked a flurry of speculation and expectation for a future shaped–and even defined– by artificial intelligence (AI).

Speculation and expectation about artificial intelligence is a pastime with considerable precedent in philosophy. Long before even the advent of computability theory, philosophers have speculated on whether a machine could think– whether intelligence could be artificially instantiated or constructed– and what it might mean to say that it can or cannot. In this talk, I will situate the current excitement over LLMs within the history of philosophy of AI. In particular, I will consider notable arguments concerning whether computing machines could think or be intelligent (exemplified by e.g. John Searle’s 1980 ‘chinese room’ thought experiment, or Hubert Dreyfus’ 1972 book, What computers (still) can’t do). I will ask to what degree such arguments still apply: what, if anything, is different about LLMs as compared with previous iterations of the technologies we have called ‘artificial intelligence’? I will offer a tentative answer in two parts.

First, I will argue that there are important ways in which LLMs suffer similar critical shortcomings with respect to attributions of intelligence as those of its technological predecessors. Second, however, I will argue that there are respects in which LLMs and comparable machine learning technologies– especially as they are understood and deployed– represent a considerable departure from their technological forebears. I will argue, in short, that what LLMs understand is, arguably, nothing; but what we understand from and about LLMs warrants urgent and careful examination