Rosa Cao

@cameronjbuckner @Ensifere Mazviita Chirimuuta has a great book review of The Multiple Realization Book (Polger & Shapiro) where at the end she reframes the MR debate for those whose primary concerns are scientific and explanatory rather than metaphysical. tandfonline.com/doi/full/10.10… (especially Section 4.1)

@Ensifere I think mechanistic approaches render much of the old MR debates in phil mind moot. If you want one of my favorite papers on MR, read this if you haven't already: jstor.org/stable/188642 . And the line between phil mind and phil science has been porous for at least fifty years.

Philosophers of science over the past thirty years developed the best accounts of multiple realizability, neural computation, and abstract mechanistic explanation ever, but couple of weeks with GPT-4 and people regress to naive behaviorism and proclaim "alien intelligence" 🤦

@gualtieropicc Lots of people have raised that question (e.g. Bill Ramsey). I think the answer depends on pragmatic considerations given by the explanatory context, and especially the robustness of the putative representations attributed (details here) dropbox.com/s/uhc54i002ugc…

@gualtieropicc Totally agree that there is interesting theoretical work for representations to do! The question is whether the activation patterns commonly identified by neuroscientists as "representations" actually play that role.

In a recent comment, @luosha asked what theoretical work can neural representations do. As I see it, there are three main mental phenomena that need explaining: intelligence, intentionality, and (phenomenal) consciousness. 1/6

@spiantado *implicit IN a good curve fit

@spiantado If you build the theory in and then re-write as above, the theory is certainly "in there". But that doesn't seem to guarantee that some theory will ALWAYS be implicit a good curve fit, whether you built it in explicitly at the beginning or not.

Large language models change everything for linguistics, starting with Chomsky. Featuring: LLMs as scientific theories, response to prior takes, "why" questions in language, acquisition... and how the field should have seen this coming. Paper is here: lingbuzz.net/lingbuzz/007180

@neuralreckoning Rick Grush has a great paper on the relevance of this q for comp neuro. and here's a book dedicated to your question philpapers.org/rec/GRUTSC books.google.com/books?id=HfhQE…

Naive question (maybe): Is there a definition of 'computation' akin to the mathematical definition of information (entropy/MI)? I don't mean Turing machines. e.g. something that could determine the extent to which a group of neurons/synapses are signalling versus computing?

@davidchalmers42 @renatrigiorese i suspect that once we get to mouse-level AI, the distance to human-level AI will be tiny. but getting to mouse-level AI requires duplicating perceptuo-motor capacities that are crucial to mouse function but difficult to make without biological bodies

@renatrigiorese one motivation: i'm interested in when if ever we might be said to have "mouse-level intelligence" in an AI system, and what the greatest challenges are in getting there.

what cognitive capacities does a mouse have that no current AI system has? (extra points for intelligence-related capacities with associated behavioral tests. the AI system can have a virtual/robot body if necessary. capacities like eating that require a bio-body don't count.)

@davidchalmers42 current "AI" systems only respond to inputs in pre-determined formats (just like mice do, given the perceptual systems that they have), but unlike in mice, those inputs don't hook up to general motivational machinery

@davidchalmers42 the ability to map a wide range of novel stimuli to approach/avoid, reward, and attentional responses in ways that tend to keep the mouse intact