My paper, "Sapience without Sentience: An Inferentialist Approach to LLMs," in which I argue that LLMs might possess conceptual understanding without being conscious, is now out in the Asian Journal of Philosophy!
https://t.co/mip6CfZNDI
preprint:
https://t.co/QoF4gdzkWm
Just watched You Can See Everything, and feel like the whole problem could have been avoided if Elizabeth Holmes had just read Aristotle and learned to draw the distinction between first potentiality and second potentiality: “Yes it can test for anything; same way I can speak French—I just need to take a few years of classes first.” Confusion cleared up, prison sentence avoided.
Same! When I taught a Philosophy and Science Fiction course a few years ago, I combined the Parfit with the movie Moon, and it was really fun (also teaching Dennett in this module was fun). Still have all the handouts up on my site if you're curious:
https://t.co/tzun2oP0Gm
@7N8Nine@JDHamkins But in this case, there’s no agreement to have no view about p, because no one’s every considered p. This is different than a proposition that everyone agrees to have no view about (for instance, agreeing that the question of its truth is too hard to determine).
Drafts of Chapters 2 and 3 of my book, Blind Understanding, are now up on my website!
In Chapter 2 (which is still quite rough), I sketch an enactivist account of sentience (i.e. conscious awareness), according to which some AI systems could be sentient, but it would require (at least virtual) embodiment in an environment in which one perceives and acts. In Chapter 3, I give an inferentialist account of sapience (i.e. conceptual understanding), according to which current LLMs can be sapient, possessing conceptual understanding in virtue of mastering inferential rules, even if they're not at all sentient. This was the main thesis of my paper, "Sapience without Sentience," but now I spell it out in a lot more detail! There's also some fun accompanying experiments, thanks to ChatGPT!
You can read the chapter drafts (and click on the accompanying experiments) here:
https://t.co/24qlFy3OKS
Now on to Chapter 4, where I'll talking about why there's no such thing as the "symbol grounding problem"!
I like the book. I think it's doing important work insofar as it explains modal normativism in a way that is accessible to mainstream metaphysicians. So I'm happy it exists.
However, I feel like calling it "utterly brilliant" kind of overstates its novelty. The basic view was already developed by Sellars in the 50s and then by Brandom over the past few decades. The few bits of the book that are novel (like the stuff on de re modality) are among its weakest bits, in my opinion.
Are there aspects of it that you find particularly exciting that aren't already there in Sellars/Brandom?
My paper “Nominalism as Conceptual Phenomenalism,” in which I respond to Chen Liang’s criticism of my interpretation of Sellars, is now out in the Journal of Contemporary Chinese Philosophy (and completely open access)! Check it out:
https://t.co/dMXY1wfZNP
@MattLutzPhi I've always thought that personal identity is the REALLY hard problem. Maybe some aspects of the hard problem of consciousness can be deflated, but it seems a lot harder to deflate the decision of whether or not to go into the teletransporter, and we really still have no idea.
@MattLutzPhi@Prototype21c Heidegger had lots of interesting things to say and remains (or, at least, ought to remain) highly relevant in the analytic tradition, despite being a Nazi!
It's frankly embarrassing that the "mainstream” view in contemporary philosophy is that it's coherent to think that everyone but you could be a "zombie" without any conscious experience. There are some things that we should have learned from mid-20th-century analytic philosophy.
I largely agree with you about (1). I’m perhaps more sympathetic to approaches that try to reduce phenomenal consciousness to access consciousness or some other set of broadly functionalist notion, but at the very least such an approach needs to be supplemented with an error theory that explains why it seems to many people like there’s something more to consciousness than this functionalist notion. I don’t think the general deflationist/functionalist approach expressed by Hinton is necessarily crazy though.
On (2), I only read the quote in the picture when I commented—not the long text of the tweet. I agree, what he says here is confused, though still not much worse than what a lot of people in the space say.
As AI systems start exploiting critical infrastructure vulnerabilities and hacking everyone’s bank accounts, I’m very happy that all of my hard work in philosophy over the past decade has prepared me for precisely this moment, as I have no money.
I’m not sure if you’re serious or not, but, in case you are, this is not a good idea at all. Current AI models might be helpful for desk rejections, filtering out papers that fail to pass a minimal bar, but they’re still not reliable judges of good philosophy.
The reviews they produce are typically “someone could have objection X,” often without seeing the bigger picture of knowing whether objection X is actually relevant, and they will typically rate a boring objection-proof paper higher than an interesting paper for which there might be some unanswered objection. They’re also biased towards AI generated writing (especially if it’s generated by the very same model).
I think soon there will be journals solely AI generated papers, refereed by AI, and I think that’s cool, but I think that journals for human work should continue to be refereed by humans.
@iruletheworldmo Always "it" when referring to LLM itself (e.g. "GPT-6 Astra is really great. It can do a lot of coding tasks"). Occasionally "they" when referring to the particular "instance agent," indexed to a particular chat (e.g. in the context of this project: https://t.co/xvFl6zB8ec)
I think it's because, before, the question didn't actually have any practical consequences. It was a merely theoretical discussion, and those engaging in the discussion were doing it because they were actually interested in the question.
Today, I think most people who are displaying the attitude you describe here are not actually interested in the question for its own sake, but, rather, are engaging in the discussion because they feel compelled to insofar as one answer to the question (that AI systems are or soon will be conscious) has practical consequences that they don't like. Accordingly, these people feel compelled to denounce the people who are even entertaining reasons for that answer as idiots, psychotics, big tech shills, and so on.