Thank you, I was searching for this lecture for years. Used to have it and knew it existed somewhere.

I believe he's still right today in 2024. Unless, of course, we define downwards the definition of 'intelligence' such that restricted-world problem solving is also 'intelligent' - similar to the internet scraping according to algorithms that passes today as AI. Internet AI is still rules and heuristic devices, except gathered very extensively. We still don't get 1) transcendent self-reflexive perspectives; 2) we don't get judgement - synthesis according to justified (normative) criteria.

The I from A.I. needs some definition. If Intelligence is considered the ability to perceive reality, to witness existence, consciousness etc; rather than an imitating mechanical function i.e. input output, then it does not matter how sophisticated the arrangement of functions, that does not indicate intelligence, true cognition. So what are we talking about? My friend had an answer phone that just said 'hiya how you doing?' I thought it was intelligent because for the input (calling him) I got an output 'hiya' that perfectly fooled me, this does not make his phone intelligent. Setting standards or tests to pass for an AI seems arbitrary, and perhaps more a test for our lack of intelligence rather than an airtight test of mechanical intelligence. What are we really talking about?

@bpatrickhoburg

Ай бұрын

I pretty much agree with you but “ai” is now more developed than simple input output. They perform more complex processes but nonetheless cannot feel in a real way, cannot sit in bed wanting a good philosophy lecture because it is storming and cannot think about a problem in a productive fashion (no pun intended). In other words, they cannot evolve themselves beyond practical problems. They can adapt in ways but cannot come close to outdoing humans. Life isn’t simply about rules and executing those rules.

@GugiMandini

28 күн бұрын

@@bpatrickhoburg Thank you for an interesting comment. It raises the idea of 'new' the 'gen' of genius true creative ability as an aspect of intelligence. Something on the point of input output, from a computer's point of view we have aggregations of functions with interfaces i.e. output = f(input). There is a branch of computational theory called functional dependencies and Armstrong's axioms that map these out, the axiom of transitivity shows that an aggregation of functions no matter how complicated is just a function. A physical library has not developed actual knowledge regardless of it size and sophistication of organisation.

@yoavco99

27 күн бұрын

How do you know computers can't be conscious?

@GugiMandini

17 күн бұрын

@@yoavco99 It's an interesting question. I would say a definition of unknown is that no inference can be made of it, the unknown is infinite in scope and possibility, like null in data systems, represented by the set of all possibilities. So although there is nothing to support that an aggregation of functions could derive intelligence, consciousness; I would also say I can not prove anything is conscious apart from me and I can not prove that anything is not.

Fantastic👏

He couldn’t foresee that AI would have the world’s knowledge to draw upon on the internet. It’s still a great lecture that points out the problems still facing AI, but ultimately he was shortsighted on the issue. Still super interesting: the realization of how much contextual knowledge is taken for granted.

There is a more recent video of him talking to an AI audience and repeating the same claims. The audience responded that AI actually does what he claimed it would never do. He seemed very surprised and asked to see it. Unfortunately, I couldn't find any follow-up videos (if there were any). Interestingly enough, Terry Winograd (who mentored Google founders) was taking Deryfus' course while at MIT. Who knows, maybe he started it all by pointing out that what Marvin Minsky was doing would be a dead end and new approach is needed.

@bpatrickhoburg

Ай бұрын

No offense for sure but I don’t know how much monitoring was achieved at Google. I’d refer you to Jaron Lanier, with all due respect!

@Danyel615

16 күн бұрын

Do you have a link to the video?

❤

Thank you… Can we say that A.I has problems understanding the way of thinking of those who are not grown up yet (kids) and Those who are well grown up (philosophical thinking)

it's so interesting, amusing and quite funny how wrong he seems to be considering where AI went.

@johnjackson917

Ай бұрын

he was right in a way you just have to understand when he says "AI" he means GOFAI - the idea of programming an AI with a set of consciously designed rules of course he's wrong in the sense that it's possible to create an AI by way of machine learning, which bypasses his objections entirely, and in actual fact turns out to work really well! but his insights with respect to formal programs i think are still true to this day. the scope of the claims needs to be narrowed is all.

@yoavco99

Ай бұрын

@@johnjackson917 Oh, I guessed he worked on a stricter definition or whatnot, that's why I said "seems". But to be honest, I simply don't know enough about this to be able to fully know what he meant in the context of where he said it. Also thanks for mentioning the term.

@daniellittlewood8471

Ай бұрын

Just to buttress johnjackson's comment, around the 33 minute mark he Dreyfus says "and so it looks to me like one ought to call into question the very idea of expertise that's in our philosophical tradition that makes us believe that these expert systems ought to work, and makes us think that only a fool would keep giving you examples, when really he should be giving you the rule by which he recognises those examples". You can't really level the same criticism against the statistical methods, because nobody really thinks they have a notion of expertise. It turns out that certain probabilistic techniques are pretty good at image recognition, and sentence construction, and that these tasks give you a surprisingly convincing human imitator. There are different philosophical objections about whether the techniques that work so well actually have anything to do with intelligence (i.e. do we learn something about how humans think by looking at how neural nets work). For instance, those techniques work very well (in fact significantly better) on junk data that has a pattern a human would never recognise. And the training data is substantially more than a human would ever have. But that's a different topic.

@user_user1337

Ай бұрын

we ought to be cautious about ANY philosopher claiming that such and such will never be possible because "I, the philosopher has thought so".

@dontbothertoreply9755

29 күн бұрын

Not at all, technology took his route anyone who has created, trained or understood AI sees that he was right.

I believe Dr. Dreyfus contradicts himself @ ~30:00 & @ ~43:00. At 30, I believe he rejects the idea that knowledge is 10,000 special cases. At 43, I believe he accepts the idea that knowledge is only so-many special cases.

@01dman

Ай бұрын

1985 Hubert thanks you

@SumNutOnU2b

Ай бұрын

Congratulations. You understood the point he was making.

@suzettedarrow8739

Ай бұрын

@@01dman XD tbh, tho, do you think i'm right? does he contradict hmself?

@lucasrandel8589

Ай бұрын

@@suzettedarrow8739 I think the interesting point here is that with the car example, the encountered situation is 'similar' to a previously encountered case. For a computer every deviation from a previous encounter would be a new case.

@suzettedarrow8739

Ай бұрын

@@lucasrandel8589 interesting. I’m pretty sure he contradicts himself. At ~49:00 to 51:00 he almost literally says the words “chess experts have only 10,000 special cases memorized. Euthyphro has 10,000 special cases memorized, too.” Doesn’t he? How do you interpret 49-51?