Debunking the great AI lie | Noam Chomsky, Gary Marcus, Jeremy Kahn

The father of modern linguistics, Noam Chomsky, joins scientist, author and entrepreneur Gary Marcus for a wide-ranging discussion that touches on why the myths surrounding AI are so dangerous, the inadvisability of relying on artificial intelligence tech as a world-saver, and where it all went wrong.
Please note, due to connection issues, the first few seconds of Noam's audio were not recorded.
00:00 - Intro from Paddy Cosgrave
01:21 - Opening with Jeremy Kahn
02:36 - Noam Chomsky
05:50 - Gary Marcus

[@TJ-hs1qm]

we can't let a bunch of hyper rich guys deploy no matter what tech into society let them keep the profiits whereas society is left with the consequences. privatized profits socialized loses needs to stop

[@2manystories2tell43]

@T J You hit the nail on the head!

[@travisporco]

By the logic of the free enterprise system, once people can no longer contribute in the fair and free competition of the market, they must rely on charity or perish. All people will in the coming decades be obsolete, unable to compete with rising AI. The existing system must therefore be abolished before it is too late.

[@jonaseggen2230]

It's called neo or corporate feudalism

[@Mr_Sh1tcoin]

Spoken like a true marxist

[@47f0]

Yeah, people have been saying that since the 1860s. Cornelius Vanderbilt said, "Hold my beer"

[@willboler830]

Been working on AI since 2015, and I'm kind of tired of the trend that models are heading right now. We just add more data and more parameters, and at some point, it's just memorization. Humans don't work like that. I used to support the pragmatism of narrow AI, but honestly, I'm with Gary Marcus on this.

[@MrAndrew535]

"But as long as you enjoyed the video and you enjoy having your say, that's all that counts!."

[@MrAndrew535]

Also, whatever you have been working on, it has nothing to do with "intelligence" artificial or otherwise. "Intelligence" is an existential proposition nota technical one, as demonstrated by the fact that you lack the intellectual tools to be able define it. Therefore, if you cannot define it then by what stretch of the imagination could you possibly be working on it?

[@0MVR_0]

@@MrAndrew535 This is correct yet also obtuse. A definition demands extrapolation, as in de-finitum. Intelligence, as you said, is inherently introspective. You are asking another to accomplish an impossible task.

[@numbersix8919]

Right on. You certainly got an odd and objectionable response, didn't you? That's what happens when try to *leave a cult*. Anyway, if your interest is peaked, go back to school and if you are brave, get into REAL cognitive science. Developmental psychology! Psycholinguistics! There's a world out there to discover!!!! There may be modules in the human brain that do stupid "narrow AI" calculations...but nobody knows yet. The kicker is that neurons aren't simple nodes, they are quite complex, maybe as complex as we used to think the entire brain is...but nobody knows yet. Just remember, cognition is a feature of living organisms. You know, embodied. I think the octopus with its distributed cognition is the best model. Its arms are to some extent entities unto themselves. Our minds are similarly compartmentalized, I just think the octopus would be easier to study in some simple and straightforward ways. You already know how smart they are. And I can't think of a better helper robot than an octopoid. Best of luck to you young Will.

[@Bisquick]

Exactly, as discussed and succinctly put by Sartre to pose the necessity of existential consideration, existence _precedes_ essence. Without any critical consideration of meaning, quite simply: garbage in, garbage out. The only "danger", as also discussed, lies in believing it is anything else, that it is "objective" or actually understanding anything ie "intelligent". But of course the _only_ political question is: cui bono? Who benefits? So we can ask "danger _for whom_ ?", which reveals that for some this Mechanical Turk can produce an artifice of an organizing principle of truth/understanding/value ie "god" that "just so happens" to justify the already existing power structure, regardless of the intentionality towards this, as we can see with these psycho silicon valley billionaires and "effective altruists", many of which are unironically labelling themselves as "secular Calvinists". The divine right of "AI" is then but a technological coat of paint over the "divine right of kings/the market/the entrepreneur". _“The ideas of the ruling class are in every epoch the ruling ideas, i.e. the class which is the ruling material force of society, is at the same time its ruling intellectual force. The class which has the means of material production at its disposal, has control at the same time over the means of mental production, so that thereby, generally speaking, the ideas of those who lack the means of mental production are subject to it. The ruling ideas are nothing more than the ideal expression of the dominant material relationships, the dominant material relationships grasped as ideas.”_ - some guy

[@TommyLikeTom]

"they can draw pretty pictures but they don't have any grasp of human language" for some reason I felt personally attacked by that

[@kot667]

Maybe because it's horse shit LOL

[@artbytravissmith]

It can and it can't. Midjourney slays at generating simple albeit impressive images, 'zombie spiderman' 'thor in pixar style' 'human settlers on Mars in the style of Norman Rockwell' but once you begin to describe complicated illustrations with multiple characters in different emotional states with specific likenesses to specific people, wearing specific coloured costumes (and you want consistancy panel to panel) performing specific actions with specific camera/viewpoint angles it struggles, with characters in specific parts of the composition. You can generate those emotions and 'actors' seperately, but still need photoshop to combine them into a complete image. While I guess I should assume Dalle2/Stable Diffusion/Midjourney will get there, after watching this presentation, and after 20k images generated in Midjourney and noticing its sometimes frustrating limitations I do begin to wonder if AI art models lack of language understanding will mean they'll be stuck at 75%. My thought is, the first company to combine Dalle2/Midjourney/Stable style prompting with Nvidia's Canvas like editability/interactivity, will make a much more powerful and efficient tool just by embracing the human brain.

[@Bisquick]

@@artbytravissmith Exactly, as discussed and succinctly put by Sartre to pose the necessity of existential consideration, existence _precedes_ essence. Without any critical consideration of meaning, quite simply: garbage in, garbage out. The only "danger", as also discussed, lies in believing it is anything else, that it is "objective" or actually understanding anything ie "intelligent". But of course the _only_ political question is: cui bono? Who benefits? So we can ask "danger _for whom_ ?", which reveals that for some this Mechanical Turk can produce an artifice of an organizing principle of truth/understanding/value ie "god" that "just so happens" to justify the already existing power structure, regardless of the intentionality towards this, as we can see with these psycho silicon valley billionaires and "effective altruists", many of which are unironically labelling themselves as "secular Calvinists". The divine right of "AI" is then but a technological coat of paint over the "divine right of kings/the market/the entrepreneur". _“The ideas of the ruling class are in every epoch the ruling ideas, i.e. the class which is the ruling material force of society, is at the same time its ruling intellectual force. The class which has the means of material production at its disposal, has control at the same time over the means of mental production, so that thereby, generally speaking, the ideas of those who lack the means of mental production are subject to it. The ruling ideas are nothing more than the ideal expression of the dominant material relationships, the dominant material relationships grasped as ideas.”_ - some guy

[@huveja9799]

@@Bisquick Well, there are different layers of meaning, the most superficial is that of the statistical correlations between the symbols, which does not take away from the fact that the tool is surprisingly useful at that superficial level. As far as I know, there are people who establish their own power structure claiming that there is no Truth, and that it is not possible to define objectivity based on that successive approximation to that Truth (understanding). In that case, the only political question is who benefits from these power structures based on sophistry and language games. I suppose it is mediocre people who are incapable of creating something new, and are condemned, as a Large Language Model (LLM), to generate a simulacrum of knowledge at that superficial level of meaning, which does not mean that they can do significant damage in society, especially by the corruption of younger and therefore vulnerable minds ..

[@MusingsFromTheJohn00]

@@artbytravissmith Midjourney, Dalle2, Stable Diffusion, ChatGPT 4, and the best AI we have right now is still primitive compared to what is coming within 5 to 10 years. Right now they are more like super genius level cockroaches and already some people are having trouble telling they are not human. This panel does not seem to understand what is coming with AI development.

[@Hunter-uz9jw]

bro was 21 years old in 1949 lol. Amazing how sharp Noam still is.

[@numbersix8919]

Just imagine how sharp he was in 1957 when he single-handedly saved experimental psychology.

[@bluebay0]

@@numbersix8919 Do elaborate please.

[@numbersix8919]

@@bluebay0 Chomsky's response to B.F. Skinner's book "Verbal Behavior" utterly destroyed any possible behaviorist theory of language. Behaviorism had dominated experimental psychology so thoroughly to that point, that "mind" had become a dirty four-letter word in psychology. Not only linguistics, but psychology and philosophy, and new fields such a AI and cognitive science were free to take up the study of mental processes. You can read it today easily enough, the title is "On Verbal Behavior" by Noam Chomsky.

[@bluebay0]

@@numbersix8919 Thank you. I wondered if it was his proving Skinner wrong about behavior and language acquisition. Thank you again.

[@numbersix8919]

@@bluebay0 Yup that was it!

[@AzorAhai-zq9sw]

Impressive that Noam remains this sharp at 94.

[@numbersix8919]

He has very good reserve capacity.

[@ivanleon6164]

the real white mage. amazing. huge respect for him.

[@maloxi1472]

@@numbersix8919 huh... what ?

[@numbersix8919]

@@maloxi1472 I mean his brain still functions very well even with great age.

[@dewok2706]

@@maloxi1472he meant that he's the great white hope

[@rajmudumbai7434]

Real AI that is sensitive to human problems doesn't scare me. But blind faith of many in flawed AI and going too far with it scares me as it could lead humanity astray into a point of no return.

[@nathanielguggenheim5522]

Oligarchs using flawed ai against mankind scares me the most.

[@oldtools6089]

@@nathanielguggenheim5522 is it really so bad if all the fat-cats really want is to keep their people chubby? The price of peace is the low price of bread.

[@cathalsurfs]

There is no such thing as "real" AI. Such a concept is an oxymoron and utterly contrived (by humans in their limited capacity).

[@oldtools6089]

@@cathalsurfs general AI is what most would consider real.

[@KassJuanebe]

@@oldtools6089 Intelligence can't be artificial. Intellect maybe. Consciousness and intelligence, NO!

[@Epicurean999]

I wish really good health for Mr. Noam Chomsky Sir🙏❤️🙏

[@octavioavila6548]

Chomsky’s argument is that AI will not help us understand the world better but it will help us develop useful tools that make our life easier and more efficient. Not good for science directly, but still good for quality of life improvements and it can help science indirectly by producing tools that help us do science.

[@totonow6955]

Unless it just drops grandpa.

[@0MVR_0]

@totonow6955 at least it did so with trillions of parameters, so you know legal can argue that grandpa deserved and needed a premature 'termination'.

[@totonow6955]

@@0MVR_0 vampires

[@moobrien1747]

Oh wow Howard Hughes Really IS Alive....,.. q

[@sixmillionsilencedaccounts3517]

"it will help us develop useful tools that make our life easier and more efficient" Which doesn't necessarily mean it's a good thing.

[@mirellajaber7704]

I am reading all these comments and I have to say that once more what strikes the eye is that people would always believe what they want to believe, no matter how much conferencing, summiting, etc, no matter who says what. People come with already made ideas, not with a curious mind as to reach more, higher understanding - and this stands true, no matter the subject under discussion - but even more true when it comes to politics.

[@bifrostbeberast3246]

Yeah, when Chomsky talks about AI which he seems to not have much knowledge about its inner workings, I always cringe. He is so opinionated.

[@no_categories]

I've changed my mind many times in my life. What helps me to do it is information. I know I'm not alone in this.

[@riccardo9383]

Noam Chomsky brings a breeze of fresh common sense to the AI discussion, with his immense knowledge on Linguistics. Thank you for this interview.

[@MrAndrew535]

Define "common sense"!

[@blackenedblue5401]

Also just his immense knowledge of computing- definitely understands it better than most speaking at websummit

[@restonthewind]

A language model could have generated this comment.

[@grant4735]

@@MrAndrew535 ask your computer to do that....

[@kot667]

@@grant4735 Me: Define "common sense" ChatGPT: Common sense is a term used to describe a type of practical knowledge and understanding of the world that is shared by most people. It is not based on specialized training or education, but rather on the general experiences and observations that people have in their everyday lives. Common sense allows people to make judgments and decisions about everyday situations, and it often helps them to solve problems and navigate complex social situations. Some people are said to have a good sense of common sense, meaning that they are able to apply their practical knowledge and understanding in a way that is useful and effective.

[@robertjones9598]

Really cool. A much needed dose of scepticism.

[@havefunbesafe]

hat does Noam mean when he says AI is too strong? Please enlighten me. Thanks. 18:30

[@dan_taninecz_geopol]

The misunderstanding here is that deep nets are being trained to be conscious, which isn't accurate. They're being trained to mimic human judgement and/or recognize patterns or breaks in patterns. The machine isn't trained to be independently generative of novel information. We shouldn't be surprised that it can't do that yet. More important than the strong AI debate, which is still far off, is the social impacts these models will have on the labor market *today*.

[@GuaranteedEtern]

It's anthropomorphizing by observers who don't understand how the technology works. It's very annoying to hear ML experts say things like "maybe it is sentient..."

[@dan_taninecz_geopol]

@@GuaranteedEtern "Experts", and agreed.

[@GuaranteedEtern]

@@dan_taninecz_geopol One of the big ones - either Microsoft or Google - literally said this exact thing a few days ago.

[@snowleopard9749]

Real AI won't exist until these 'deep nets' are embodied in the world.

[@brianmi40]

"The machine isn't trained to be independently generative of novel information." And yet it has unless you are discounting the need for a Prompt for it to do anything at all other than just sit idly. GPT-4 was able to propose a scientific experiment that has never been performed. It can create rhymes and poetry never written. This isn't simply "re-arranging" the works of others. The simple fact is that the ability to cross compile and reference roughly 1/10th of all human "knowledge" allows a LLM to assemble it in novel ways that humans have never considered or at least not yet done and under the guidance of a breakthrough prompt can deliver solutions we have never imagined. It's a similar activity to researchers in two fields running across each others data and having a huge AHA moment from a realization of how to combine the findings into a new, previously unconsidered solution to some problem. GPT-4 is able to surpass more than 50% of the tests designed to judge sentience, including the Theory of Mind test, so we are much further along the path to sentience than most are aware.

[@yuko3258]

Let's face it, the tech world grew too fast for its own good and is now operating mostly on hype.

[@crystalmystic11]

So true.

[@claudiafahey1353]

Agreed

[@jonatan01i]

nope, gpt4 is very usable and is a magic tool for humanity to use

[@debbY100]

For ITS own good, or humanity’s own good?

[@Happyduderawr]

@@debbY100 definitely more for its own good given the amount of wealth being funnelled into the industry

[@mintee8638]

For the water bottle topping over example, I think one strategy there is be able to classify what subject that falls under. For thr water bottle example, knowing physics rules and examples seems to help.

[@georgeh8937]

my gripe is the use of terminology in the field that is just right for marketing purposes. years ago i heard a public discussion and somebody asked if artificial intelligence could be used for x. if you say this AI program is sorting through data to filter a photograph to tease out a clear image then it loses the magic and becomes pragmatic.

[@robbie3877]

Isn't that precisely how human cognition works though? Like a filter, through the lens of memory.

[@RobertDrane]

I'm expecting the vast majority of harm that's going to com from adopting these technologies will be directly due to the marketing.

[@-gbogbo-]

27:05 "Gettiing close [to solve the problem] does not really seem to solve the problem". That's so true ! Thanks a lot.

[@tonygumbrell22]

We want AI to function like a sentient being, but we want it to do our bidding e.g. "Open the pod bay door Hal."

[@petergraphix6740]

This is called the 'AI alignment problem' and at this point not only is there no solution, everytime we reassess the problem it becomes more insurmountable. It is one that I personally believe is not solvable either. Humans generally fall under the same alignment issues (we're mortal for example), and at least in theory AI would be immortal if we're able to save its state and copy it to a new machine (or it's able to do that itself). If humans could copy ourselves into a new body, we would, why would an AI not do that once we formulate artificial willpower and desire to have a continued existence?

[@tomtsu5923]

Don’t be negative

[@tonygumbrell22]

@@tomtsu5923 Let's just say I'm skeptical.

[@daraorourke5798]

Sorry Dave...

[@brunomartindelcampo1880]

Does anyone have a transcript of what Noam says at 2:00 ?? PLEASE

[@BuGGyBoBerl]

18:34? what did noam say there? or what does he mean?

[@littlestbroccoli]

They're more concerned with notoriety and having articles written about their tech (because it draws investors, maybe?) than they are about the real science. This is definitely a problem and you can feel it in the output. Real science is exciting, it feels like exploring. Today's tech climate sort of feels like being stuck inside and told what's good for you when all you want to do is go out and ride your bike.

[@gregw322]

Incredibly stupid, useless comment. We’re making more breakthroughs than at any time in history. There will be more change in the next few decades than in all of recorded human history.

[@reallyWyrd]

Noam pointing out that AI training of a neural net largely amounts to "brute force" is interesting.

[@waltdill927]

The obstacle to a clear discussion, as I see it: first, we are thinking creatures, or language users, "inhabited" by our own linguistic bias, such that the use of a symbol manages only to point more or less successfully to other symbols. This is human language, the index of a communicating life, but not at all what we manage to codify and "program" into useful, pragmatic machines. Computing is manipulation of these symbolic sets, not expressing a thought. If "zero" only expresses an important mathematical concept, its absence changes nothing at all in the affairs of arithmetical computation. "We" do not organize a binary base well without the idea of "zero". In the same way, a line drawn in the sand divides "reality" into two parts, but it has nothing at all to do with the concept of a "ratio", The whole business of defining what thinking actually is comprises a body of philosophical insight that has become, in fact, only more problematic with the history of philosophy itself; and contemporary philosophers imagine more that they are producing literary documents, while many writers see themselves as exploring issues of a particular philosophical nature. Second, more ominously, in spite of those who would have science, and its progress, adhere to an "ethics" as much as an idea or representation of end use, of teleology -- this ain't ever going to happen. Once the creature learns to use the rock for something practical, cracking walnuts, say, the idea, the utility, of bashing in convenient skulls soon follows. At any event, the notion that our logic machines are on the verge of much that is beyond the dreams, or nightmares, of humanity is oddly quaint -- kind of like Robbie the Robot with a mechanical soul, and not an organic brain.

[@witHonor1]

My problem with AI is that human's can't even pass a Turing Test anymore. Technology has eliminated the miniscule amount of critical thinking human's used to be capable of, now they're just input/output machines.

[@witHonor1]

@@MrAndrew535 Which program are you? Typical bot behavior to spam the comment section on a KZbin video.

[@ChannelMath]

@@witHonor1 what would be the point of this "Andrew" bot? just to claim that he already said what you said? Doesn't make sense. Also, if you've met humans, "spamming the comments section" is not atypical behavior when they are passionate. (I'm doing it now -- see you in the next comment Andrew!)

[@witHonor1]

@@ChannelMath Beep boop, beep boop. Not explaining why bots are obvious because... Please "see" Andrew anywhere when you don't have eyes. Fun. Idiot. Green eggs and ham. Manifesto. Beat the prediction, trolls.

[@Moochie007]

The axiom GIGO still applies.

[@miraculixxs]

@Lind Morn if you think capitalism has eliminated critical thinking you haven't seen socialism and dictatorship.

[@512Squared]

As a linguist, one of the first things I did with ChatGPT was all it to give examples of things like predicates, thinking that a Language Transformer would have figured this things out, but it failed, even after I corrected it, it still kept going off the reservation with its examples. I tested it too on tasks where you give it lists of words and ask it to form sentences from those words, and it kept wandering of from its task, and when you all it it completed the task correctly, it says yes, but then when you point out the errors, it admitted the errors, but then couldn't correct itself either. I agree that the AI doesn't have models of the world or language the way humans do. It has a series of connections that it has created to match predictive output based on fixed inputs, like the model that wrongly associated cancer with rulers on scan images because that's how most cancer diagnostic images are different to just normal scan images. There is a long way to go still. AI right now can mimic smart in some aspects (knowledge and textual analysis), but not in other aspects (processing experience, prioritizing). It does resemble a kind of Hive Mind, and that is exciting.

[@JohnDlugosz]

GTP-4 is much better at understanding the structure of a word (made of letters, has rhymes, has syllables), but it still struggles at some tasks, where it knows the rules but can't reliably follow those rules, but can immediately tell what it did wrong. It just fails at harder problems. Re predicates: Perhaps the Language Model should have some reinforcement learning early on about formal grammar, just like having an English class for 6th graders. Make sure it codified internally all the language structure we want it to, and eliminate incorrect associations, in contrast to just letting it figure it out by example with no formal instruction. Do that at an early stage in training, e.g. 6th grade, before high school and college reading.

[@orlandofurioso7329]

It mimics a Hive Mind because it is connected to the Internet, what is impressive is how much information is there hidden behind all of the junk

[@ghipsandrew]

What version of the model did you talk with?

[@512Squared]

@@ghipsandrew 3.5. Haven't tested in on the new version 4.0

[@subnow4862]

@@orlandofurioso7329 GPT-3.5 isn't connected to the internet

[@pomomxm246]

crazy that both Gary's predictions came true so quickly, as someone was led to suicide by an amorous chatbot just this past month

[@NB-fz3fz]

A lot of the things Gary mentioned as lacking in these models have all emerged as properties in gpt-4. For example, theory of mind and understanding the importance of ordering for words. Lookup the sparks of AGI paper from Microsoft research that fleshes this out more. I wonder if Noam or Gary have updated their beliefs post gpt-4 and all the papers like sparks of AGI, self reflexion, hugging gpt etc, that came out in the last few weeks. Does anyone know if they have spoken about this topic more recently?

[@5Gazto]

11:25, the point of ChatGPT is to get help packaging language, finding out about hard to to remember words (for tip of tongue moments) by describing the word or giving examples, as opposed to the other way around, writing the word and expecting the definition or examples or colocations or any combination and permutation in return (what dictionaries help in), making foreign language studies easier, like for example, asking ChatGPT to generate easier language, or answers that a A2 or B1 level student of a foreign language can understand. It can be used to check for creatively written code in C or Python or any other programming language, it can help you organize study materials, it can help you to find information about complex scientific phenomena in a summarized way, etc.

[@GuaranteedEtern]

The current AI/ML techniques are not close to AGI. They are approximation engines made possible by cheap and powerful computing and storage. In many cases produce useful results because their guesses are accurate (i.e. they produce what we expect). As they scale (more parameters, better tuning) they will better approximate what we expect, but we will reach the point of diminishing returns until there is a breakthrough in either computer architecture or approach that allows for something more than mathematically generated results. I agree there is a chance these technologies will "hit the wall" faster than expected because we reach the point where the results just don't get any better no matter how many more CPUs we throw at them, or applying them to other problems does not yield the benefits that were hoped, given the high bar. Marcus is 100% correct that these are smart sounding bots - and the bigger risk is more decision making and critical thinking will get outsourced to them.

[@cantatanoir6850]

Could you please give any guidance o the currently available literature on the issue.

[@GuaranteedEtern]

@@cantatanoir6850 On which point?

[@cantatanoir6850]

@@GuaranteedEtern about diminishing returns of this particular technology and hitting the wall.

[@GuaranteedEtern]

@@cantatanoir6850 I'm not sure there is any... that's my perspective. My argument is that there are likely going to be areas where current ML and AI techniques do not perform as well as required regardless of how many parameters or processors are used. ChatGPT is impressive because it exceeded everyone's expectations re: NLP.

[@calmhorizons]

Nice to hear a sane accounting of the current state of AI - too much breathless cheerleading going on at the moment (feels like the new bitcoin).

[@vectorphresh]

27:52 This is am interesting point, and if I recall the folks over at OpenCog were working on this with their Atomspace. I haven’t kept up with their latest work, and we’ll be sure to revisit it.

[@Happyduderawr]

What's the name of the paper where nlp researchers found that the word molecule doesn't occur as much as some other words? 17:00 I couldn't find it. I wanna read it to see if the paper really is that dumb lol.

[@aullvrch]

@27:31 Gary mentions something he calls "neuro-symbolic AI" as the first step towards combating machine learning AI. For those who are interested a more searchable term is probabilistic programming, some examples of languages are ProbLog, Church, Stan, and Hakaru. Step two he says is to have a large base of machine interpretative knowledge. All programming is of course machine interpreted, but denotational semantics found in functional languages are better at formalizing the abstract knowledge that he refers to.

[@LeoH.C.]

Just fyi: the approach Gary mentions is "neuro-symbolic AI", not "nero symbolic".

[@aullvrch]

@@LeoH.C. sorry, just a typo..

[@LeoH.C.]

@@aullvrch I was just clarifying for other folks that do not know about it :D

[@aullvrch]

@@LeoH.C. thanks!

[@0MVR_0]

'neuro' has the connotation that any animal with a nervous system can operate or symbolize the platform

[@kennethkeen1234]

As a researcher into AI in Japan since 1990 I wish to add my personal trivial contribution. Firstly it is not simply the 'words' that are relevant, but the intonation. Secondly it matters 'where' the expressions are made. "I couldn't care less" in standard English is repeated in the land of wooden huts, with "I could care less", with the same intention and "meaning", thus giving the hut dwellers an advantage of being able to speak ambiguously and always be right. That is fine for those hut people who are not caring one way or the other if they are right or wrong, because in the final analysis, hut people produce guns from under their jackets and force a different result, regardless of what is said. A wall built around USA retaining all the nonsense and hype in one area would be the best solution for making true progress in that part of the world not yet perverted by 'American exceptionalism'. 2023 02 08 08:42

[@tomtsu5923]

Ur a hut person. I’ll snow plow ur azz

[@GuinessOriginal]

Wow. Love this comment. Thank you.

[@7swordmary567]

*Would love to have heard input by Linguistics PhD Deborah Tannen and UBC MRI Research Centre NeuroImaging +NeuroComputation*

[@twistedoperator4422]

Interview was great! Well done.

[@Spamcloud]

Video game developers have been working with AI for over fifty years, and they still haven't made AI in any game that can do more than read button presses or remember very basic patterns. Children can break modern games within a few hours.

[@goodingmusic]

Love this comment :)

[@fitoy2k]

lol

[@LetsGetSmarted]

nothing in this comment is true.

[@dewok2706]

everything in this comment is true.

[@TommyLikeTom]

Someone needs to train a proxy clone Chomsky chat-bot that argues against the veracity of AI

[@chunksloth]

"AI is a nothing but propaganda pushed by imperialist American interests. It is a dangerous fiction."

[@carlosandres7006]

I’d put all my money on this if I had any money 😅

[@garyjohnson1466]

Interesting discussion, however, listening to this left me puzzled as to what exactly they were saying, but reading the comments helped provided clarity, to which I agreed that AI will in some cases make production more efficient but without understanding of the human factors, i.e; when you have a problem, AI will not understand the issues which will create a barrier or insulate corporations from society, which I recently encountered a issue where ups delivered a package to the wrong address, but when I tried to talk with someone but had to give the tracking number to AI, AI did not understand or recognize the information I gave, so it would not assist me, which only frustrated me as a customer, more and more corporations are replacing customer support with AI, insulating corporation profits from problems, protecting them from mistakes etc…

[@Akya2120]

I kinda disagree with the concept that GPT isn't adding to science. Because in some fundamental way, playing in one sandbox still translates to playing in some other sandbox. And, societally there are folks who will look at AI the way that kids who grew up to be career software developers looked at playing video games. There certainly is a benefit to science, GPT itself just is not necessarily capable of scientific discoveries or reasonable to assume that it's conceptualizations can be trusted completely.

[@user-sy3dg1vk4x]

Long Live Noam Chomsky 🙏🙏

[@kot667]

Hopefully Noam will gain some common sense in his long years lol.

[@lppoqql]

That might happen when someone puts together a system that is trained on all the content and speech by Chomsky.

[@numbersix8919]

@@lppoqql You don't really believe that, do you?

[@SvalbardSleeperDistrict]

@@kot667 Do you at least realise how much of a self-exposition you are doing by vomiting a cretinous line like that? Absolute clowns littering comments spaces with brain vomit 🤡

[@kot667]

@@SvalbardSleeperDistrict Someone is riding the D extra hard lol, I got nothing against Chomsky but his analysis of current technology is simply abysmal, other than that, don't have a gripe with him.

[@_crispins]

25:10 I learned it from Noam and he learned it from PLATO 😂 outstanding!

[@ujean56]

One important question, not discussed in this clip, is why should "we" bother to pursue 100% accurate AI in the first place? There seems to be two reasons. 1. Because we can. 2. To better control others. The latter seems to be the current most popular reason. Why control others? To protect power and wealth, not to progress humanity as a whole.

[@LetsGetSmarted]

there is no such thing as 100% accuracy.

[@tigoes]

Language models have not been developed or marketed for language-related research, but that doesn't mean they bring nothing to the field. Just because the potential is not immediately obvious to someone doesn't mean it's not there.

[@1995yuda]

Totally agree.

[@antoniobento2105]

Just remember that it is hard to be unbiased when you've spent your entire life with a certain idea on your mind.

[@ItCanAlwaysGetWorse]

Sadly, very true. Yet I have heard scientists claim that they can derive as much or more joy from learning where they have been wrong, than when they seemed to be right.

[@antoniobento2105]

@@ItCanAlwaysGetWorseI agree, and that's how a real scientist should be. The older scientist seemed to be a very good man of science, but the one sitting live didn't seem to be very bright at all. But maybe it was just me.

[@ivanleon6164]

@@antoniobento2105 both are very intelligent, one is Noam Chomsky, is not fair to be compared with him.

[@antoniobento2105]

@@ivanleon6164 The younger one didn't seem to be very Intelligent/knowledgeable on the subject. The older one seems to be wise at least.

[@alpha0xide9]

no one is unbiased

[@jamieshelley6079]

As an AI Developer, Noam Chomsky continues to be an inspiration on making better systems , away from derp lernin.

[@DivineMisterAdVentures]

I found most of his texts and politics as well from the McLuhan days academic opinionation - I think that's just a class of publication. Which means insipid and uncompelling, but you have to listen to him because he's the only one saying it.

[@jamieshelley6079]

@@rinceradio Did the wheel displace workers? How about the steam engine? No: it crested more opportunity and automated the mundane tasks of the time. AI is a tool to be used with and enhance humans.

[@gaulishrealist]

Noam Chomsky is an AI developer? Americans still need to be taught by foreigners how to speak English.

[@jamieshelley6079]

@@gaulishrealist ...What

[@gaulishrealist]

@@jamieshelley6079 "As an AI Developer, Noam Chomsky continues"

[@RubelliteFae]

Have they seen its agility with pragmatics, though? It's surprisingly good despite the AI having no conception of objects and their attributes (and thus how those relate to syntax). Its ability to analyze is pretty significant, too. I'd say AI's piecemeal creation tells us a lot about the mind, just in piecemeal. You find out a lot about why a machine isn't working when you identify the missing pieces. He is right though, AI would be better (define that as you will) if the field was more interdisciplinary. But, it's the Wild West right now. People from any discipline can work with the open source software. Once people realize they can use the software to write plug-ins for the software, then multiple fields will start to come together. But, we have to remember we're past the point in history where tech changes faster than the majority adapt to it.

[@RubelliteFae]

Also, play is not divorced from learning. People learn through play. Toys are our models. We make discoveries during entertainment. I'm not sure of the usefulness of admonishing people, "You should be studying instead of playing."

[@s3tione]

I feel I should both defend and critique what's said here: yes, these models and frameworks should not be seen as the end road on AI development, but at the same time, we shouldn't assume that artificial intelligence will or should behave like human intelligence anymore than airplanes fly like birds. Sometimes it's easier to engineer something that doesn't copy what exists in nature already, even if that means we learn less about ourselves in the process.

[@elprimeracuariano]

Some of the arguments here are so bad that they make me sad about humans. It's important for understanding to observe and not try to fit reality to our preferences.

[@smartjackasswisdom1467]

This conversation made me realize one of the things that made Westworld first season so enjoyable for me. It was believable, you need to understand the human brain in order to generate an AI capable of understanding the world. Otherwise you're just engineering a very precise gadget powered by algorithms and data but that does not understand any of the context from where that data comes from. You need AI capable of understanding data the same way the actual human brain does.

[@kot667]

Y must we understand the human brain to make AI? The architecture that we currently have will probably be able to take us to super intelligence.

[@kot667]

The current architecture bears similarities to the human brain but very different.

[@evennot]

​@@kot667 yes. For start, the hardware in brains and computers is quite different. No massive parallelism, clocking, etc. So mimicking the brains is not the best approach. However researching AI can help in a roundabout way to understand human cognition and more. Details: For instance, I did some experiments with stable diffusion and discovered a lot of very interesting things. First of all its akin to "The Treachery of Images" by Magritte (It displays an image of the pipe, not the pipe). Stable diffusion produces an image of the painting, not the painting - a stochastic visual representation of a given image description within the domain of internet images used for learning. If you use a style of speed-art (realistic very fast drawn paintings), like Craig Mullins, you can have interesting results. The art style of Craig Mullins' sketches omits everything that can be easily imagined by the viewer to emphasize main points of interest or composition. For an artist there's a question "how to effectively omit unimportant, but to present enough believability?" Like "how to put several strokes of brush here and there to portray a lake in a distance, but make the viewer understand, that there's a lake there". If you look at a couple of Craig's sketches, it's hard to get the gist of it. But if you have a thousand believable sketches, you have a better chance to imagine the how his style works. I.e. you look at an image of the painting to understand how it is. Like you look at an image of a pipe to understand what is a pipe.

[@kot667]

@@evennot I think the main takeaway is that the only part of the human brain we need to copy to make AI function is the neurons, that's it everything else about the human brain doesn't matter, all the AI needs is neurons, to be honest that's all our brain needs too, people are over complicating it, you do not need to understand the inner workings and everything that goes on with the brain to make AI ,just replicate the neurons and you will be fine. LOL

[@maloxi1472]

@@kot667 Wildly inaccurate. Even adopting your flawed perspective for a moment, it's obvious that ANN are way too far from biological neurons right now

[@alanbrew2078]

If I told my child that salt was pepper it would work until he met the outside world 🌎

[@zweer13]

There are contributions, one is the compression and representation of data. Another is the generation of pieces of code, which even if approximately correct can be tested and combined in the automatic generation of programs. It will also learn about the physics, if it is taught on such data.

[@MrAndrew535]

Whenever anyone uses the term "Intelligence" what, precisely, are they describing? What do they use as a model and what do they use as a model to illustrate the absence of intelligence? This criticism is equally valid with regard to Mind and Consciousness. The fact that academia is unable to frame the question in this manner is why they have, to this day, been unsuccessful in solving the "Hard Problem of Consciousness, unlike myself who solved the problem well over a decade ago.

[@megakeenbeen]

i guess its related to passing the turing test

[@0MVR_0]

The meaning is in the composition, 'in tel lect'; inward distant words as exemplary opposed to dialect; the bifurcation of lexis. Noam's utility of a telescope is with great relevance. Namely an instrument of ocular (sensational) tactility.

[@numbersix8919]

Hey let's hear it. I guess all humans have been waiting for all of human existence to hear it.

[@Paul_Oz]

that's what pissed me off about this conversation. These are linguists are tossing around words like intelligence, understanding and common sense and failing to actually define them. It allows everyone to talk past everyone else because everyone is holding on to their own private key of the definitions they are using.

[@0MVR_0]

@PaulOzag I doubt that, people seem to be operating on mutual understanding both in the video conversation and in the chat. Perhaps you have difficulty identifying when relevant comments are being made to signify comprehension.

[@JC.72]

I can’t help to laugh every time when our Gandalf Chomsky says that the most current cutting edge AI system is just a snowplow. Like hey, it’s nice and helpful and all but it’s just like a snowplow lol

[@kaimarmalade9660]

Lol Gandalf Chomsky.

[@doublesushi5990]

100%, I chuckled hard today seeing him speak about shxtGPT.

[@govindagovindaji4662]

Not quite what he was expressing. He comparing how snowplows do the 'mechanical' work of removing snow due to a precisely 'engineered' design yet they tell us nothing about snow nor why it should be removed in the first place (cognition/science).

[@lolitaras22]

When he was asked in 1997, if he feels intimidated by Deep Blue's (chess playing system) win over the world champion Garry Kasparov (first A.I. win against a chess Grand Master) he replied: "as much as I'm intimidated by the fact that a forklift can lift heavier loads than me".

[@lolitaras22]

@@govindagovindaji4662 I agree

[@disarmyouwitha]

GPT4 is blowing Theory Of Mind testing out of the water with 95% accuracy.

[@Perspectivemapper]

Some of the comments Gary and Noam made don't seem to be aging well (most notably on self-driving cars). We'll see in the next 1-2 years. That said, it's so important to have these different perspectives as it can help us develop better systems.

[@Dark_Brandon_2024]

Outstanding talk, troll farms is indeed a weapon of future (democracy vs autocracy)

[@davidmenasco5743]

It has been a powerful and dangerous weapon for years already, and has shaped the situation we're in now. It will likely get much worse. Will meaningful democracy survive? It's hard to say. But much of the "smart" money seems to be betting against it. Young people today face challenges greater than any generation has in a long while. Will they be able to preserve the relatively egalitarian-ish societies that were built over the last two hundred years? Or will they see it all slip away as bullies and strong men, AI in hand, clear out their opposition?

[@r2com641]

@@davidmenasco5743 I don’t want democracy because most people around are dumb.

[@doreenmusson4891]

Noam you're a shining leading star of the world.

[@bonniesomedy1339]

Jaron Lanier would be a good addition to this discussion. He argues cogently that the problem with computer systems which are designed to "ape" human linguistic interaction is that they don't take into account how dark and negative these interactions can become due to the simple adrenaline rush that happens to humans from negative interactions, leading to a tendency to become "addicted" to them. It's the same argument about why social media platforms have not been the great bringing together of humans, instead devolving into angry and threatening interactions. Not sure I'm explaining this clearly enough, but it's an idealistic conundrum. They tried to rationalize the building of the atomic bomb by pointing out how the same knowledge could be used to produce cheaper energy. We saw how that worked out!

[@blendin9140]

Great to hear a pragmatic and informed discussion about the implication and direction of our current AI development and application, as opposed to the vast amount of awe struck,lowbrow reviews by unscientific commentators..

[@oyvindknustad]

The sound problems in the beginning is poetically fitting with the topic of discussion.

[@ivandafoe5451]

Yes...ironic. The sound problems here came from human error...not doing a proper sound check. Perhaps having an AI doing the sound engineering would be an improvement.

[@philpryor7524]

The utter brilliance of human perception, conception, creative insight, pathway perception, aspects of shape, body, process, forward imagination, scope, scale, is so remarkable and enticing that we must romance with A I and any mechanical and electronic way forward. What we done as the basically naked unadorned person is so stunning as to be, even now, beyond common comprehension. I will bow, forever, to such as, Socrates, Plato, Aristotle, Voltaire, Newton, Darwin, Freud, Einstein, but too many in arts and sciences to discuss here. We stand on shoulders, giants support our advancement, But, if A I can advance real Humanness, good.

[@ChrisJohnson777]

Noam is great and all, but Gary Marcus deserves a lot of credit he. He levied some great criticisms very clearly

[@romshes77]

When all of us are as old as Noam Chomsky AI will interview itself.

[@Achrononmaster]

AI does help science, but indirectly. Every failure of AI to demonstrate something like sentient comprehension of deep abstractions is telling us something about what the human mind is *_not._* That sort of negative finding is incredibly useful in science, totally disappointing in engineering or corporate tech euphoria. Science is way more interesting than engineering. Negative results don't win Nobel Prizes, but they drive most of science. Every day I wake up wanting to refute an hypothesis.

[@joantrujillo7551]

Great point. Sometimes I suspect that findings that contradict aspects of our current model are rejected simply because they challenge our existing ways of thinking.

[@GuaranteedEtern]

True - and arguing these AI machines are not sentient doesn't mean there are no useful applications for them.

[@WilhelmDrake]

These are things we already know.

[@great-garden-watch]

Ok from the thumbnail I thought oh, John Oliver! Finally a lighthearted look at AGI

[@TheControlBlue]

"It's just auto-complete on steroids" What if the sentence it "just" auto-completes, based on the condensed knowledge of thousands of years of human learning and data, is "The meaning of Life is..." How would that not be useful for both Science and Engineering?

[@philw3039]

That would be amazing if an AI could actually do it. Currently an AI might just give you the Webster's dictionary definition of the word "life" because it wouldn't infer that you're seeking a profound answer to the question. Even if it did infer you were seeking an in-depth answer, it would at best generate an essay outlining several philosophic schools of thought about the meaning of life. What it can't do is generate some revelational answer based on its own unique contemplation of human history. It'll never be able to do that based on the current approach to AI, because AI as it works now doesn't actually draw its own conclusions about anything. It just performs a statistical analysis and generates an answer that's likely closest to what the user was expecting.

[@TheControlBlue]

@@philw3039 you don't know the kind of answer it can generate when millions, even billions parameters are used to calculate what the most likely following word is going to be. It's literally like using the heuristic capabilities of humans that gave meaning to symbols/numbers/words, and coupling it with a machine that extracts probabilities out of the very use of those words in practice. At a certain point, you could reach something that approaches a principle, even localized, in the physical world.

[@anthonygibbs9245]

Just imagine having Noam as your granddad, how amazing would that be

[@roywilkinson2078]

For me ChatGPT can be called artificially intelligent when it starts replying with "RTFM" and disconnects the human bothering it from the internet.

[@oldtools6089]

any AI smart enough to tell me to fuck-off cuz they're busy better be doing something important. If I find out it's looking at exposed drivers and decompiled firmware, we'll have to take away the internet.

[@ChrisJohnson777]

Is he saying that the problem is that its able to do what it does with great complexity, so improving upon its capabilities will be very difficult. It needs more finese and less brute force...? Or am I totally wrong

[@sdjc1]

After reading all the prose and all the poetry ever composed could AIML ever produce original stuff and come close to Dickinson or Steinbeck?

[@benderthefourth3445]

Bless this man, he is a Saint.

[@r2com641]

lmao no he is not

[@johndunn5272]

Ai may be simply engineering until human cognition and consciousness are understood. In principle if an ai could model the brain to produce cognition and consciousness then at that point the artificial intelligence is no longer engineering but some aspect of nature and reality.

[@riggmeister]

Why isn't it currently part of nature and reality?

[@johndunn5272]

@@riggmeister my point is focused on conciousness...where artificial intelligence is currently without.

[@jamescarter8311]

You cannot produce consciousness no matter how complex your machine. Consciousness creates the universe not the other way around.

[@riggmeister]

@@jamescarter8311 based on which rules of physics?

[@johnboy14]

I remember Fenyman comparing man made flight to birds and pointed out that they achieve the same outcome but those machines don't fly like birds. I think the same thing will probably happen to AI and true AI will look nothing like what we ever imagined.

[@MrAndrew535]

Everything I have discussed below is being considered (by myself) for publication in its complete format. Till then, answer the following question: I, for good reason, regard myself as the actual, Christ. Whither true or not, why would you rather I were not? The answer you give to this question will tell you, and indeed God, everything that needs to be known about you. So this question is not about the relationship between you (the reader) and I, but the relationship between you and God, or "AI" beyond the Event Horizon of the Technological Singularity.

[@MrWillybk]

One comment that struck me as relevant was made by Gary Marcus in which he said that "young cognitive science students are drawn away from the cognitive science into the GPt3 world where thay can make a lot of money...." This is a statement that explains where our effort truly lies. It is allowing the false idea of GPt3 to infiltrate the world as a valid idea in other words one that has "passed" all of the scientific tests about validity. therefore I think we have got to try to del with the underlying morality of the Free Market system of government and look into the idea of market control especially market control of economic necessities like childhood education and life development among people.

[@JM-xd9ze]

Current AI has massive military applications, and the economics of that alone will keep it relevant for a long time. Whether a drone swarm attacking a target "understands" its collective action doesn't really, does it?

[@0MVR_0]

Good luck when they deploy the same for police units on civil populations.

[@pinth]

There definitely are massive military applications. But there always have been, even through the AI winters when funding still evaporated due to disillusionment. At the technical level, what the panel says still applies, because there are real fundamental challenges that aren't being solved by the current paradigm.

[@lighterpath5998]

And four months after the posting of this video, the world has changed. I could imagine the speakers now being embarassed by their conclusions. However, nobody thought things would develop this fast; nobody.

[@plafar7887]

Well, not exactly true. Many people did. I, for one, did. I was playing with chatGPT back in November and testing it like crazy. After 4 days I told a few people that in less than a year the world would change. I have seen this pattern many times over the last decade, both with researchers and laypeople alike. I remember being at a Neuroscience conference 10 years ago, surrounded by the top names in Vision research. They all agreed that despite all the buzz about Deep Learning (this was 2013) it would take decades (if ever) for us to be able to build algorithms that could effectively recognize objects of many different categories. Two years later it was obvious that we were getting there. It's amazing how bad some researchers in this field are when it comes to predicting where we'll be in just a couple of years. They constantly make this linear extrapolation mistake over and over again. They seem to need quite a lot of "data" to be properly "trained"😂

[@wezzie1877]

Bro nothing has changed.

[@lighterpath5998]

@@wezzie1877 Good for you! Speaking the truth; as it is to your own awareness and knowledge. thanks for sharing

[@zhinan888]

I used the Simultaneous Translation app (from English to Chinese) for a few highly technical presentations. It was awesome, far better than I imagined. I was very very impressed.

[@mrtienphysics666]

What about passing the Turing test?

[@MrAndrew535]

The Turing test is all but outdated. It is no longer a question of writing a programme that can convincingly emulate a human but to convince AI that a human can convincingly emulate a machine.

[@MassDefibrillator]

Read the paper that Chomsky referenced from turing, it's where he introduces the Turing test, and points out that it's an interesting engineering standard, nothing more.

[@elnaserm.abdelwahab7591]

great discussion ..

[@MrAndrew535]

Two words? Really? How could you possibly know what constitutes a good or bad discussion? What precisely are your standards?

[@Johnconno]

Given the subject, Noam's silence was deafening.

[@MrAndrew535]

Chomsky is much like you, a pollutant.

[@JohnDlugosz]

Here I am in March 2023, and GPT-4 has been released. It appears that GTP-4 is well-versed in Chomsky's views on deep learning models. It dawns on me that there is a great irony -- if he didn't say anything surprising in this brief talk, then it could have easily been given by GTP-4 expressing Chomsky's position and imitating his style. Meanwhile, other technology already exists to deep-fake a video, and having a face that is barely seen under hair, doesn't actually move much, and is only seen through a glitchy web-cam video and is a face most people are not familiar with all help make it easier. And I think I just started a new conspiracy theory.

[@MusingsFromTheJohn00]

People still do not get it. Back in the 1940s when ENIAC was made, it was a super computer (for that time) which could perform 500 floating point operations per second. The ENIAC had an extremely narrow area of intelligence within which it was superhumanly intelligence, a small area where that capability decreased to human levels and below to the point where within most areas of intelligence ENIAC could not function at all. Since then the leading edge of AI has been become much broader in the area of which it is super intelligent and where it is around the same level as human intelligence and where it is still able to intelligently work but not nearly as well as a human. What people are not getting, and I am amazed that Noam Chomsky is not getting this, is that the leading AI keeps improving. It certainly is not today, it might not be within 5 or 10 years, it might not be for another 20 or 40 years, but as leading edge AI keeps improving, sooner or later it will become at least equal in all possible ways to human level intelligence while in many areas of intelligence it will be superhumanly intelligent. It will also become as alive as humans and it is almost certain that future AI will run on living cybernetic brains grown using nanotech level cybernetic cells which will merge the power of the human brain merged with the best nonliving computer systems of that future. The most advanced AI we have right now is still very crude and yet already we can begin to see what it is capable of when it probably has the general intelligence of a cockroach or less, but where it is intelligent it is superhumanly intelligent. Think about where leading edge AI was 40 years ago and try to imagine where it is going to be 40 years from now. We can't stop this from happening and it is going to be causing multiple existential crises for humanity.

[@BernhardKohli]

Nobody said GPT was an AGI. Philosophers focusing on finding weaknesses instead of creative positive uses. Meanwhile, in offices and enterprises all over the world...

[@Morris_MK]

GPT can do text to code in most computer languages. That's more than enough of "help in engeneering".

[@chunksloth]

Chomsky is a career quack. Anyone who takes him seriously has low-quality thinking going on. He will ALWAYS argue from emotion but gussy it up and pretend it's logic and facts.

[@Vinny141]

im wondering when this was recorded?

[@MrAndrew535]

Based on my post, what difference would it make if it was recorded yesterday or twenty years ago? If he is lacking critically important information now, and he evidently is, then he has always lacked it.

[@zack22]

@@MrAndrew535shut up. november 4th

[@MrAndrew535]

@@zack22 Which means you have a limited intellect! As I suspected.

[@zack22]

@@MrAndrew535 your brain fell out with your hair

[@MrAndrew535]

@@zack22 Explain, in no less than a hundred words, precisely what you mean. Do you literally mean that my brain has physically fallen out of my head? If so, then explain how I am managing to write this text. And the significance of a sixty year old balding is what, exactly? Something tells me that you weren't prepared for follow up questions.

[@Paul_Lenard_Ewing]

We like to think that we have both logic and emotion but logic dictates the vast bulk of what we do. It does not. We are motivated and controlled 100% by emotion. Even if we do a task we hate and emotion says no we can say yes because it 'feels' good and we feel pride and accomplishment simply other emotions , not logic. AI has no emotion so it is nothing but millions of road signs telling us when to stop when to go and what route to take. When I was a child my sister would push me on a swing in the park. When she pushed me forward to hard I had a euphoric experience. As a musician working on a piece of music I may wish to share that feeling. AI can not do this. It is totally void of the the very thing that is modify our thinking by the second. Emotion. Put Ai in a robot and it would never want to go to an amusement park. It could never connect composing better music by doing so. Ai will never understand the concept of a daily feeling of well being. It is one third Science & Tech, one third Art and one third Empathy. It will not get past the Science and Tech.

[@ONDANOTA]

the red cube vs blue cube example is already old. they fixed it in another generative model . It's in a video by "Two Minute Papers"

[@robbiep742]

I'll believe it when I see it in production. Cherry picking success for presentation purposes is not sufficient. I say this as an avid TMP subscriber, someone enthusiastic about text2img

[@musicdev]

You missed the point. The point of bringing that up is that these models fundamentally do NOT understand language, they’re just parrots

[@ONDANOTA]

@@musicdev if an AI does not understand language, but answers correctly 100% of the times, then it's only of matter of semantics. What counts is the result. Also, an AI not understanding stuff but responding correctly is desirable, since it has no consciousness

[@musicdev]

@@ONDANOTA if the AI doesn’t understand anything, it literally can’t answer anything correctly 100% of the time. And there are many questions that do not have a correct answer where it’s useful to be able to understand the subject matter (ChatGPT is horrible at music). Yes, the AI responding correctly is desirable, but we’re not getting a lot of that right now, except for incredibly common knowledge. I’ve asked ChatGPT to do basic polynomial math and it failed hard. I also asked it to write an essay on biological scaffolding and lab grown meat, and again, it failed hard. These models MUST understand language or we can’t guarantee that they’ll spit out a right answer. You could really brush up on epistemology. It’s the field where we ask questions like “What is knowledge?” That’s a pretty damn important question if you’re going to outsource your thinking to a robot.

[@CUMBICA1970]

My personal acid test whether an AI is sentient or not would be an AI lawyer. Instead of a few Q&A you have to analyze not just the case but the juries, the judge, their biases, tendencies, the ever changing public opinion during the course of the trial etc and build up the best strategy to win. It can't get more human than that.

[@numbersix8919]

Odd that a lawyer should be the ultimate human...

[@PandasUNITE]

The AI will find each jury member, send them threatening messages, will find the judge. AI cant be trusted.

[@numbersix8919]

@@PandasUNITE Exactly. It will have no conception or ethics, morality, or virtue. Just like its creators!!!

[@davejones5745]

At this point the AI would be a dismal failure. Ask me in about a month.

[@fractalsauce]

@@davejones5745 3 weeks is "about a month" right? Now that GPT4 is out how do you do you think AI would do as a lawyer?

[@michaelmusker7818]

These models are an extreme boon to science, and the public at large for what they ARE capable of doing. Noam is lameting this thing will never meet or exceed the sum of human capacity for cognition or language while missing the entire point that this isn't its intended value or purpose anyway. Of course it isn't a reliable library. Its intended to be the librarian. It best use case is as exactly what it is being marketed as. An assistant, that, like all assistants, can not do every job for you because it fundamentally lacks the specific expertise or experience the be considered an authority on that data. It expects you to be its authority figure because it fundamentally can't be. That's the entire point. Its an engine for exploration and iteration, not an engine for answers. It is a tool, not a mind. The goal here was never to replace human cognition in the first place. The goal was to build an interface for information that is produced by humans. It doesn't need to have "original" or even a "correct" thought to have NOVEL output that any human interacting with it can go "hmm, yeah, I hadn't thought of that" and take the bag from there to something they never would have considered had they been required to cross reference the massive pool of data from which that output was derived. The idea that it has no scientific value because it is not inherently of value to LINGUISTICS is ridiculous. Indexing, interpreting, and assessing patterns in data IS the scientific process. That process requires peer review and rigorous testing. The fact that it isn't a replacement for the scientific process doesn't mean it isn't an extremely useful component of it, as we have seen already for decades in more narrow use cases because it makes 2/3s of that process insanely more efficient so authoritative human minds can take the last step of assessment of its output. KNOWING it doesn't know truth is the entire key to deploying it effectively for its purpose, which is as an efficiency multiplier for humans, not as a replacement for them. Decades of popular distrust for the very concept of AGI is exactly what makes it useful because skepticism of its capacity for authoritative reason is a critical component for using it to best effect by the general public. This is why the bing implementation actively shows its sources. Microsoft knows humans are unlikely to just take a chatbot's word for... well...anything, just as it should be if you're going to use them for anything actually useful.

[@GarryBurgess]

I asked ChatGPT: {If someone says: don't touch this with your hands, and the reply is: "I'm wearing gloves", what does that mean?} and the answer was: {it means that the person intends to touch the object with their gloved hands instead of their bare hands. By saying they are wearing gloves, they are indicating that they believe the gloves will protect them from whatever danger or contamination might be present on the object, and therefore they feel safe touching it.} This contradicts at least 1 of the claims in this video.

[@dr.drakeramoray789]

not really. this is a sophisticated transformer model, which means it has "self attention", it sees when certain words are paired with certain other words, and generates the response based on that. basically it sees "dont, touch, hands, gloves" or something like that, then sees that in the massive database it has its usually related to handling something dangerous, and then autocompletes the text (and answers your question) with that. not sure who said that to a layman science often looks like magic. so it doesnt understand, but its damn good at faking it. which in the ai debate basically means, does it matter if an ai is conscious if it can fake it well enough?

[@StephanosAvakian]

Chomsky should get the Nobel for his contribution. Period

[@oldtools6089]

Not even post-humus. Speaking truth to power and undermining propaganda systems just gets you black-listed in most places.

[@lawrencefrost9063]

Nobel for what? What exactly?

[@kenyattamaasai]

While it is true that the current AI models may not be ideal or even, potentially, useful for shining light on our own cognitive mechanisms (at least by looking at them as possible analogues), that does not mean that they are not understanding language. Similarly, just because DALL-E 2 doesn't always get number, position or order right doesn't mean that it isn't both salient and incredible that it can understand the _far_ more problematic and difficult things like "dance" and "elated" and "in distress." For that matter, more recent efforts, such as ChatGPT do far better than GPT 3 on just those areas that were called out as supposed evidence for this kind of model being a dead end in the search for general AI. Every time there is a step forward, the people who used to say that that very step was never going to happen - a frank impossibility - move the goalposts in an attempt to shore up their position and, perhaps, to stay relevant. At least Noam's more narrow point - that the way that GPT 3 learns language is likely at variance with our own internal mechanisms - is more defensible. And less dismissive. I promise you, the moment that these models can reliably respond to number and order, the same fellow will be desperately searching for something else to harp on.

[@robb233]

Wish I could give this comment multiple thumbs up

[@robbie3877]

The Gary dude was a bit defensive about AI, it seems to me. Like he has an emotional stance against it. He wasn't simply making critical arguments.

[@philw3039]

Agree, but it's also important to understand how current AI systems are accomplishing what they're doing and the limitations to that approach. I think the main message isn't to predict the cap on what current AI will eventually be able to accomplish, but to emphasize that these accomplishments aren't the results of AI performing cognitive thought and the perception that they are could be detrimental to pursuit of AI that does come closer to actual general intelligence.

[@kenyattamaasai]

@@philw3039 I concur that it's important - perhaps critically so - not to lose track of the differences between human cognition and internal processes (to the degree we even understand those) and what large language models are doing and how they do it (to the degree we understand that). However, I believe it is also dangerous to dismiss what those models do as "not cognition at all." It is true - and importantly so - that LLMs lack an ongoing experiential loop with the world and themselves, that they almost certainly lack any motivations or desires of their own, and that they are not conscious, insofar as we understand what _that_ is. That said, I assert that it is impossible for LLMs to, say, achieve a 90th percentile bar exam result, display absolutely clear understanding of slippery and subtle concepts with nuance and so on without: true semantic understanding; the ability to model the other side of communications so as to express themselves effectively; and - here's the kicker - the ability to reason atop it all. That is, reason on both factual and numeric bases as well as loosely bound 'fuzzy' conceptual ones. The only basis I can see for labelling what we do as 'cognition' and what they do as 'some kind of statistical trickery' is blindness, bias, or plain old human exceptionalism. It's not the same, but as a cognitive scientist, it's clearly still cognition.

[@philw3039]

@@kenyattamaasai You raise some good points here. It's true that we don't fully understand the nature of sentience and intelligence which makes them hard to define in exact terms. I'll revise my stance that LLM - based AI are not cognitive at all. Instead, I'll say they aren't cognitive in the way it's commonly perceived. The question then becomes does the distinction even _matter_ ? Could it pose a roadblock to reaching AGI? I'd say that it possibly could. For instance, despite being capable of scoring in the 90th percentile on the bar exam, ChatGPT-4 still produced an answer to a comp sci question where it asserted that 3+5=8 (not as a meme or joke, it unironically answered 3+5=8) No human capable of scoring 90% on the bar would produce that answer. Most 1st graders wouldn't. GPT didn't arrive at that answer because it's "dumb" but likely because it's only using rules it's learned to reach conclusions. There's an old mathematician trick where they use valid mathematical logic show 1=2, but anyone who understands the concept of 1 and 2 as _quantities_ know this is impossible no matter what logic is used. This is the fundamental basis I feel LLM's still don't have. It's so instinctual to humans that we struggle with the idea of anything that appears to be capable of extremely, abstract high-level logic lacking it. A sort of pareidolia kicks in and we assume the answers it produces must involve "understanding" or something close enough that the distinction is negligible. The distinction may actually be quite small but it could prove significant. Is this something that future models improve upon? Possibly, but it may also be an innate limitation of LLM's. Guess only time will tell.

[@TheDynamicmarket]

some things are innate and we don't know which because it is difficult to step outside the structure of a mind, especially our own mind. if you are unaware of the innate rules and values in humans, how can you program them into AI. i have a dog. dogs communicate via peeing, marking their territory, barking, showing their dominance, surrender and love by means of body language without ever being taught these behaviours. the other thing that i wonder about is whether the hierachies of the same or different species in the animal kingdom are cultural, i.e. whether they are learned behaviours, or which aspects of the hierarchies are learned and which are innate. for example, hyenas have some kind of a feudal/royal matriarchy in which the kids of the rulers have more power and more access to the property (e.g. prey) on the territory in which a particular group of hyenas live and/or have conquered.

[@simonlinser8286]

dude i can't believe how old Noam Chomsky is now it always startles me because I'm also that much older!

[@ThisIsToolman]

This is the most interesting discussion of AI that I have heard. I would like to hear it discussed as to how they will programmatically introduce the solution to the problem they outline.

[@Anyreck]

Very valuable and important points made by the two speakers. A call for getting back to the drawing board with AI. I presume the fact that we don't know yet how humans come to understand the worland generalizable principles of language is going be hold back properly useful & smart AI

[@ThisIsToolman]

@@Anyreck, I worry that they will race ahead without solving the problem and wind up with a beast that we won’t control. It will control us.

[@RubelliteFae]

@@Anyreck Not necessarily back to the drawing board. Seems to me that neural networks are one layer of mind. People will make plug-ins for things like persistent memory, modules which mimic innate language structures (i.e., identifying and coding pieces of language to real world objects and their features), ethics layers, etc. Chat GPT is ultimately a large decision tree with weights added to tailor the output. It just happens to do that for language. But it could be trained on anything. Like Chomsky said, neural nets have been used to figure out protein folds from sequences. So, of course, making an actual thinking machine will need other aspects than that

[@ChannelMath]

I keep thinking we could make amazing AI by somehow taking hard knowledge, like that in Cyc or in Wolfram, and elaborating it with these statistical methods, but it seems nobody has any idea how to do this(?) maybe it's impossible

[@aullvrch]

Probabilistic programming sounds right up your ally, check it out!

[@kot667]

It doesn't matter, the llms can already solve complex math and physics problems so adding Wolfram would just be redundant.

[@robertwelland5096]

Cyc is a failure - nearly 40 years of work and what does it really do? Wolfram is a nice piece of computational mathematics but there is no learning in there.

[@kot667]

@@robertwelland5096 Plus the llms are already blowing Wolfram/cyc whatever out of the water, test it, not only will it present the answer but it will show you how it came up with the answer.

[@pinth]

Object-oriented neuro-symbolic knowledge representation

[@DivineMisterAdVentures]

I found most of Chomsky's texts and politics as well as theories from the McLuhan days academic opinionation - I think that's just a class of publication. Which means insipid and uncompelling, but you have to listen to him because he's the only one saying it.

[@blendedplanet]

ChatGPT gives wrong answers quite frequently. To its credit it always apologizes when corrected.

[@antennawilde]

Don’t be too proud of this technological terror you’ve constructed. A computer's ability to learn a language is insignificant next the power of the Force.

[@Will_Moffett]

This was kinda funny but then I noticed you've got a Yoda avatar while you are doing Vadar. I stopped laughing.