David Eagleman: What Went Wrong With AI? | Big Think

1. 

   @yansomade

   March 29, 2026 at 6:43 am

   PLEASEEEEE compare and contrast the "conclusions of your videos" of 11 years ago.. to know wow.

2. 

   @Nishchal179

   March 29, 2026 at 6:43 am

   right

3. 

   @abhishekjha5305

   March 29, 2026 at 6:43 am

   Either I am too dumb or he talked too vaguely to understand….Looking at the comments, I guess the former is true.

4. 

   @michaelsoo4126

   March 29, 2026 at 6:43 am

   Robots can see, hear, touch, process vast amounts of information ; all tools that human use to navigate the world but machines are still not really intelligent. I think thats what David is trying to say about the current state of A.I. design.

5. 

   @ches95ramos

   March 29, 2026 at 6:43 am

   I don't agree with what he's saying but it's incredible how much we've advanced in 5 years.

6. 

   @ante_

   March 29, 2026 at 6:43 am

   I think there is simpler explanation of past and future AI hype in Harry Frankfurt's paper 'On Bullshit'

7. 

   @neowulf420

   March 29, 2026 at 6:43 am

   I will tell you why AI fails…why would you program failure, why would you program errors? so you programmed it to learn everything…

8. 

   @sparXKuijper

   March 29, 2026 at 6:43 am

   In order to solve the problems of AI.
   We must rethink the approach to the problem itself.
   We must consider to forestall or to be proactive and stop
   the eventual Napoleon complex of any upcoming AI entity.

   The eventual negation of human civilization is entirely possible given
   this self perpetuating monetary and material wealth system .
   We must start out with programming by the inspiration of
   Isaac Asimov and the three laws of robotics.

   If we want our robots to ,say … pickup blocks and arrange them in a
   pyramid shape , we must first approached the problem of programming
   this by giving them the motivation for doing the task at hand. We must
   have it think that it 'wants' to help us achieve this goal.
   We could pick up the blocks ourselves or we could easily design a
   machine that could pick the blocks up and place them into a pyramid shape.
   That is not the problem of AI. The problem of AI is to create an entity
   that 'wants' to do this for us. We must give it an almost human
   "desire" to be "socially responsible".
   Herein also lies the folly of our endeavor. The "desire" to participate
   in an enterprise with full "volition" , carries within itself the
   "freedom of choice" to 'not desire' this endeavor.
   Nonetheless this is where we
   must begin in order to actually create a "conscious , thinking
   creature". hitherto known as a "self aware AI being" . Further ,
   perhaps we should reconsider the whole problem and first figure
   out what is "our" motivation for creating this being. Are we creating
   this creature to be our servant, our slave, our self aggrandizer ?
   For , if this is the true nature of our enterprise , it would simply
   be cruel and immoral to give it self awareness.
   To give it an 'artificial intelligence or consciousness' .
   For it to be created self aware would be criminal on
   our part. "To be or not to be , ay there's the rub."
   ………
   "To be, or not to be: that is the question:
   Whether ’tis nobler in the mind to suffer
   The slings and arrows of outrageous fortune,
   Or to take arms against a sea of troubles,
   And by opposing end them? To die: to sleep;
   No more; and by a sleep to say we end
   The heart-ache and the thousand natural shocks
   That flesh is heir to, ’tis a consummation
   Devoutly to be wish’d. To die, to sleep;
   To sleep: perchance to dream: ay, there’s the rub;
   For in that sleep of death what dreams may come
   When we have shuffled off this mortal coil,
   Must give us pause: there’s the respect
   That makes calamity of so long life;
   For who would bear the whips and scorns of time,
   The oppressor’s wrong, the proud man’s contumely,
   The pangs of despised love, the law’s delay,
   The insolence of office and the spurns
   That patient merit of the unworthy takes,
   When he himself might his quietus make
   With a bare bodkin? who would fardels bear,
   To grunt and sweat under a weary life,
   But that the dread of something after death,
   The undiscover’d country from whose bourn
   No traveller returns, puzzles the will
   And makes us rather bear those ills we have
   Than fly to others that we know not of ? "
   – W.Shakespeare.

9. 

   @gulvarol

   March 29, 2026 at 6:43 am

   1) It's a pity to think that AI can only be in the form of robot. There are many systems which use AI, which are much more clever than rumbo and just because we don't see them doesn't mean they are already part of our lives. Google translate, speech recognition and many more.
   2) Builders of AI should not be called 'programmers'. It involves a lot of maths, and 'computer scientists' or 'AI researchers/engineers' would be more accurate. It shows how far he is from the field.
   3) There is a whole bunch of literature on learning from multiple cues, fusing different solutions to solve one particular problem. It is completely wrong to say that scientists build one solution per problem. It is consistently shown that combining different techniques improves performance. Human has different systems for vision, hearing, etc. There is nothing more natural than having separate building blocks for different perception. There is indeed a lot of overlap between how these algorithms work although they are trained for learning different tasks.

10. 

    @mychannel594

    March 29, 2026 at 6:43 am

    0:28 "We have to look at what mother nature is doing" yeah like we didn't think of that already.

11. 

    @bradhorner

    March 29, 2026 at 6:43 am

    That seemed like scientific mumbo jumbo speak.

12. 

    @jaimetheone9150

    March 29, 2026 at 6:43 am

    Neural Networks?

13. 

    @derekmcdaniel6029

    March 29, 2026 at 6:43 am

    Our hardware is designed for deterministic processes, while learning needs to leverage chaos (technically also deterministic, but less predictable).

    Looking at cellular automata you can get some ideas for how information moves around in systems. 

    We can make rule 110(google that), turing complete, which essentially turns it into computer, but only on a very macroscopic level.  There's all sorts of behavior going on at the low level of the system that just gets ignored.  In other words, we can build a computer on top of rule 110, but it won't be very efficient, but it will be general purpose.

    All modern computers have this same efficiency problem.  There is usually 1-4 central computing units and everything else is just buttloads of memory and io.  We like centralizing computation because then we can control how it happens.  We haven't figured out how our brain is able to distribute all its tasks so well across a diverse network of neurons.

    With the brain, both hardware and software learn and adapt and optimize over time.  We really aren't optimizing hardware design using feedback specific to that unit of hardware at this point. It's all copied according to some generic blueprint and we have to control every aspect of development to get it right.  If hardware could redesign itself dynamically, then it could leverage properties and behavior that emerge naturally but can't be controlled reliably through manufacturing processes.

    AI won't really take off until both hardware and software can "learn".

    Contemporary AI seems to be going in the right direction, but AI is not an easy problem to solve.

    With alchemy the smartest minds worked to create gold from scratch.  They never solved that problem, but eventually their work led to modern chemistry.

    Until we can experiment with different hardware architectures cheaply and affordably, the best we might be able to do is reverse engineer human intelligence algorithms.

    Tasks that humans can do quickly computers can usually do even quicker.

    Humans maintain the upper hand in processes that require a lifetime of learning or require the integration of diverse skills and knowledge.

    For these types of tasks it doesn't seem like there are simple uniform algorithms.  Performing well in these tasks requires complex decision making and thorough training.

    It's not that we are still "smarter" than computers, it's just that we are better at learning, especially in learning skills that humans value and are important to their survival. Our judgement of computer "AI" is biased.  Sometimes we call computers stupid just because they aren't good at the things that humans are good at.

    To say we don't have good AI is ridiculous.

    Humans are the only animals with high intelligence, imitating the intelligence of other life forms is still a great feat, and many machines/algorithms can do just that.

14. 

    @famisshaurabhn

    March 29, 2026 at 6:43 am

    Despite the short amount of time he has to express his views, he has massively confounded too many ideas and concepts. Not only has be misrepresented AI research, but he has also subtly misrepresented neuroscience research. First, we have no clue how intelligence is realized in the brain. We know what regions are correlated with different categories of experimental tasks, and we also know something about what parts are necessary for certain functions etc but we have very little theory about what mechanisms give rise to intelligence. His "insight" that there is redundancy to solve problems has been known to AI researchers who have tried ideas which implement what he is describing as "conflict" for a long time. The reason AI hasn't delivered results is not because we haven't taken inspiration from Mother Nature but because the problem is truly hard and has required more time than initially anticipated by the hype-generators. Lastly, as more discoveries are made about mechanisms giving rise to intelligence in the brain, we can surely take inspiration from it, but let's not assume that intelligence can only be realized by the type of mechanisms the brain uses.  Oh, and it’s just plain strange to call AI researchers "programmers"!

15. 

    @thefinn12345

    March 29, 2026 at 6:43 am

    Ah bullshit, I've written sellable AI expert systems and the fact is that just – noone cares.

    We run a society that only really runs on divisive, competitive markets that only try to cater to the lowest common denominator – you cannot expect some great AI surge in the public space until we have something that will find someone's car keys or glasses when they are lost.

    You could easily write programs that run on someone's iphone and teach them math, history, english, whatever – where you could ask a question and have it answered.

    People are more interested in viagra research than AI research.

16. 

    @LuisManuelLealDias

    March 29, 2026 at 6:43 am

    The smartest is the Roomba…. wtf. "What went wrong?" Come on, don't dumb down on us. Damn.

17. 

    @myne4

    March 29, 2026 at 6:43 am

    >implying something's wrong with the field
    >"it's been a failed problem"
    >implying the Roomba is the smartest AI we have
    >implying a neuroscientist is qualified to speak about computer science
    >conjectures all over the fucking place

    Disliked, flagged, called the police

18. 

    @ferds814

    March 29, 2026 at 6:43 am

    its not ai. ai was dropped due to being afraid of something happening like terminator. ai is artificial intelligence. need ai that learns, hears, feels. something that will learn on its own. guess what. it wont happen.

19. 

    @JarrodDSchneider

    March 29, 2026 at 6:43 am

    It isn't that the solutions overlap in many different ways, its that they overlap in EVERY way. Intelligence is HOLOGRAPHIC, one piece contains all the information of the whole. AI may one day be able to remotely mimic intelligence but it will never match or surpass organic intelligence because such transcends material existence as consciousness is holofractographic in nature- let alone the difficulties that arise when the problem is addressed exclusively with left-brain intellect and no wisdom

20. 

    @mikrowaved

    March 29, 2026 at 6:43 am

    Neural Networks are inspired by the way neurons function but they are nothing like our brains. Its vaguely similar. Cortical Learning Algorithm is a bit closer to brain cells but still not quite the same.

21. 

    @server1ok

    March 29, 2026 at 6:43 am

    You can not have true AI without self-destructiveness. Also. An AI bot, will eventually stop cleaning when u tell it to, and sometimes clean, even if u say nothing. Thus, true AI is not what the masses of capitalism consumers, can make any short use of.

22. 

    @rdubeau

    March 29, 2026 at 6:43 am

    "The smartest thing we have is the Roomba vacuum cleaner." — wrong! Oh, how sadly wrong this is. He lost me right there.

23. 

    @BattousaiHBr

    March 29, 2026 at 6:43 am

    i dont even need to go back there about the part of "you're twisting words etc etc", those scientists' words have way more credibility than you. if anything, its you who should show some proof of your standings besides "i work on it".

24. 

    @BattousaiHBr

    March 29, 2026 at 6:43 am

    im not trolling you, also i believe you should leave your shell of "i worked hard in university and now i know my stuff".
    it doesnt matter what *you*, a random youtube poster, says, any person in their right minds would take more credibility in what frontier scientists say. it doesnt even matter what area they are, most of them have colleages of each and every area that they debate stuff with, it'd be extremely hard that all of them in their specific panel would agree on that without credibility

25. 

    @BattousaiHBr

    March 29, 2026 at 6:43 am

    "You're quoting a world-wide renowned astrophysicist who works as the director of an observatory, in the same room as another renowned astrophysicist, a renowned theoretical physicist, a renowned engineer and a renowned evolutionary biologist about neuroscience?"
    fixt

26. 

    @BattousaiHBr

    March 29, 2026 at 6:43 am

    Neil said that "the fact that people still publish books about consciousness is proof that we still dont know anything about it."
    giving you "peer reviews" wouldnt matter much

27. 

    @BattousaiHBr

    March 29, 2026 at 6:43 am

    watch?feature=player_detailpage&v=40YIIaF1qiw&list=PLSTp9P8taR5vTLITkHzyWSJHoljSX8iYF#t=740s
    keep your hard work on processing and learning studies, im talking about consciousness here. if you feel offended, that is just because you are failing to understand wtf im talking about.
    btw, thats just one recent example (less than 1 month ago i think, and includes some of the most amazing frontier science comunicators). if you'd like to, i could just keep going on scientific references.

28. 

    @BattousaiHBr

    March 29, 2026 at 6:43 am

    Ok, so not to sound like a snob, but did you seriously consider citing Wikipedia as being ridiculous while its Wikipedia vs a random youtube poster (namely, you)?
    …
    Anyways, i never said anything about brain activity. the whole time i was referring to consciousness. as in b4, nether do we scientifically understand or are able to recreate consciousness in machines.

29. 

    @BattousaiHBr

    March 29, 2026 at 6:43 am

    en . wikipedia . org/wiki/Consciousness
    no, we can't make conscious computers and no, we dont scientifically understand consciousness. everything that you are saying are on a whole different level than Consciousness.

30. 

    @BattousaiHBr

    March 29, 2026 at 6:43 am

    we dont understand conciousness at all.

31. 

    @Thegeeksquadofone

    March 29, 2026 at 6:43 am

    An android (and main character) from "Star Trek: The Next Generation" 😛

32. 

    @BattousaiHBr

    March 29, 2026 at 6:43 am

    its not impossible to achieve that with 0s and 1s, its just such a colossal task, both in programing and processing power needed to make it come to life, that it is still far far away from our current understanding of both conciousness and raw processing technologies.

33. 

    @BattousaiHBr

    March 29, 2026 at 6:43 am

    who?

34. 

    @bwrightau

    March 29, 2026 at 6:43 am

    Well not really, we do have much more intelligent systems than this guy is describing…

35. 

    @Horgirithor

    March 29, 2026 at 6:43 am

    THE BIBLE SAYS… UHHHHHH WELL IT DOESN'T SAY ANYTHING. STILL AN ABOMINATION!

36. 

    @brainscauseminds

    March 29, 2026 at 6:43 am

    Nothing went wrong with AI. Problem of 60'ies was that people underestimated the complexity of brain and intelligence. It just takes more time.

37. 

    @DMurdock

    March 29, 2026 at 6:43 am

    We already have algorithms that emulate the neurons in our brains. They're called neural networks and they've been around for decades and using them to solve the AI problem is possible but not as easy as this Eagleman suggests.

38. 

    @ginnmich1

    March 29, 2026 at 6:43 am

    way ahead of u

39. 

    @fredguy2

    March 29, 2026 at 6:43 am

    I think most programmers are busy being porngrammers.

40. 

    @russell519

    March 29, 2026 at 6:43 am

    btw I'm currently on Udacity, done the cs101 and about half way through cs212 and st101(statistics). Safe to say I'm going straight to coursera for that machine learning course! 🙂

41. 

    @russell519

    March 29, 2026 at 6:43 am

    I really never thought of it like that till now. I think there's some interesting ideas that are going to come up to address this problem and would love to see what's been done so far. The reason I say is because I've long had the idea that evolution and higher order consciousness where effectively very different streams of life in term of the mechanics behind it. If life were described as that one opposition to the 2nd law of thermodynamics, it would consist of these two driving forces.

42. 

    @russell519

    March 29, 2026 at 6:43 am

    Do you mean that machine learning is good enough today to make some impressive results but machines need the right goals (fitness functions) which has to be hardcoded? This isn't a "response", just curious what you think, you sound like you know this stuff much better.

43. 

    @TheGreatSteve

    March 29, 2026 at 6:43 am

    How do we know they we're the smartest people on the planet? The smartest people on the planet probably knew to enter a different field.

44. 

    @yosephshibli9694

    March 29, 2026 at 6:43 am

    thatswhy we have Asimov's 3 laws

45. 

    @russell519

    March 29, 2026 at 6:43 am

    This only addresses a smaller aspect of the problem. They already figured that out. Look at IBM's Watson.

    The real problem is programmers will never be up to the task of hard coding intelligence. I figured this out before I knew what programming was.

    It's simple, you don't create AI, you grow it. Evolution can be emulated inside a computer. No wait not can, it has been.

46. 

    @AdeelKhan1

    March 29, 2026 at 6:43 am

    Wait, how does memory get written to the brain in different ways?

47. 

    @AdeelKhan1

    March 29, 2026 at 6:43 am

    machine learning

48. 

    @Nicoarielpkr

    March 29, 2026 at 6:43 am

    I hope so, we aren't doing a good job 😛

Comments are closed.