back to article Meet the man building an AI that mimics our neocortex – and could kill off neural networks

Jeff Hawkins has bet his reputation, fortune, and entire intellectual life on one idea: that he understands the brain well enough to create machines with an intelligence we recognize as our own. If his bet is correct, the Palm Pilot inventor will father a new technology, one that becomes the crucible in which a general …

COMMENTS

This topic is closed for new posts.

Page:

      1. xperroni

        Re: Let a thousand flowers bloom

        [N]ot so long ago the main driver for results was the military, and I don't think that private enterprise's goals are much more worthy. Better to strive for a better understanding of who we are as humans than settle for models that can help us to destroy or one-up each other.

        And yet you're making your opinion known over the Internet.

        The world is never simple...

      2. xperroni

        Re: Let a thousand flowers bloom

        I will say that trying to emulate pre-cortical brain structures is unlikely to elicit much excitement from the general populace (...).

        Of course, if the general populace could tell the difference between cortical and pre-cortical brain structures, teachers should get a raise. Hell, give me a scalp and a brain to dissect along those lines, I'm bound to make a fair number of mistakes myself.

        More to the point, there are a number of skills you'd want an "intelligent" machine to have (such as task selection and motor control) that stem from pre-cortical / sub-cortical structures, so you'd at least want to have a look into how they work, if you're working on a neurologically consistent model of intelligence.

      3. psyq

        Re: Let a thousand flowers bloom

        Well, pre-cortical brain structures can do impressive things as well.

        Lizards and frogs do not have neocortex, but are doing pretty well in surviving. Even octopuses are pretty darn smart and they do not even have brain parts even the lizards have.

        Today we are very far even from the lizard vision (or octopus vision if you will), and for that you do not need an enormous neocortex. I am pretty sure that something on the level of lizard intelligence would be pretty cool and excite the general populace enough.

        These things are hard. I applaud Jeff's efforts but for some reason I think this guy is getting lots of PR due to his previous (Palm) achievements while, strictly speaking, AI-wise, I do not see a big contribution yet.

        This is not to say that he shouldn't be doing what he is doing, to the contrary, the more research in AI and understanding how the brain works, the better. But too much hype and PR can damage the field, as it happened before, as the results might be disappointing compared to expectations.

    1. John Sanders
      Facepalm

      Re: Let a thousand flowers bloom

      Listen if anyone were on the right track with understanding how the brain works by know we should have being able to understand how the nervous system of something much smaller and simpler works, lets say an ant.

      Or a ladybug, or even an earthworm.

      Nothing that we can fabricate or model now doesn't even compare to what the brain of humble caterpillar or a spider can achieve, let's not talk about a bird or a small mammal, and no, we're not talking about being self aware.

      We're trying to understand the brain from the complete wrong angles, it is not so much about how it is cobbled together, but how does each cell know what it has to do from the moment the brain begins to form.

      We do not have even the flimsiest idea on how does nature encodes information. We barely understand that there are instructions on the ADN, but all we do is the same we did with the atom, throw things on it to see what happens.

      And for atoms at least we have some crazy theories.

  1. Truth4u

    ai already exists

    this comment is written by ai program v 0.3

    1. John Sanders
      Devil

      Re: ai already exists

      Chomskybot is that you?

      1. Vladimir Plouzhnikov

        Re: ai already exists

        Shhhhh! He is still in alpha...

  2. JaiGuru

    This man is a traitor to flesh. Advanced AI is being developed to replace YOU who read this comment. You don't have a share in the future it will bring about. Anyone who betrays flesh is guilty of crimes against humanity.

    1. Destroy All Monsters Silver badge
      Headmaster

      You may want to consider removing yourself from the the genepool with the appropriate application of a large chunk of meat before you morph into Sarah Connor.

      The "OH NO COMPUTERS! MUH JOBS!!!" discussion has been had in the 70s/80s/90s.

      Turns out the jobs that went out of the window were the heavy industry ones ... who of the the heavily unionized fellowship would have thunk it?

    2. Don Jefe

      Are you certain that man is, in fact, flesh? How could you validate your conclusion? Is it possible that any verification you received was not actually verification, but was only an illusion of verification created by the AI inhabiting what you believe to be your skull? Is it possible that you are actually the AI and have created a virtual world filled with what you believe to be Humans who believe that your virtual world is the real world that is being run by what the believe to be an AI disgusting itself as a real Human? How would you verify it?

    3. Crisp Silver badge

      Re: Advanced AI is being developed to replace YOU who read this comment

      Someone's developing an AI that drinks coffee and post comments on El Reg all day instead of writing code?

      Who would do something like that?

    4. Truth4u

      RE: This man is a traitor to flesh

      Had to give a thumbs up to counter the 7 thumbs down as the globalists are planning to replace all jobs with bots and machine intelligence by 2035, and no we won't be allowed to live a life of leisure in a post-work society, most of us (90+ %) will be killed with weaponised ebola.

      Only the globalists will have received a clean vaccine. Leaked documents have repeatedly showed that health workers and emergency services will be vaccinated against the ebola but will receive a time delayed cancer virus in their vaccine to kill them after the host population has died and after they have done their job of cleaning up the dead bodies.

      The globalists will be the only ones allowed to enjoy a life of leisure in a post-work society, this is what they have been planning at their annual Bilderberg meetings for over SIXTY YEARS, they have the money and the expertise to do it, and most of the plans are going through the final stages of review. They have war gamed this from every angle and even if it doesn't work or they get caught, they have enough food and water in their bunkers (read luxury underground palaces) that they could start a nuclear war and live underground with their rich disgusting friends for the next 300 years.

      1. Vladimir Plouzhnikov

        Re: RE: This man is a traitor to flesh

        Globalists? You mean environmentalists, surely?

  3. John Smith 19 Gold badge
    Unhappy

    So software implementing a model on a serial computer.

    No I don't think I'll be fearing TROTM anytime soon.

  4. DerekCurrie Bronze badge
    Devil

    To Point Out The Obvious: AI is beyond human capability at this time

    …However, Artificial *Insanity* is WELL within our grasp. Expect it.

  5. gerdesj Silver badge

    Time gentlemen, please.

    I suspect that by including time as an integral component of his model, that will make it a much better model than others that don't. Time is incredibly important to us and how we operate.

    However I don't think that time is an inherent component of the operation or intelligence of the brain. I think it is an emergent property that arises from its operation. I think our perception of the passage of time is just that - a perception. There is a passage of time, it's just not how we perceive it. I certainly don't have an ntp daemon in my noggin.

    By including time as part of the model may cause the model to improve to the point where it can eventually be removed as part of the input and processing and emerge as an output instead - now that would be seriously impressive.

    Cheers

    Jon

    PS Has anyone else noticed the Feynman reference

    1. TechnicalBen Silver badge

      Re: Time gentlemen, please.

      Time is important, but no more so than in current computing. In that, packets, OS, calculations, all have "calls" and "requests" and have time limits (latency) for returns to work well.

      The brain probably does go a step further, with using other temporal methods to work things out, but current software does already do this.

      It's an additional variable on a system that uses many variables.

    2. Paul Shirley

      Re: Time gentlemen, please.

      Time is an inherent requirement of the way neurons work, both as individual analogue 'devices' and networks of them. Accumulating charge and generating pulses are inherently time domain behaviour.

      While it's reasonable to supposed evolution has worked out use that to create intelligence, it's foolish to dismiss alternatives as wrong. Brains may be our best known example of intelligent hardware and well worth study and replication, but that's all they are - 'the best known example'.

      Also surprising to hear that he's the only one considering time as vital to AI, it's implicit in everything that can learn yet built, since it's difficult to learn anything without some acceptance of cause&effect.

    3. John Sanders

      Re: Time gentlemen, please.

      There is an ntp server in your head, the brain uses time in a similar way a board uses a clock, it is used to sync other parts of the brain.

      It is well understood that the brain is kind of slow processing sensory input, and cheats using time to alter your perception.

      Time is nothing but change from a state to another, anywhere in the universe and including your brain, so it is not difficult to imagine that the brain would exploit a mechanism it is intrinsically bonded to.

  6. Anonymous Coward
    Anonymous Coward

    "...mimics our neocortex..."

    Interesting, thought-provoking article. However, there are a couple of historical precedents about "artificial X" which might need some consideration. The first is that successful "artificial X" is seldom remotely like its "natural" predecessor. Compare a pidgeon with a Boeing 747.....some of the aero is somewhat similar, but everything else is quite different. Compare a horse and buggy with a one horse power golf cart......similar function, same wheels.....everything else completely different. Why should we think that "artificial intelligence", if achieved, would have the brain as a direct model?

    Incidentally, the disparaging comment above about the Turing Test misses the point....Turing was trying to provide a simple test by which we might identify "intelligent behaviour". To the best of my knowledge, even today, there is no widely accepted definition which captures the very wide range of "intelligent behaviour". Indeed, every advance in the technology of "artificial intelligence" seems to show more about how LITTLE we know about intelligence.

  7. cortland

    Informed intuitive leap?

    I am intrigued by his approach, and more than that, wondering, because even though I'm near 70; I learned only recently of an undiagnosed "autism spectrum" disorder.

    We humans vary from inconsistently logical, to consistently dogmatic; indifferent to irreverently enthusiastic. Some are idiots and savant at once. Creative intuition, artists and writers, Pirsig's instinctive (if disorganized) mechanic, Zen, Yoga, mathematics, poets, programers and painters; we do not in fact know what intelligence is, preferring to define it as something successful humans have, and I suspect we miss a good deal of our world by assuming that's all there is to it. I suspect that many among us, perhaps "late bloomers," have been people whose intelligence was expressed differently than others and our pedagogues expect, where learning was or is perhaps handicapped not by a different intelligence, but a different apperception. Perhaps this research will opens more eyes to the possibilities in all that we are, all of us, every one.

  8. Anonymous Coward
    Anonymous Coward

    No mention (that I could see) of Ray Kurzweil.

    I read his book "How to create a mind" recently, and one of the main thrusts of his ideas mirrors Hawkins' ideas of time-based analysis/processing.

    Kurzweil, if I read him correctly, thinks that memory is essentially sequence based. That memories are time-based sequences, they may well find that they are talking about the same thing.

    1. psyq

      Actually, the fact that the memory is temporal is known for quite a long time.

      At least since early 90s, after the discovery of spike timing dependent plasticity (STDP) - http://www.scholarpedia.org/article/Spike-timing_dependent_plasticity it became obvious that the neurons encode information based on the temporal correlations of their input activity. By today, our knowledge has been greatly expanded and it is known that the synaptic plasticity operates on several time scales and its biological foundations. There are also dozens of models of varying complexity with even some simple ones being able to reproduce many plasticity experiments on pairs of neurons quite well.

      Since early 90s there had been lots of research into working memory and its neural correlates. While we do not have the complete picture (far from it, actually), we do know by now very well that the synaptic strength is heavily dependent on temporal correlations and that biological neural network behaves like auto-associative memory. There are several models that are able to replicate simple things including reward-based learning, but all in all, it can be said that we are really just at the beginning of understanding how the memory of the living beings works.

      As for Ray Kurzweil, sorry but anybody who can write something called "how to create a mind" is just preposterous. Ray Kurzweil has no clue how to create a mind. Not because he is not smart (he is), but because NOBODY on this planet has a clue how to create a mind, yet. Ray does, however, obviously know how to separate people from their money by selling books that do not deliver.

      If somebody offers to tell you "how to create a mind" (other than, well, procreation, which pretty much everybody knows how to do it) just ask them why is that they did not create it, but instead they want to tell you that. That will save you some money and quite a lot of time. While I do not disprove motivational value of such popular books, scientifically they do not bring anything new and this particular book is just a rehash of decades-old ideas.

  9. Faye B
    Paris Hilton

    New thinking or rehash?

    I can't feeling that Hawkins is simply rehashing previous research in the hope that something will work if you build it big enough. Hierarchical learning ideas have been around a long time ( I did my degree paper on hierarchical learning in rule acquisition using a Pole and Cart simulator) and much of what he is proposing can be equated to the ART algorithms of the 1970s. I am also surprised that he considers the use of binary inputs into neural networks to be effective unless he is using his 'data streaming' approach to replace 'weightings'. Personally I think that AI should be looking at harmonic resonances in neural 'circuits' as an approach to recognition and temporal processing but sadly my mathematics is insufficient to the task of writing a paper on it.

    Paris? well we are talking about an artificial intelligence here..

  10. Pascal Monett Silver badge
    Pint

    "If his bet is wrong, then Hawkins will have wasted his life"

    I simply cannot let that line pass. That is so primitive a reflection as to be ridiculous.

    Whether Hawkins succeeds or fails to build an AI is actually irrelevant. Whatever the end result of this endeavour, he will have succeeded in furthering our knowledge of the brain and the corresponding neuroscience. For that alone, he deserves recognition.

    Personally, I fail to understand how anyone can hope to build an artificial brain without understanding how a real one works. If we have impressive car simulators today it is because we have a very good understanding of how a car works. Without the practical knowledge we have of tire grip, shock absorbers, torque and power, how could one possibly build a proper car simulator ? Building an artificial brain must be the same.

    And Hawkins' remark that the brain does not come with a set of predefined instructions hits the nail squarely on the head. If the opposite were true, we would have no trouble raising children and only one book on the subject would ever have been written.

    I don't know if Hawkins will succeed in his quest, but I sure wish him the best of luck - if only to shut up the naysayers with their precious math.

    Now, supposing he does succeed, I have one question. What will the first AI's favourite distraction be ?

    1. Anonymous Coward
      Anonymous Coward

      Re: "If his bet is wrong, then Hawkins will have wasted his life"

      > Personally, I fail to understand how anyone can hope to build an artificial brain without understanding how a real one works.

      Indeed, the oft used comparison with the aeroplane is apt here.

      We don't build giant birds but we needed to understand the principles of bird flight in order to build aeroplanes and you can only really do this by studying birds.

  11. Anonymous Coward
    Anonymous Coward

    I have spent the whole daily elReg reading quota on this six-page article (I won't read about "Selfied are OVER...", "Boss at 'Microsoft' tech sca", "Tokyo to TXT wanin..." and "Google confirms Turk...", sorry dear tabs). Now please tell me anyone: did you also watch the 25-minute long video on page 5?

  12. Neil Stansbury

    Pattern matching

    I suspect the brain will turn out to be much simpler than everyone imagines, and that the challenge wont be in modelling the brain but in matching its parallelism.

    If the brain is basically one giant pattern matching machine, and If every neuron has around 7000 synaptic connections to other neurons, just changing the order that the synapses fire on one neuron would be enough to encode vast amounts of pattern information throughout the brain.

    If the hippocampus acts as the adaptable fast changing memory, then it becomes like the 5th wheel on an Enigma machine - multiplying all the potential slower learned encoding possibilities even more so.

    Modelling 10 of billions of neurons with their 1000s of re-ordable synapses, might be doable. Modelling how 100's of millions of them fire simultaneously every second is a whole different ball game that clever software alone won't solve.

    1. John Sanders
      Thumb Up

      Re: Pattern matching

      Agreed.

  13. Bruce Woolman

    Companies should pay an earning tax for every robot. This tax should go into a fund to retrain people in jobs that only people can do. Mainly the arts and athletics. So in the future the robots will do all the work and the people will eat delicious robot-made tucker and put on plays for each other and write poetry to each other. We will all be good at sports and belong to many leagues. Everyone will know the Karma Sutra forwards and backwards... especially backwards. We will paint and declaim and write art reviews. Indeed the leaders of society will be the reviewers. It will be a world of poetry and abstract art and street theater. In short. The future will be an artsy fartsy LIVING HELL. Except for the Karma Sutra part, of course.

  14. Anonymous Coward
    Anonymous Coward

    Wear and Tear.

    Machines wear out at varying rates depending on how they are used. For instance, an automobile driven on nothing but highways at moderat speeds will last for at least two decades, but an off-road vehicle is probably going to be used up inside the first five years of use. Electrical components fail as well as a function of hours of use and the currents handled. We will never be able to construct an indestructible automoton capable of independent thought. We will, I think, eventually be able to construct machies with useful lifespans that are much shorter than our lives. Machines self-destruct because of a series of small positive feedback loops. A bad bearing increases its rate of wear over time, for instance, but human bodies constantly repair themselves and there are only certain tissues that must be repaired by surgery once they are damaged.

    What this means is that our creations will be more mortal than we are. They might well go out on strike for better wages, just so that they can afford a complete overhaul after twenty years. I seriously doubt that they will simply allow us to shut them down and throw them on the scrap heap for recycling the way we do other machines. We already buy insurance to repair our automobiles. Expect to be paying for insurance on your friendly family robots.

Page:

This topic is closed for new posts.

Biting the hand that feeds IT © 1998–2019