Not signed in (Sign In)
    • CommentAuthorbarryhall
    • CommentTimeJan 5th 2010 edited
     (7519.1)
    Professor Henry Markram, a doctor-turned-computer engineer, has announced that his team would create the world's first artificial conscious and intelligent mind by 2018.
    What Markram's project amounts to is an audacious attempt to build a computerised copy of a brain - starting with a rat's brain, then progressing to a human brain - inside one of the world's most powerful computers.
    This, it is hoped, will bring into being a sentient mind that will be able to think, reason, express will, lay down memories and perhaps even experience love, anger, sadness, pain and joy.
    'We will do it by 2018,' says the professor confidently. 'We need a lot of money, but I am getting it. There are few scientists in the world with the resources I have at my disposal.'

    Link Sorry it's the Daily Mail.

    Great quote - "Markram is not mad, but he is certainly unsettling."
  1.  (7519.2)
    Well, I can't see how this could possibly go wrong.
  2.  (7519.3)
    Sheesh, we don't even fully understand how brains work and this guy thinks he can recreate a functioning one by using binary and a map? Without all of the stimuli, nerves, environment, genetic instructions, and everything else the brains needs to serve it's function as a control center?

    Evolution is how you make a brain. That's what they should be trying to mimic virtually. This is putting the cart before the horse and not even bothering with the horse.
    • CommentAuthorFlabyo
    • CommentTimeJan 5th 2010 edited
     (7519.4)
    This is ultimately the final goal of the neural net school of artificial intelligence. The idea that if you can perfectly replicate the structure and mechanism, intelligence just sort of falls out of it. The other school being 'classic' AI, which takes the approach of replicating the way people reason using computer logic, and believing that if you build a complicated enough model, intelligence sort of falls out of it. These two schools of thought don't get on that well, and they both roundly hate what I do for a living (videogame AI, which isn't AI at all but gets called that).

    The stumbling block has always been 'mechanism' rather than structure though. There's just too much weird and wacky stuff involving various neuro-chemicals that we have barely scratched the surface in understanding to make much progress.

    Although we do have a pretty solid idea of how the visual cortex works, and can replicate that to a fair degree (albiet with far lower ability than the human visual cortex).
    • CommentAuthor256
    • CommentTimeJan 6th 2010
     (7519.5)
    @William George - Not sure if it's this same guy, but there is a project underway to simulate the biological structure of a rat brain and one of the stated goals is to help to understand what makes the brain work.

    The essence of the problem, as I understand it, is that even though we are getting to understand what the structure of the brain looks like, the question of how a thing with that structure spontaneously gives rise to the effects that we call "thought" is a practically closed black box. If we can simulate a rat brain perfectly and then the simulation starts to think like a rat then we have something very interesting on our hands.

    On the other hand, if they switch on the perfect simulation of a rat brain and it doesn't, what you have there is a very black day for the science because that would imply that whatever it is that makes the whole thing work is some effect or interaction that we haven't worked out how to see yet.

    i.e. this is a test of our understanding of how the brain works - if it works we're on the right track, at least.
    • CommentAuthorSteadyUP
    • CommentTimeJan 6th 2010
     (7519.6)
    Honestly? Someone else'll beat him to it by then. Either via an evolutionary model or something else.
  3.  (7519.7)
    Didn't the Russians already do this? EDIT: Nope, can't find the article anymore.

    My uncle develops AI with voice recognition.One time a woman called the computer (which could answer the phone line so he could educate it while driving) on a wrong number, and had a brief conversation with it before hanging up convinced she'd talked to another person. I wouldn't say it passes the Turing test, mind you, more of a puzzled "Hi, how are you?" exchange, but still.

    You can hear it here.
    • CommentAuthorSteadyUP
    • CommentTimeJan 7th 2010
     (7519.8)
    Dude.
  4.  (7519.9)
    Fools! They called me mad, but I'll show them all! BWA HA HA HA HA HAHAHAAAAAA!!!!
  5.  (7519.10)
    Years of movies in need of a scary monster-villain have told me this can't possibly end well. Who am I to disbelieve?
  6.  (7519.11)
    Skynet will kill us all.
  7.  (7519.12)
    @Brendan, holy shit. That is amazing. Given time it will become HAL, but hopefully without the murderous and sociopathic tendencies.
    •  
      CommentAuthorGhostboy
    • CommentTimeJan 8th 2010
     (7519.13)
    Skynet, huh? Well, I want to live long enough to see those cool bikes shooting out of a giant robot's shins. Then die happy.
    • CommentAuthorthescruff
    • CommentTimeJan 18th 2010
     (7519.14)
    That's amazing & definitely something to keep an eye on. I can't even imagine the advances in technology we will see in our lifetimes or the struggle to deal with the new.
  8.  (7519.15)
    There are few scientists in the world with the resources I have at my disposal.

    And a chill runs down my spine
    • CommentAuthorKosmopolit
    • CommentTimeJan 19th 2010
     (7519.16)
    As far as the Turing Test goes. I say get two neural nets - train one in folling humans, train the other to discriminate between humans and AI.

    Then hook them up and let them evolve.
    • CommentAuthorPhranky
    • CommentTimeJan 19th 2010 edited
     (7519.17)
    I was having a discussion with a friend about the development of AI and whilst he was skeptical I'm pretty confident in the chances of memristors (newly invented electrical components that behave like human brain cells) being able to give rise to a true artificial intelligence. It uses actual hardware to simulate a brain as opposed to using huge banks of super computers to emulate portions of a rat's brain.
    Recently, a simple electronic circuit[38] consisting of an LC network and a memristor was used to model experiments on adaptive behavior of unicellular organisms.[39] It was shown that the electronic circuit subjected to a train of periodic pulses learns and anticipates the next pulse to come, similarly to the behavior of slime molds Physarum polycephalum subjected to periodic changes of environment.[39] Such a learning circuit may find applications, e.g., in pattern recognition

    First computers will have the intelligence of slime molds and the next...
    •  
      CommentAuthordr_ether
    • CommentTimeJan 19th 2010
     (7519.18)
    This is ultimately the final goal of the neural net school of artificial intelligence. The idea that if you can perfectly replicate the structure and mechanism, intelligence just sort of falls out of it. The other school being 'classic' AI, which takes the approach of replicating the way people reason using computer logic, and believing that if you build a complicated enough model, intelligence sort of falls out of it. These two schools of thought don't get on that well, and they both roundly hate what I do for a living (videogame AI, which isn't AI at all but gets called that).

    The stumbling block has always been 'mechanism' rather than structure though. There's just too much weird and wacky stuff involving various neuro-chemicals that we have barely scratched the surface in understanding to make much progress.


    @Flabyo I utterly agree. The issue is not how brain cell connection is organised, but how information is transmitted and the means. Neural Networks have been a focus of my work for almost 4 years and their attempt to mimic brain cells is ok, good enough for pattern recognition. But for Intelligence to just fall out of a simulated brain. I'm not sure. There could still be many biochemical mechanisms that have not been identified or correctly characterized. There may well be chemical signals that rely on quantum tunneling in order to exchange molecular/atomic/electronic signals, such as the transfer of an electron, the shuttling of protons or the shift in conformation of a protein or small molecule.

    I mean good luck to the man for trying but I do not have too much hope in seeing it succeed. But then every failiure offers new discoveries.
    • CommentAuthorPhranky
    • CommentTimeJan 19th 2010
     (7519.19)
    The problem here is trying to build an AI which behaves like an organic being. It will never be an intelligence that we could see as human because it could never be that. A technological intelligence (without the variety of biochemical impulses that make up a human being) would be something new and quite different.

    We should just build something out of those memristors and other technology and give it the ability to learn and grow to see what happens.
    •  
      CommentAuthordr_ether
    • CommentTimeJan 19th 2010
     (7519.20)
    The problem here is trying to build an AI which behaves like an organic being. It will never be an intelligence that we could see as human because it could never be that.


    So long as the model can represent all the biochemical signals (either explicitly using atomistic simulations) or representative (numerical model representation of the signals) then I see no problem with it being a smulated intelligence. The problem will be that the intelligence can only be human like if it has access to some form of input that is representative of the real world and the 'body' that the intelligence is supposed to inhabit.