Not signed in (Sign In)
    • CommentTimeDec 18th 2007
    ....okay! ....

    Secondly, as Sara alludes to, there are many different kinds of AI research out there. The vast majority of it is still classical AI with its heavy roots in algorithmic search. An example of this would be the chess program that searches through every single possible series of moves and responses from the current board in order to look for a way to win. (This is a greatly simplified description. There are many elegant ways in which search algorithms speed things up and determine which branches aren't worth searching through and so on. However, the gist is that you are trying to search through all the possibilities in a methodical way.) This is not how humans play chess. Humans do not have the resources to enumerate all possible moves and all possible moves in response to those moves and so on. Computers kick our asses at this kind of thing now that they have so much processing power. Computers are great at methodically going through the lists of all possibilities. Research into improving the efficiency of how they go through the lists is generally considered to be AI research.

    Even within connectionism (of which artificial neural networks are an example), the AI is at best biologically inspired. I mean, human brains have lots of simple little things (neurons) which can't do that much on their own, but are interconnected in a way that results in massive processing power and a huge range of capabilities. Artificial neural networks have simple little processors that are interconnected in a way that gives much greater processing power than the individual components, but if you get down to anything much more detailed than that it really all falls apart. It's a quite highly abstracted model, and like Sara pointed out they get used for a lot more than just cognitive modelling work. As someone who interprets trained networks, I can also tell you that they don't necessarily organise themselves internally in a way that relates to the current theory of how a human performs the same task.

    Mind you, humans and other animals are really great at a lot of tasks that computers still have a really hard time with. Processing image data, for instance, is something we just trivially unthinkingly do all the time. Computer vision research is a hugely complicated field with amazing work coming out of it but nothing even close yet to what we (and many other animals) can do.

    That link from @hemlock_martini to a model of the rat neocortical column at the cellular level is interesting but I think the last part of the piece does a good job of addressing the rather optimistic time estimate for a model of the entire human brain as well as pointing out that this mapping does not translate into a full understanding of function. It is also worth nothing that it is a model of the neocortical column at the cellular level not the molecular level. It is not until it is expanded to the molecular level that they will be at the phase of being able to do simulation drug testing. This has already been commented on others here too so I don't think I have much I can usefully add.

    @Willow I'll chime in with some others to say I don't see it as a choice between mimicking or making entirely new things. Both kinds of work are being done. Simulation and modelling have their place - part of which is helping us test out current theories of how our brains work. If you build a model based on your theory and it generates different responses than human subjects then you know you need to revise. If responses are the same then you just might be onto something and should try testing it some more to see where it fails.
    At the same time there is a great deal of research into intelligent ways of, say, searching through large sets of data that aren't human-like and don't try to be. It's still arguably intelligent when you have some sophisticated algorithm that no longer even resembles brute force. Or it is to some ... intelligence itself is a badly defined term as @turing mentions. We even have a lot of trouble deciding what counts as human intelligence. Just consider all the failings of standardized testing and the intentions of measurement behind them. So! I agree with you that it is important to look into other forms of intelligence, but I think there's room for investigating the human kind as well.

    Of course, this is a totally stale thread so it's possible you all hate me for resurrecting it but ummm... look behind you! Robots!
    Beep boop beep!
    • CommentAuthorKosmopolit
    • CommentTimeDec 19th 2007
    Hey, if we don't build in the capacity to feel pain and fear they'll be moving into our neighbourhoods and trying to date our women.
    • CommentAuthorlex
    • CommentTimeDec 21st 2007
    And we'll date theirs, so what ;-)

    Thanks, Vanessa for explaining the different kinds of AI research, I was too lazy to do it.
    Unfortunately (or fortunately, for the frankenbots fanatics), it does not even stop there if we want to have intelligent robots. There are many unsolved problems in different areas to achieve humanoid-aequivalent robots:
    - sensors – do you even know how awesome the human eye is (or cat eye)? It adapts incredibly fast do different lighting environments, of which current cameras are not capable. Same for touch, smell, taste sensors.
    - power – How long can a robot go autonomously without enslaving humankind for batteries?
    - motorics – Oh yeah, that asimo bot can climb stairs, but that's about it. Make it climb El Capitan and I'm impressed.
    All this is not strictly AI, but definitively as challenging, if not more.

    You may read Hans Moravec's 'Mind Children'. It's slightly outdated, but still fascinating and thought-provoking. I believe it is even cited in Transmet at some point...
  1.  (44.44)
    Psi as in like "mental powers?" Wouldn't that just be wireless networking? :-)

    I just wrote that on a post-it for future reference. Because its SEXY...
    • CommentTimeDec 22nd 2007
    Psi as in like "mental powers?" Wouldn't that just be wireless networking? :-)

    I just wrote that on a post-it for future reference. Because its SEXY...

    everything I do is sexy
  2.  (44.46)
    everything I do is sexy

    Even your avatar pic. Except that's in a sexual miscreant kind of way.
    • CommentTimeDec 22nd 2007
    I'm having a time out until I can learn some manners.
    What's possibly more scary is a robot built to <i>enjoy</i> pain.

    <a href="">Poor Elmo!</a>
    • CommentTimeDec 22nd 2007
    that was awesome