It may interest some people to know that Steve Grand (him of "Creatures") is also attempting to 'raise' an entity. In this case it's a robot primate which he initially tried to bring up at the same rate as his niece (I think). He came to our university recently and gave a talk about it. Unlike the program, it's embodied and situated. He's using his own methodolgy, based (if I remember) loosely on subsumption architecture:
But that's just it - it doesn't - and can't -"know" anything. It has a set of algorithms that manipulate symbolic internal representations, in this case, strings of characters. This just isn't how our brain works, on any level - our brains operate more as a dynamical machine. There's no grounding of symbols in this program, it just relates string of characters together with no meaning behind them.
Would you expect it to be intelligent if it, for example, read an encyclopedia? To make it operate equivilent to an 'older' level, the New Scientist article indicates that you have to give it more and more complicate algorithms. The writer also implied (only in a fun way!) at the end of the article that they considered that it could in some way be alive - no way! (But maybe it was just a jokey way to end)
This stuff is very GOFAI (Good Old-Fashioned A.I.) rather than NFAI (New Fangled A.I.) which deals with ALife and evolutionary/adaptive systems. I'm not saying it isn't going to be useful and fun to use, though!
I urge people to read Brooks paper that related to this, Intelligence Without Reason, great stuff! It highlights other attributes that may be important before we can say if something is intelligent, things animals and even plants have, but this program, and others like it, doesn't.
OK, I accept that his world may be real to him. But he's expected to talk to us about our world - unless he can react to a 'ball' in some way (even on completely his own terms), surely he'll always be manipulating the word "ball", and not actually mean a ball. He'll be faking it! Chinese room anyone?
It is interesting that they are interacting with him on a one-to-one level.
Also, about the N.N. method - at the moment, they're not very biologically plausable anyway. In dynamic recurrent/Boltzmann/even B.P. N.N. models (I guess there're the ones most people mean here) an artificial neuron represents many real neurons, isn't spiking, etc. So, it's not just a matter of numbers of neurons, we also don't understand them so well (but I may be wrong).
Just my thoughts!
Doesn't the program seem a bit limited though as an A.I. though? The program is not embodied or situated in any world, particularly not the same world as an adult, and it has no symbol grounding - it can manipulate terms like 'park' as a word, but it can't actually derive anything from it or interact with the world in any meaningful way.
Slashdot Mirror
http://citeseer.nj.nec.com/41420.html
http://www.cyberlife-research.com/Lucy/index.htm
Would you expect it to be intelligent if it, for example, read an encyclopedia? To make it operate equivilent to an 'older' level, the New Scientist article indicates that you have to give it more and more complicate algorithms. The writer also implied (only in a fun way!) at the end of the article that they considered that it could in some way be alive - no way! (But maybe it was just a jokey way to end)
This stuff is very GOFAI (Good Old-Fashioned A.I.) rather than NFAI (New Fangled A.I.) which deals with ALife and evolutionary/adaptive systems. I'm not saying it isn't going to be useful and fun to use, though!
I urge people to read Brooks paper that related to this, Intelligence Without Reason, great stuff! It highlights other attributes that may be important before we can say if something is intelligent, things animals and even plants have, but this program, and others like it, doesn't.
Er, guys, I think he was joking! You were joking too, right? Just teasing? Weren't you? Guys?
It is interesting that they are interacting with him on a one-to-one level. Also, about the N.N. method - at the moment, they're not very biologically plausable anyway. In dynamic recurrent/Boltzmann/even B.P. N.N. models (I guess there're the ones most people mean here) an artificial neuron represents many real neurons, isn't spiking, etc. So, it's not just a matter of numbers of neurons, we also don't understand them so well (but I may be wrong). Just my thoughts!
Doesn't the program seem a bit limited though as an A.I. though? The program is not embodied or situated in any world, particularly not the same world as an adult, and it has no symbol grounding - it can manipulate terms like 'park' as a word, but it can't actually derive anything from it or interact with the world in any meaningful way.