Three interesting articles on science.slashdot today

Post by **MarauderIIC** » Tue Feb 01, 2005 9:14 pm

DARPA Contracts For AI Technology
Tuesday February 01, @09:14PM
heptapod writes "USA Today is reporting that DARPA has contracted two professors from RPI to develop artificial intelligences that can learn by reading and understanding natural language. Interesting taking DARPA's Grand Challenge into account. Mentioned in the article is Cycorp, Inc. which has been pursuing this goal since 1994!"
Read More...

HP's Crossbar Latch... Next-Gen Transistor?
Tuesday February 01, @01:38PM
moojin writes "CNN.com reports that "in a paper published in Tuesday's Journal of Applied Physics, HP said three members of its Quantum Science Research group propose and demonstrate a "crossbar latch," which provides the signal restoration and inversion required for general computing without the need for transistors.""
Read More...

Solar Super-Sail Could Reach Mars in a Month
Tuesday February 01, @11:00AM
ti-coune sent us a story running on newscientist describing solar super sails and how they could one day get us to Mars in a month. The key is a special new paint. The cast of Trading Spaces is unavailable for comment.
Read More...

Post by **JS Lemming** » Wed Feb 02, 2005 7:29 am

Very interesting, especially the transistors deal.

Post by **Falco Girgis** » Thu Feb 03, 2005 8:00 am

The first one interests me most. AI that can read and learn from English. That'd be pretty cool, but is this really realistic? Do you guys think it's possible? I mean it is possible, but you think it'll ever happen?

Post by **JS Lemming** » Thu Feb 03, 2005 8:11 am

I think its very possible. But I think people are going for it the wrong way. I believe that to successfully create a "learning" program, it must start from scratch instead of being programmed to know english from the start. I think they should first create a pretty much brain dead thing, place it in an artificial environment (because that is how we learn) and give it senses. With the ability to compare senses. It would also need a will to survive and curiosity.

Remember ALICE? She really wasn't learning or thinking. She was programmed to just work with text. What we need is something that essentailly programs itself at an insanely low level.

Post by **MarauderIIC** » Thu Feb 03, 2005 3:23 pm

Seems to me that the problem is in making it learn. Computers either learn everything, nothing, or things they are programmed to. So if its programmed for everything -- what about bad things, or irrelevant information? Things they are programmed to -- what if new useful info comes along?

Anyway, I could probably spare more thought for this post and maybe do some actual looking up of stuff if I wasn't about to run to work.

Post by **JS Lemming** » Thu Feb 03, 2005 4:33 pm

You can't make things learn, thats were curiosity comes in. I've read some stuff on these kinds of topics and have found some interesting ideas. In order for the program to determine if someting is boring or not, it could compare paterns of inputed data, if the patern is similar many times, then the data is "boring" which is bad. Something is considered interesting if the data is only similar, the patern that is.

Post by **Falco Girgis** » Thu Feb 03, 2005 5:54 pm

WTF. Yeah, and by doing that you're making it learn.

You can make things learn.

Orgodemirk · Post by **Orgodemirk** » Thu Feb 03, 2005 7:31 pm

they should just get two humans and have them teach a robot about everything and then have that robot learn how to built robots and before you know it we're all dead.

Post by **JS Lemming** » Thu Feb 03, 2005 8:22 pm

GyroVorbis wrote:WTF. Yeah, and by doing that you're making it learn.

You can make things learn.

I don't think so. How would you force a computer ro learn, punish it if it doesn't? It wouldn't care. It wouldn't have a need too. You can force it to acquire information, but I don't believe it goes the same way with learning.

Post by **MarauderIIC** » Thu Feb 03, 2005 9:01 pm

You can't make it learn, but you have to tell it how to learn with programming -- which also teling it how to figure out how to use the information its learned.

Teach a computer math... but how do you make it draw connections?

Post by **Falco Girgis** » Thu Feb 03, 2005 9:32 pm

MarauderIIC wrote:You can't make it learn, but you have to tell it how to learn with programming -- which also teling it how to figure out how to use the information its learned.

Teach a computer math... but how do you make it draw connections?

You program it to (somehow, not really saying it's completely possible). You program it to use information. You program it to acquire information.

How is that any different than us? We learn and use information as we were created. We use our brains. How do we make the connections? We were created/formed with that ability.

How would you force a computer ro learn, punish it if it doesn't? It wouldn't care. It wouldn't have a need too. You can force it to acquire information, but I don't believe it goes the same way with learning.

Are we punished if we don't learn? WTF are you talking about there? You program it to accept information and use it. When you say "it wouldn't care" I have no freaking clue what you're talking about. Are you trying to act as if we're talking about a human here?

How does it not go the same way with learning? How do you think humans learn? We analyze information and do stuff with it. If you think about it, humans could be programmed just like machines.

Post by **JS Lemming** » Fri Feb 04, 2005 7:35 am

The whole point of it to learn is to eventually break away from programming the "LearnNow()" function in which you tell it how to learn.

Post by **Falco Girgis** » Fri Feb 04, 2005 7:59 am

Oh really? Since when did you decide that that's the point of the project?

Humans could all still be using "LearnNow()" functions. The only basis that you'll disagree with me on is because of your religous beliefs.

Give me a counterexample that proves that humans can't be manipulated by "LearnNow()" functions.

We find information. We take it in and then we apply it. I'm not saying any of this is realistic, but it's certainly not "impossible" and yes, a machine can learn.

Post by **Falco Girgis** » Fri Feb 04, 2005 7:12 pm

Actually, it looks to me like JSL and I agree.

The thing we disagree on is the definition of "to learn".

I think that that would be considered learning. JSL doesn't. We aren't actually disagreeing on anything else.