Artificial Intelligence

[TheSolitaryOne]

Is it really possible to build a thing which has artificial intelligence and can learn things with time and become better?

[Z3R0]

http://www.youtube.com/watch?v=P9ByGQGiVMg

This is about as close as it gets for now. I'm a little skeptical about it though...who's to say someone wasn't giving it instructions during the video?

[gh0st]

its impossible unless you code something really big but really it would take too much time if you are thinking about a humanoid robot but a simple robot that does simple functions already exists maybe in a future

[Kulverstukas]

I imagine that all those calculations require huge amounts of processing power and probably petabytes of storage if you want the bot to be learning every day and storing the information somewhere.

[Huntondoom]

I imagine that all those calculations require huge amounts of processing power and probably petabytes of storage if you want the bot to be learning every day and storing the information somewhere.

jup, but dont forget how much our brain is doing, cause with everything you see, it looks how it build, what shape, distance to one other etc etc etc,
and it also does that with memory, smell, feel, and thoughts
but that is the advanced our brain does multi-thread where each thread is done by a small group or/and a single brain cell

[ElectricNoodle]

Im reading a book at the moment, by Premier Press, called:  AI Techniques for Game Programming,

If you're interested in AI then read it!! He talks about the standard types of "AI" that use hundereds of case statements... then goes on to talk about how ai is changing and that its no longer acceptable for enemies/ ai players in a game to act stupid! It also delves into programming with neural nets... its an all round good read

[Deque]

In my opinion it is not a matter of speed or storage space. Our brain works with only 14 Hz (means a neuron fires about 14 times per second). That is nothing compared to today's processors. The strengh of the brain is in its massively parallel computing (10^10 neurons, each one has 100000 to 200000 connections to other neurons), which means it can process a lot of data at once. Also the brain's associative storage - most of the storage a computer uses is address based. A computer is usually very exact, but fault intolerant, so that it is bad with problems that have faulty or incomplete input. A brain is just the other way around: inexact and very fault tolerant. We still need more algorithms or concepts that work well with inexact data.

But to answer the question:

Is it really possible to build a thing which has artificial intelligence and can learn things with time and become better?

Yes, i.e. artificial neural networks can do that or evolutionary algorithms. But usually those are only good at one thing. To build an AI that is good with several tasks (like walking and talking and face recognition by one AI) is pretty difficult today.

[Axon]

You are talking about Strong AI

https://en.wikipedia.org/wiki/Strong_AI

It could be possible in the near or distant future ,it has a bright side and a dark side. Strong AI machines can exceed human intelligence thus making humans inferior to them,you don't want to be a slave to a robot ,do you ?

[ande]

Just had a bit of Déjà vu here.

This topic have been up before. Where I replied that an AI does not necessarily have to be a huge and complex project. Maybe the other way around. It might help to look into the human brain, its not as complex as it seams. Its really just a bunch of the same type of cell, doing the same thing. Therefore, maybe one could adapt that to a computer system. Making it learn like a human, little by little. etc etc.

Search for the topic(s). gh0st was the one posting them earlier. I posted some useful PDF's and videos there to. If you cant find it, look into Jeff Hawkins book and website called On Intelligence and numenta.com.

[Jboeggs]

http://cleverbot.com/


^ Its pretty cool (Just don't tell it that or it gets happy)
see what I did there

[techb]

I've been wanting to write a chat bot for about a month now. Been reading up on A.I. myself. The biggest issue we have with ai right now is having a hot actually understand what's going on. Sure, we can simulate understanding, but in the end it's just redundant instructions being processed.

There is a guy I saw on TED not too long ago that is trying get computers to understand, and is making great progress with wolframalpha.com, I don't have a link to the talk, and don't wanna go look for it since I'm on my phone. If you search wolframalpha I'm sure you can find it.

I also seen a talk where they developed a simulated brain, with all the neurons and how it processes things. They fed it an image of a rose, and creeply it started forming inside the brain as neurons processing what it was seeing. This is a big leap in technology and science, we might be able to reverse engineer the brain and cure all diseases. With the same technology, we can develop true advanced ai.

But for now, we can't even simulate human speech well enough to fool peoples. We have come close in turing tests, but no one has been 100% fooled. In the bot I'm designing I'm going to have it learn by listening to human conversion in an irc channel. I'll also have "training" sessions to give it correct responses. I might have it crawl google, wiki, and wolframalpha to build a response based on factual data. That way maybe it can develop a sense of understanding.

Natural language processing mixed with neural nets could yield some interesting results too.