• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Artificial Intelligence

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

Foxie3a

Normal Member
Joined
Sep 7, 2003
I can't remember ever seeing this brought up here, and it is relating to computers... so I thought that I'd see what you guys know about it.

I am doing a lot of learning in the AI feild, and its pretty hard learning. I am a member over in the AI forums... (ai-forum.org) and its pretty hard to keep up with some people there.

I love the idea of creating something intelligent. Although to do so you must study so many different subjects over such a long period of time. I'm just starting... :-/

So I was just wondering if anyone else here was into it like I am, and what you guys thought about it.

I'm starting to go hardcore with this.. Going to start learning multiple programming languages, and go deeper into the brain and the way that people think. Sorry guys, but overclocking just isn't my favorite thing anymore. ;)
 
What is the most advanced AI out right now? Do they have any PCs that can fool people into thinking they are real people over the net or anything like that yet?
 
Yes, of course. There is a test named after its creator called the "Turing" Test. And its purpose is to do just that - to see if a person can tell whether or not it is a person or machine.

There are bots out there for AIM that you can talk to, and could be a human, except for the exceptionally fast typing skills. ;)

There are some popular AI's like HAL, and Alan, I dont really know too much about either though. You can download HAL(sorry, not link), and it can help you with your schedule, even has an AIM server which is a lot of fun to play with. I once had it talking to my friend, unfortunately my friend didn't know it wasn't me and thought I was being a jerk...I guess it passes the Turing Test.

What's limiting AI today is our little knowledge of how WE think. Because we dont fully understand how we think, and how our brains work(look up neural networks) we can't implement that into our code.

I did a little research on neural networks before and it isn't like I thought biology would be. The Neurons are our transistors, just like a processor, sort of speak. I highly recommend looking it up, its fascinating.

Once we understand how we think, and can express that in code we will have much more intelligent, intelligence.

I personally think that we need to focus on the AI's thinking ability. Once it can think to itself, even slowly thinking, it can better itself, or maybe even TEACH another bot what it knows. The sharing of information in the electronic age...

A nice way to finish my night. Time for bed. :)
 
Hopefully these new machines that think for themselves don't start thinking that Humans should die or something crazy like that. (Yes I watched the Terminator/Matrix/Dr.Who too many times...LOL)

I know machines will be very very far from thinking on thier own that way. However, I bet that once the machines learn to teach and better themselves then thier intelligence will grow at a exponetial rate.
 
I once had to do a programming project when I was a freshman in high school, where I had to write an AI that could play tic-tac-toe. It was freaking hard, but it worked eventually (and it was even a real simple AI that only gathered a database of knowledge, checked out what you were doing, and determined if it had ever seen anything like what you are doing). Like you said, the real problem IMO is not the programmer, but the lack of knowledge of how learning and intelligence actually works.

Good luck learning the languages and with AI. It's an extremely interesting field of research, and likely to eventually gain some more "real world" use :)

JigPu
 
Foxie3a said:
I can't remember ever seeing this brought up here, and it is relating to computers... so I thought that I'd see what you guys know about it.

Actually, we did do this but before you joined. Back in the old days, we had a forum just for this type of discussion. It had to go away for a while because we lost control of it during the run-up to the war. Thanks to Nihili, the best threads from the debates forum are still around (after it was locked he spent a few days doing nothing but saving the stuff so that we would still have it).

This thread has been a real walk down memory lane to reread. There are a lot of absolutely stellar posts there.
 
just wondering...

I'm on AIM and MSN, isn't there a bot i can IM, and have a chat with?

if so, whats its Screen name? i want to test one of these things out just for fun.

its 2am here, and i'm bored!
 
Don't know if it's been mentioned to you yet, but you should also look into 'Fuzzy Logic' as that is how some researchers are getting their AIs to learn.

Good luck

DWolf:cool:
 
I havn't come across fuzzy logic, but I have heard of it.

AIM: "Smarter Child"
 
I always felt that there could never be a true AI because if we program it, it would still be predictable and could never be genuine or compuilsive i always thought that it would have to create itself which would be impossible i guess its really interesting to ponder though
 
Another thing that I think is interesting, are the computer programs where you says "computer create some thing that uses a,b, and c and completes functions x, y and z. On NOVA (I think) i saw a thing where they said "ok, computer make a structure out of legos that can hold 1 kg. and use the fewest pieces possible. and it did that, IT was awesome!!
 
That's right, its not actual intelligence.

Something that I just thought about...

We can't create a program that has the ability to fully learn, and there is the debate of whether or not we can...

Lets just say that we could. If we could, then couldn't we pretedict what it would do, since we could follow the code, and behave in our minds like the code might. Is being able to predict everything something does take away the computer's intelligence?

Oh boy, I need to ask that to my AI buddiies! lol
 
Well, in the other thread I posted a link to an article about a computer program that models electronic circuits. Unfortunately, SciAm no longer allows links to old articles. However, the program has developed a number of patentable circuits and the USPTO is processing the paperwork as fast as the program develops new stuff. While the program itself is the intellectual property of the programmer, Should he get the credit for the work done by the program?

Think about that for a moment. The program is inventing new stuff. That begs an interesting question. By what standard will we judge AI? Will there be a defining moment when someone flips a switch and the first AI is born? Or will it creep up on us as our machine get better? It may be that there is no defining moment of birth for AI.
 
Hey Foxie, your post arrived while I was typing my last one. Let me ask you a question: Who says that we can follow everything that a potential AI can do?

The reason why we have check codes and error correction in today's programs is due in part to the fact that switches on a processor core sometimes arbitrarily flip over. The main cause of this is impurities in the wafer such as uranium decaying. As the switches get smaller and cores are more densely packed the errors become more common.

Could you ask your AI forum guys if they are concerned about this being a problem in the future? Conversely, could that be something they are counting on as it would add a bit of arbitrary behavior to their creations?
 
Ooh, a thread about AI! Sounds like something I can contribute some interesting information to.

Here's the big difference between computers and humans: humans recieve massive amounts of information, process it, and merge it with other information to formulate the way we think about things. Computers, on the other hand, take the information they are given and process it in a way that is given. Let's look at a simple situation that may shed some light on the fundamental difference between people and computers:

You have two people in a room. One is controlled by a human brain and the other is controlled by some sort of AI. The task of each of these people is to be able to escape that room before the other person. The first person to escape gets fame and fortune. The other person gets a gruesome death. This room is 10ft x 10ft. One wall looks like a regular wall in a house. Three walls are made of brick. One of the walls has a steel door. The ceiling is 30ft high and there is a 2ft x 2ft opening about 25 feet up one of the brick walls.

So now you've got the room description. Lets say there are several lengths of rope and wire of various lengths, several dozen small hand tools such as wrenches and hammers, a lawnmower, a tank of propane fuel, and a spear.

Now lets look at the things the human may do: he may tie the ropes together and use the spear to climb out the window. He may use the hand tools and rope to make a rope ladder. He may use the hammer to break through the regular wall that may be made of simple drywall. He may use the hammer to break through the brick walls. He may use the propane and hammer to bust through the steel door. He may even use the lawnmower or spear to kill the other person and then take his time trying to break out.

Now lets look at what the AI may do:

The AI sees the hammer. It knows the hammer is used to drive in nails. There are no nails so the hammer is useless. It sees the propane but no tools that use propane. It sees the lawnmower but no lawn. It sees the ropes and it sees the window but the ropes are too small and the AI requires a grappling hook to get the rope up to the window. It sees the steel door but it is locked and there are no keys or lockpicks.

Now that I've stated that odd situation, you can probably see where the problem lies with artificial intellilgence: you need a MASSIVE database with a MASSIVE amount of relationships that are all manually entered into the database by a human in order for the AI to react appropriately to any situation. In other words, you need the AI to start out as an infant. You need to teach it that drywall can be easily broken through and that hammers have a wooden handle that can be used as ladder rungs. You need to teach it that, in life and death situations, killing the opponent may be a viable course of action.

***************************

Now that I have my little opinion of where AI needs to go to be effective, I'll let you all know that I have taken a course in artificial intelligence recently. We learned a couple of interesting algorithms such as how a robot (or a bot in a video game) may choose to travel a path to get to a goal quickly.

The most interesting project we had was a real world application to a spam filter. You know all those email accounts you have on yahoo? You know that "this is spam" button? What that button does is it makes some program on yahoos servers analyze several characteristics of the email and then enter into a database that those characteristics are characteristics of a spam email and gives them a spammyness rating. When you recieve an email, yahoos program analyzes all the characteristics, gives it a spammyness rating, and if the rating implies that is is sufficiently spammy, then it goes into your bulk mail folder.

For example: You get an email that has the phrase OMG THIS IS SUCH A GREAT PRODUCT!!!!!!!!!!! The number of capitalized words may make this email have a higher spam rating. The use of several exclaimation points may make it spammy. The word "product" may make the email spammy (you know, like the words viagra, v1agra, meds, porn, teens, and mortgage).

Anyways, the goal of that email scanner is simply to get a computer to be able to look at an email and decide as easily as you can that it is spam. It's pretty easy for a human to decide if an email is spam. It's pretty friggin hard to get a computer to make the decision accurately and it tends to require a pretty big database to get it to behave accurately.
 
Krusty, that's sounding a little bit like another problem...

Our brain is made up of a neural network. It allows us to filter things out immiediately.

And yes, DBases take up a LOT of space...And just imagen inputting all that data... That's why people are trying to create an intelligence that has the ability to learn on its own. Maybe if it was an AI with physical properties it could be taught how to read, and you could give it a book.

Just imagen giving your robot a book on Diesel engines, and it reads it for you in a few minutes, and then gives you a nice summary, or maybe it'll go over the entire book, but at your pace. Robots could be extremely useful.

Now we just need to figure out how to code them.
 
Back