Sorry your browser is not supported!

You are using an outdated browser that does not support modern web technologies, in order to use this site please update to a new browser.

Browsers supported include Chrome, FireFox, Safari, Opera, Internet Explorer 10+ or Microsoft Edge.

Geek Culture / Creating a "good" AI

Author
Message
Three Score
20
Years of Service
User Offline
Joined: 18th Jun 2004
Location: behind you
Posted: 21st Apr 2007 07:39
By human-like, I mean it actually acts like a person would...

but yea, The problem I think with making a real AI is that once you made it to where it could "learn" it would take as long as it would take a human, to get it to act like one...

I mean, with a "smart" learning system though. This way, when it does something you can tell it something simple like "yes" or "no" without having to say something like "yes, unless it is 3:00..then no"
something that just can learn

Also, an assistant for this would be something like a slap on the hand for "bad" and giving it food for "good"

Really, this is too big of a project for me to even design, much less develop..

but anyway..
anyone think this is a good idea?

Open86 --My Emulator (now with it's first super alpha release
I'm addicted to placebo's...I would quit but it wouldn't mean anything! lol
bitJericho
22
Years of Service
User Offline
Joined: 9th Oct 2002
Location: United States
Posted: 21st Apr 2007 07:45
Start wayyyyy smaller

But heck ya, ai is always fun.

Osiris
20
Years of Service
User Offline
Joined: 6th Aug 2004
Location: Robbinsdale, MN
Posted: 21st Apr 2007 07:55
Here is an open source "Intelligence program"

Read this, its very interesting

Your signature has been erased by a mod because it's larger than 600x120....
Zotoaster
19
Years of Service
User Offline
Joined: 20th Dec 2004
Location: Scotland
Posted: 21st Apr 2007 13:45
AI that actually thinks is probably not that far in the future, atleast at the rate technology is progressing just now. Let's face it, anyone with programming experience can make simple learning AI. And through the use of Artificial Neural Networks, the concept of human-like AI is starting to look less and less like science-fiction.

Already people have made robots that literarly have the intelligence of insects, and those are small and cheap to built, but they provide the ideas that we have to follow to make something alot bigger and better.

I think what people should be aiming for is for a robot to do something that literarly hasn't been programmed in, and I'm not talking about just replying, but for example, rolling around and having fun, without being told to, or programmed to.

One of the hardest elements is getting it to recognise things. You might think that a robot could recognise an apple, because if it looked at an apple previously and you told it it was an apple, it would remember the shape, so when it sees an apple again, it can say "That's an apple". With currect technology this is possible, but what if you took a bite out of the apple. Would it still recognise it as an apple, but with a bit missing?

Three Score
20
Years of Service
User Offline
Joined: 18th Jun 2004
Location: behind you
Posted: 21st Apr 2007 16:07
The real question would be does a baby(really young) recognize it as an apple with a bit missing? We really can't be sure as to "how" we ourselves develop...

The problem I have is how to store learned data..
my best idea for recognizing objects would to use a combination of vector and bitmap pictures..like a bitmap that is just the color, and a bitmap that has the texture and it is filled with those.. but really, I'm not sure that will work...

btw anyone know where I can at least learn the basics of AI, as I'm unfamiliar with even the terminology(neural networks?)

Open86 --My Emulator (now with it's first super alpha release
I'm addicted to placebo's...I would quit but it wouldn't mean anything! lol
Zotoaster
19
Years of Service
User Offline
Joined: 20th Dec 2004
Location: Scotland
Posted: 21st Apr 2007 16:14
Another problem with recognising objects is with different shadows and stuff.

Neural networks are a simple simulation of a human brain. You can train them by adjusting "weights", so they will usually (about 90% of the time) come out with the correct answer.

Osiris
20
Years of Service
User Offline
Joined: 6th Aug 2004
Location: Robbinsdale, MN
Posted: 21st Apr 2007 19:56
Well you could always store an image like a dog, but then have the AI break it up into claws fur and such, and when it tries to learn lets say a cat then well cats have fur and claws, only a different shape, it it would take less and less time to learn things, also 3D images would be better to learn off of for an AI as well, because think about it, babies and little kids don't learn by looking at one side of things, they pick it up and look at all sides, and if its big like a chair, they crawl or walk around it until they have seen all sides, because well the right side of a dog may look like the right side thats not always the case, and the front and back don't look like the sides, to it one were to let the AI see it all it would have a much better chance at learning, and sooner or later it would be learning very fast using already established patterns to recognize new ones, like a face, we all have eyes and ears and a nose and mouth, but they all look different, so the best way to go about it would be to break items up into groups such as that.

Well thats my two cents anyways.

Your signature has been erased by a mod because it's larger than 600x120....
Three Score
20
Years of Service
User Offline
Joined: 18th Jun 2004
Location: behind you
Posted: 21st Apr 2007 20:39 Edited at: 21st Apr 2007 20:45
A genius idea, but how would anyone go about storing the 3-d object.. best might be to have it look like a cube of the pictures of the sides..

really I am thinking that the AI being blind might be easier to implement..

The easiest stimuli to emulate could be something like
text, pain/joy(for punishment and reward), but then what? you have to have something for the AI to text about...

but really, I could even get started with just those 2 stimuli above and make something kinda cool..

Open86 --My Emulator (now with it's first super alpha release
I'm addicted to placebo's...I would quit but it wouldn't mean anything! lol
Osiris
20
Years of Service
User Offline
Joined: 6th Aug 2004
Location: Robbinsdale, MN
Posted: 21st Apr 2007 21:07
Well think about how we as humans do it, we don't think about something as 3D all the time, we more or less think about it in 2D but we know its part of the same thing, like say a cube, we know it has six sides, but we cant see them all at once, only two or three sides at once, so its best to store it like this:

Side one=A
Side two=B
Side three=C
Side four=D
Side Five=E
Side six=F

And store them so it knows well those are all different (Cube=A through F), but they are part of the same object. and then it could maybe start to imagine things in a 3D way.

Your signature has been erased by a mod because it's larger than 600x120....
Peter H
20
Years of Service
User Offline
Joined: 20th Feb 2004
Location: Witness Protection Program
Posted: 21st Apr 2007 22:08 Edited at: 21st Apr 2007 22:08
Quote: "I think what people should be aiming for is for a robot to do something that literarly hasn't been programmed in, and I'm not talking about just replying, but for example, rolling around and having fun, without being told to, or programmed to."

programming as we know it would have to change... (as in NOT procedural which is really all we have now disguised in one form or another)

One man, one lawnmower, plenty of angry groundhogs.
Jeku
Moderator
21
Years of Service
User Offline
Joined: 4th Jul 2003
Location: Vancouver, British Columbia, Canada
Posted: 21st Apr 2007 22:40 Edited at: 21st Apr 2007 22:41
I think AI has been moving in the wrong direction historically. For example the IBM chess playing software has millions of games in its memory and can calculate the best possible move for every situation. But then a human being has to accumulate the list of moves and store it in the AI's memory.

I read a really good book called Blondie where two guys made a bot that knew absolutely nothing about checkers. But they played the game with the bot many, many times and eventually the bot "learned" how to play the game, much like a human would. Finally it ended up competing in championship games and doing far better than the developers thought it would. *This* is the right direction for AI in my point of view.

I think game developers realize this as only a decade ago game AI was basically "run toward player and fire weapon". But now we have games like Black & White where the AI tries things out to get results.

I read a fascinating article in Wired about 5 years ago about how a creature in Black & White was upset and the way it felt better was to heal people... yet, there were no sick people to heal. So the creature picked up a human and threw him up in the air to injure him, so that he could heal him afterward. The developers did *not* put this behaviour into the game, and were surprised to see it themselves.

Three Score
20
Years of Service
User Offline
Joined: 18th Jun 2004
Location: behind you
Posted: 21st Apr 2007 22:57
Quote: "I read a fascinating article in Wired about 5 years ago about how a creature in Black & White was upset and the way it felt better was to heal people... yet, there were no sick people to heal. So the creature picked up a human and threw him up in the air to injure him, so that he could heal him afterward. The developers did *not* put this behaviour into the game, and were surprised to see it themselves."


That is amazing! A bad behavior, but still, very neat!

Open86 --My Emulator (now with it's first super alpha release
I'm addicted to placebo's...I would quit but it wouldn't mean anything! lol
Hobgoblin Lord
19
Years of Service
User Offline
Joined: 29th Oct 2005
Location: Fall River, MA USA
Posted: 22nd Apr 2007 10:23
Well to make it act more human like you need to teach the AI to cheat, pout and whine about how you got lucky rolls or how they were so close to beating you but something went wrong. Having an AI chess player get mad and toss the board over would be neat.

Zotoaster
19
Years of Service
User Offline
Joined: 20th Dec 2004
Location: Scotland
Posted: 22nd Apr 2007 15:44 Edited at: 22nd Apr 2007 15:45
Jeku, I totally agree with you there. As I said, unexpected results is what should be expected in AI. Also, the way people always focus on a bot in limited domains, and perhaps merging them all together to make a human-like robot.

When we were doing AI in school, we had to learn about a game where you have to move matched from one pile to another till you got 8 matches in each pile, and there were some rules to follow. When my teacher told us how it would work, I didn't think it should be the way AI should be worked on, because all this was, was a simple algorithm, and the way it solved it was totally different to how any living creature would.

As I said, Neural Networks FTW!


[edit]

Another interesting story about Black & White was of an animal who got really hungry and there was no food around, so it tried to eat it's foot, and that wasn't programmed in.

Three Score
20
Years of Service
User Offline
Joined: 18th Jun 2004
Location: behind you
Posted: 22nd Apr 2007 16:54
lol..I have never even heard of black and white..

Now what if we taught it to speak in 1337! that'd just be horrible, or if they learned it on their own!

Open86 --My Emulator (now with it's first super alpha release
I'm addicted to placebo's...I would quit but it wouldn't mean anything! lol
Osiris
20
Years of Service
User Offline
Joined: 6th Aug 2004
Location: Robbinsdale, MN
Posted: 22nd Apr 2007 18:10
Funny ones from TES: Oblivion.

1. One character was given a rake and the goal "rake leaves"; another was given a broom and the goal "sweep paths," and this worked smoothly. Then they swapped the items, so that the raker was given a broom and the sweeper was given the rake. In the end, one of them killed the other so he could get the proper item.

2. Another test had an on-duty NPC guard become hungry. The guard went into the forest to hunt for food. The other guards also left to arrest the truant guard, leaving the town unprotected. The villager NPCs then looted all of the shops, due to the lack of law enforcement.

3. In another test a minotaur was given a task of protecting a unicorn. However, the minotaur repeatedly tried to kill the unicorn because he was set to be an aggressive creature.

4. In one Dark Brotherhood quest, the player can meet up with a shady merchant who sells skooma, an in-game drug. During testing, the NPC would be dead when the player got to him. The reason was that NPCs from the local skooma den were trying to get their fix, didn't have any money, and so were killing the merchant to get it.

5. While testing to confirm that the physics models for a magical item known as the "Skull of Corruption," which creates an evil copy of the character/monster it is used on, were working properly, a tester dropped the item on the ground. An NPC immediately picked it up and used it on the player character, creating a copy of him that proceeded to kill every NPC in sight.

Your signature has been erased by a mod because it's larger than 600x120....
Three Score
20
Years of Service
User Offline
Joined: 18th Jun 2004
Location: behind you
Posted: 22nd Apr 2007 19:36
Am actually building a bit of a prototype now..
It's pretty fun making something think..

I'm doing it all in C
If I ever get anywhere with it, then I might make a SF project page or something..

all I'm trying to get it to do is ask for food or comfort, and to ask not to be slapped..but all without explicitly programming it in

really quite simple..I have no idea what I will do when I start storing it's memory on a file though..(right now it's just a linked list of memory)

Yea I know it will probably never get anywhere, but it's fun so, I'll work on it till I'm bored..

Open86 --My Emulator (now with it's first super alpha release
I'm addicted to placebo's...I would quit but it wouldn't mean anything! lol
Zotoaster
19
Years of Service
User Offline
Joined: 20th Dec 2004
Location: Scotland
Posted: 22nd Apr 2007 20:37 Edited at: 22nd Apr 2007 20:44
I read something up on the Black & White AI. The idea (and it sounds relativelly simple), is that the creature has desires, and an intensity for each desire, so it choses the desire with the highest intensity and goes with that. For each desire, you get some sub-data that goes with it, for example, if the creature is hungry it will scan all objects around it, so there might be a rock, and a sheep. It can try and eat the rock, and decide that the taste for that is not so good, so the rock's intensity goes down. It might then try and each a sheep, and find it tatses really good. So over time, the desires become more and more refined, so it wouldn't just be "I feel like eating", it would be "I feel like eating a sheep", As time goes by, desires are added to a list, and the cycle begins all over again, so it already knows sheep are good to eat, and rocks arent, so it might try and eat a normal sheep, then a fat one, and the fat one tastes better, so, eating a fat sheep is added to it's desires.

The intensities of each desire is not just affected by taste or whatever, but how the player treats it when performing an action. Other ways are how the creature watches the player and tries to copy it, so obviously the creatures have a data base where knowledge can be added to.

What I'm guessing about the creature that hurt a guy just to heal him again was, that the creature somehow learned that big impulse = pain. What it wanted to do was to heal someone, but someone had to be in pain. So it appears, that desires can link to each other somehow. That bit, sounds hard

Three Score
20
Years of Service
User Offline
Joined: 18th Jun 2004
Location: behind you
Posted: 23rd Apr 2007 01:28
Probably the hardest thing is to teach it to do math, like adding and subtracting
Anyone know of AI that is good enough to know math?

Open86 --My Emulator (now with it's first super alpha release
I'm addicted to placebo's...I would quit but it wouldn't mean anything! lol
Venge
18
Years of Service
User Offline
Joined: 13th Sep 2006
Location: Iowa
Posted: 23rd Apr 2007 01:55
You should talk to the Alice bot online.
A bit strange, but the good version costs money.

Gil Galvanti
19
Years of Service
User Offline
Joined: 22nd Dec 2004
Location: Texas, United States
Posted: 23rd Apr 2007 06:43
Quote: "Probably the hardest thing is to teach it to do math, like adding and subtracting
Anyone know of AI that is good enough to know math?
"

That seems like it would actually be fairly easy, considering math is built into a programming engine, so it wouldn't really require any simulation. It depends on what you would want it to do, but if your talking about where if the player asks "what is 1+1?" then it would just read and analyze the different parts of the sentence. It would see the plus sign with two numbers on each side, signifying it to add them.


dab
20
Years of Service
User Offline
Joined: 22nd Sep 2004
Location: Your Temp Folder!
Posted: 23rd Apr 2007 09:48
Quote: "You should talk to the Alice bot online.
A bit strange, but the good version costs money."


I put the free version into a Windows Live Messenger Plugin. Its great! Some people think I'm talking to them. The only problem is I can't save the "brain". It restarts as msn restarts.

The dude guy
20
Years of Service
User Offline
Joined: 3rd Aug 2004
Location: In the streets of sasatuin
Posted: 23rd Apr 2007 14:11
Quote: "I read a really good book called Blondie where two guys made a bot that knew absolutely nothing about checkers. But they played the game with the bot many, many times and eventually the bot "learned" how to play the game, much like a human would. Finally it ended up competing in championship games and doing far better than the developers thought it would. *This* is the right direction for AI in my point of view."


Oh no the robot is beating us in checkers!

The Alice thing gave me about 10 minutes of entertainment, but then I got bored, what's the entire point of it again?
GatorHex
19
Years of Service
User Offline
Joined: 5th Apr 2005
Location: Gunchester, UK
Posted: 23rd Apr 2007 15:01 Edited at: 23rd Apr 2007 15:04
AI will never make computers intelligent.

It's all based on statistical weight which means it will never expect the unexpected.

To see how simple AI is look up the Tic-Tac-Toe (Naughts and crosses) game that can be played with nothing more than 9 match boxes!

They based the film War Games on it, where an AI computer decides to destroy the world with nuclear weapons until a teen hacker teaches it that like Tic-Tac-Toe the best you can achive is a draw



http://www.KumKie.com http://bulldog.servegame.com
Dazzag
22
Years of Service
User Offline
Joined: 26th Aug 2002
Location: Cyprus
Posted: 23rd Apr 2007 15:04 Edited at: 23rd Apr 2007 15:05
Anyone still have that link to that online "game" where you think of an object and the computer asks you questions? If it gets it within something like 6 guesses then it wins. So you get something like:-

Is it an animal?
No.
Is it plastic?
Yes.
Is it bigger than your head?
Yes.
Is it edible?
No.
Is it a giant phone.
Yes.

Ok, so not quite with the phone, but it was amazingly uncanny. Even if you tried to throw it off track a little it still got there. Think it took a hell of a lot of people using it. Still fun though. We threw everything at it in work (good afternoon that one), and it got like 99%. Even when you tried to be rude it knew you were being rude and told you off. Tops.

So anyway, someone find it (I'm working now ) as it is loads of fun.

Cheers

I am 99% probably lying in bed right now... so don't blame me for crappy typing
Current fave quote : "She was like a candle in the wind.... unreliable...."
Zotoaster
19
Years of Service
User Offline
Joined: 20th Dec 2004
Location: Scotland
Posted: 23rd Apr 2007 17:58
That's a relativelly good simulation, but I still think the idea of having such limited domains is a bad one.

GatorHex,

I'm afraid noone can know for sure, but sometime in the future, I think it will be possible. The lightbulb was invented around 100 years ago. Look at us now. Imagine what things'll be like in another 100 years time!

David R
21
Years of Service
User Offline
Joined: 9th Sep 2003
Location: 3.14
Posted: 23rd Apr 2007 21:46
Quote: "The lightbulb was invented around 100 years ago. Look at us now"


Still using pretty much the exact same incandescent form of lightbulb that's only about 30% efficient. Yeah, real progress there :p


Zotoaster
19
Years of Service
User Offline
Joined: 20th Dec 2004
Location: Scotland
Posted: 24th Apr 2007 18:24
I wasn't talking about the lightbulb in particular

If you want a better example, the first calculator was made in 1972... that means people could get to the moon without a calculator. "Look at us now"!

Login to post a reply

Server time is: 2024-11-18 13:49:51
Your offset time is: 2024-11-18 13:49:51