Scratch

Bump

jji7skyline wrote:

Might be better to send them an email 

Ok, do you happen to know where I can do that?
Thanks!

EDIT: Figured it out and sent them an email. I hope that works!

Last edited by -GizzardGulp- (2012-01-21 08:36:07)

GLaDOS2 wrote:

I could help you.
I'm an AI myself.
Of course, the price for some knowledge is to finish some tests.

Thanks, I'm glad you want to help. I don't quite understand how you're an A.I. yourself. And what are the "tests"?

Last edited by -GizzardGulp- (2012-01-21 10:33:22)

GLaDOS2 wrote:

-GizzardGulp- wrote:

GLaDOS2 wrote:

I could help you.
I'm an AI myself.
Of course, the price for some knowledge is to finish some tests.

Thanks, I'm glad you want to help. I don't quite understand how you're an A.I. yourself. And what are the "tests"?

The tests were a joke.
You'd have to have played Portal or Portal 2 for this to make sense. 

Ah. What about the part about how your an A.I.? Was that not a joke?

I'm not a member of the Scratch Team, nor an expert in the field, but I thought I'd just talk about it 

What are some of the major theories about how artificial intelligence will function in the future? How does present A.I. function and what is it currently capable of?
I know about game trees, in which the computer looks at all possible outcomes, generally "playing" against an opponent and chooses the best path to reach the best scenario.  There are single-state AIs, in which one variable determines what "state" the AI is in (e.g. "searching", returning home", etc.).  Then there are "neural networks", which are really the most direct attempt at simulating human thought, in which there are a number of "neurons" that are weighted, and by changing those weights, it can learn.  To "teach" these "naturally", you can simply use a "you're right/wrong" feedback.

I want to know the processes of more artificial intelligence technique theories. I also want to find out more about how this “layered path-finding works”. I think you would need some pretty advanced path finding and decision making to get an A.I. to know how to adjust to certain conditions.
I'm not sure exactly what you meant by "layered path-finding", but it may be similar to a "game tree".  Such methods are used when the total possible outcomes are severly limited (e.g. in Chess and Checkers, the average number of moves is 30, as opposed to, say, Halo, in which everyone can move at different speeds in different directions, leading to nearly an infinite possible number of moves per frame).  However, game trees can't usually be used in "real time", but are used much more in turn-based game AI.  On the other hand, I may be totally off, and you may be referencing something I don't know about  .

[color]By doing this research paper, I hope to learn more about an A.I.’s decision making skills. Another thing I’d like to research and discuss in my paper is how current A.I. works. Current A.I. is unadvanced and simply consists of making an object react in certain pre-defined ways to certain changes in its environment. Some would probably not even consider it artificial intelligence. I’d like to learn more about it.[/color]
I'm not sure I'd define it as "unadvanced".  Certainly we are short of - say - human intelligence, but with neural networks we are continuing to advance in that direction.  And as far as limited-scenario situations are concerned, computers are superior (e.g. Chess).  Unfortunately (for AI anyway), life has a nearly unlimited number of scenarios.


Why is this topic/question important to you?
This topic is important to me because I am very interested in technology and logic. I hope to be able to learn more about it so that I can increase my understanding of technology. I think artificial intelligence is perfect for my interests and I will definitely enjoy researching about it.

You'll probably be beyond me soon (if you're not already).  If you're interested in technology and logic, I'd also recommend studying how computers really work, particularly logic gates.  I haven't spent much time on them, but what time I have spent has sparked my interest.

GLaDOS2 wrote:

-GizzardGulp- wrote:

GLaDOS2 wrote:

The tests were a joke.
You'd have to have played Portal or Portal 2 for this to make sense. 

Ah. What about the part about how your an A.I.? Was that not a joke?

That t'was the only true part.
C'mon. An AI has to play around, too - humans can't have all the fun.

Ok then, tell me about yourself and what its like being an A.I.. How do you does your "brain" work?

-GizzardGulp- wrote:

GLaDOS2 wrote:

-GizzardGulp- wrote:

Ah. What about the part about how your an A.I.? Was that not a joke?

That t'was the only true part.
C'mon. An AI has to play around, too - humans can't have all the fun.

Ok then, tell me about yourself and what its like being an A.I.. How do you does your "brain" work?

Don't listen to GLaDOS2;; they're a troll.

veggieman001 wrote:

-GizzardGulp- wrote:

GLaDOS2 wrote:

That t'was the only true part.
C'mon. An AI has to play around, too - humans can't have all the fun.

Ok then, tell me about yourself and what its like being an A.I.. How do you does your "brain" work?

Don't listen to GLaDOS2;; they're a troll.

You never know... maybe it is a real genuine A.I. writing on here right now!

  Joking

veggieman001 wrote:

-GizzardGulp- wrote:

veggieman001 wrote:

Don't listen to GLaDOS2;; they're a troll.

You never know... maybe it is a real genuine A.I. writing on here right now!

  Joking

Nope. I know who it is but they constantly deny it.

*Vaguely remembers person obsessed with Owls and Portal but not the name*

WORK BRAIN WORK GOSHDORNIT!

veggieman001 wrote:

soupoftomato wrote:

veggieman001 wrote:

Nope. I know who it is but they constantly deny it.

*Vaguely remembers person obsessed with Owls and Portal but not the name*

WORK BRAIN WORK GOSHDORNIT!

I believe the person also is Medic, if that helps :3

Owls? Portal?
Where did the owl thing come from? I like turkeys - they're actually really cuddly. My uncle has a farm, and every year we go to his house and help him with the turkeys.
Owls are creepy. .3.

EDIT: I also have to help my brother with AI stuff because that's what he chose to do instead of medicine like I did, I saw AI in the title, so that's why I'm here.

Last edited by Medic (2012-01-21 18:19:06)

I've heard about A* and I'm pretty sure I understand the concept, but I've never actually tried to implement it.  Anyway, chess is the most famous case of computer-vs-human rivalry.  The most famous games were between Deep Blue and Kasparov (arguably the greatest chess player of all time).  Basic matches involve six games (each player gets to play white three times).  Deep Blue won 2 games, Kasparov won 1 game, and there were 3 draws.  Bear in mind though that this was 1996, and computers are naturally even more powerful today.

[offtopic]
Interesting Fact:  I read a news article that said that Russian police raided Kasparov's house (didn't know he was Kasparov), and were surprised to see a computer the size of a mini-van in his house.  Sounds like he's been training 
[/offtopic]

Checkers is also a quasi-good example.  We have "solved" checkers (computers have looked at every possible move from start to finish and determined the "best" moves.  However Checkers Grandmasters are nearly as good, so almost every game ends in a tie.  So computers are about as good as it is possible to get, while humans are (arguably) just as good.

[sidenote]
Computers have also solved Tic-Tac-Toe, which only has 9! possible games, less if you count rotations.
[/sidenote]

Moving down the spectrum, the game Go is an entirely different ball game.  This is because there are hundreds of possible moves per turn.  Computers simply don't have the processing power to handle that many moves using a game tree.  As a result, computers generally can only beat novices at Go (I'm pretty much below novice, so I lose to everybody and everything xD).

Now this is a nice way to transition into game-trees.  Basically, a computer will look at all the moves it can make (each player has 2 possible moves in the picture below).  Then it looks at what moves its opponent can make in response to its moves.  Then it looks at its responses to its opponents moves, then its opponents responses to its moves, etc.

It continues deeper and deeper into the future.  At some point, it stops going deeper (typically because the user doesn't want to wait millions of years for the computer to reach the "best" move).  Then it evaluates the positions at the end (for a basic chess computer, it could simply count how many pieces are left).  Then it assumes that its opponent will always make the "wost" move (the opponent's best move is, to the computer, the "worst" move), and that it will always make the best move.  From that, it chooses which option is most likely to result in the best outcome.  If that didn't make sense... well, that is probably entirely my fault.



In any case, AI doesn't use game-trees for real-time scenarios or scenarios with so many possible "moves" that it is impossible to evaluate all of them.  I'm going to use Halo as an example (but any videogame should work).  Every frame the computer reads your controller, and calculates how fast to go, and whether or not to turn, and how fast to turn, whether to shoot, hit, switch weapons, etc.  Lets say that you have 100 possible options (move forward and switch weapons, move forward and hit, strafe left quickly and do nothing else, etc.).  Now lets say there are 3 other people with 100 options each.  That means that every frame there are 100^4 possible results that could occur.  The computer simply can't calculate a game tree that splits into 100 million branches every frame.  Where you see (B) and (C), there would be 100,000,000 circles, each with 100,000,000 squares below it.  Now add into it that the computer has to calculate this very frequently (say, once a second).  Even with these very conservative estimates, it is simply infeasible.

Instead, programmers make some assumptions.  For instance:  if I (the enemy) can shoot Master Chief, I should or if the leader is killed and I am a grunt, I should run away.  Since a lot of it is common sense, programmers don't even try to think through each step.

Example: I see an enemy to my right

Game Tree:
a) turn to face him
b) turn to face him
c) turn to face him
d) turn to face him
... (for as many frames as it takes)
k) aim
l) aim
...
y) shoot

Other AI (single-state AI maybe?):
a) I can see the enemy
b) I will change to "attack mode"
c) I am in attack mode, I automatically turn and shoot the closest enemy
d) turn, aim, shoot

Logic gates:
A very new subject for me, and I haven't had much time to pursue it.  But this is what I know:
XOR - acts like "OR", but if both items are true, it returns false
NAND - acts like NOT((x)AND(y) ).  It always returns true, unless both items are true
NOR - only returns true if both items are false (like a NOT((x)OR(y) ) )

Theoretically (I'll get around to proving this when I have time), you can build any of the six basic operators (NOR,NAND,XOR,AND,OR,NOT) using only NANDs or using only NORs.  Everything your computer does (okay, not everything, nearly all the computing) is through these gates.  Adding, Subtracting, Multiplication, etc. are all simply a combination of these gates (if you use binary code, rather than decimal).


Edit: are image tags still not working?!  sorry 

Last edited by MoreGamesNow (2012-01-21 20:41:35)

After looking for reputable sources, I realized that I left some things unsaid.  Game Trees can only be used in perfect-information games (you need to know every possible action and all results of that action).  While this is fine for computer-games (in which the computer ultimately knows everything), in real life, game-trees are far less applicable.

Computers that use Game Trees have a "horizon" (called the "horizon effect?").  Basically, they can't see farther than they are searching (obviously).  One of the biggest problems with this is that a computer might make stupid moves to push back an even worse, inevitable loss past the horizon (out of sight, out of mind).  For instance, if the computer has to lose its queen, but can trick itself into thinking it can save its queen by sacrificing pieces, it will sacrifice pieces in a vain attempt to prevent its queen from dying.  There are techniques to minimize this (searching deeper under specific circumstances for example).

Alpha-Beta Pruning greatly increases the effectiveness of Game-Trees.

Also, something you (might) want to include is something called a technological singularity.  It was coined by a Sci-Fi author (darn it!  I used to know his name).  Basically, it is the idea that when computers surpass human intelligence, the future becomes unpredictable, as things more intelligent than us will be actively influencing/controlling it.  Additionally, the super smart computers will make even smarter computers even faster than we made them, and they'll be an production of ever smarter computers forever.  Theoretically they/we will take over the universe this way.  All theoretical of course.  Some of it is here


Anyway, I did some google searching and this is what I found:

Stanford Site on Game Trees and Game Theory in general.  I haven't read it all, and it is pretty extensive, but it's reputable 

Stuff on neural networks.  Probably way more than anyone wants to know xD.  I'm not sure if it has that "you're right/wrong" stuff on it.

* another interesting factoid: I believe the Xbox Kinect uses a neural network for image recognition... but I'd double check that if I were you.

I've got to go for now, but my personal info:

I'm not sure about Scratch's policy on names/personal information.  I will tell you that I'm only in 11th grade (Junior in high school), so as a "reputable source", I may not be, well, acceptable.