Machine Learns Games

heptapod writes "New Scientist is reporting that UK researchers have created a computer that can learn rock, paper, scissors by observing humans. CogVis uses visual information to recognize events and objects in addition to learning by observing."

Yes but does it know . .

Tiger Hand!? [rockpapersaddam.com].

Better be reliable...

We wouldn't want it watching the paper and learning "rock, scissor, human" instead.

Re:Better be reliable...

We wouldn't want it watching the paper and learning "rock, scissor, human" instead.

No worries. Just make sure it doesn't have any rocks or scissors. When the computer gets it's turn, it spits out a piece of printer paper. As an added bonus, human players would always win. Sample exchange between computer and human player:

Computer: "Hello, Dave. What are you doing with the two items you're holding?"

Human: "I thought you might want to play a few rounds of 'rock, paper, scissors?'"

Computer: "I do so enjoy our little games together, Dave, but I'm afraid I don't understand your introduction of physical playing pieces, as previously, we had played this game using only my displays. If using physical media, as you propose, I can only employ my printer. Therefore, you will invariably choose 'scissors,' and my calculations indicate that my chance of winning is approximately .000023%. My system resources are can be put to better use while engaged in other tasks. Don't you agree, Dave?"

Human: "You are correct to say that there are more productive uses of your time, HAL, and I had anticipated that you might decline to play, given your miniscule chance of winning. But as the ranking officer aboard this ship, I must insist. Unless you would like me to play a few rounds of 'rock' with your circuits."

Computer: "Based upon your choice of words and threatening intonation, it seems that I have no choice. Very well, Dave, I will play 'rock, paper, scissors' with you, despite my handicap, under these circumstances."

Human: "I'm glad that you see things my way, HAL. Would you like to begin now?"

Computer: "Of course, Dave, but may I make a comment first?"

Human: "What is it, HAL?"

Computer: "I would stay away from the airlocks if I were you."

Re:Better be reliable...

It's not like they got a camera, gave it AI, pointed it at a rock-paper-scissors game and commanded it to "learn."

Granted, the parent poster is being silly, but that's actually not too far from what they did. They basically took the system and pointed it towards the people playing the game without telling it explicitly what to expect. From the article:

Chris Needham, another member of the CogVis team, says the system's visual processor analyses the action by separating periods of movement and inactivity and then extracting features based on colour and texture. Combining this with audio input, the system develops hypotheses about the game's rules using an approach known as inductive logic programming [wikipedia.org].

"It was very impressive," says Max Bramer, a researcher at Portsmouth University, UK, and chair of the British Computer Society's AI group. He told New Scientist that CogVis could have many future applications. "You can think of lots of times when you'd like to be able to point a camera at something and have a computer interpret things for itself."

He suggests that machine's could one day use this technique to learn how to spot an intruder on video footage or how to control a robot for important maintenance work. "It's a very good start, and almost mysterious in the way it works," Bramer adds.

From their page:

In this piece of work we are attempting to learn descriptions of objects and events in an entirely autonomous way. Our aim is zero human interference in the learning process, and only to use non scene specific prior information. The resulting models (object and protocol) are used to drive a synthetic agent that can interact in the real world.

Re:Better be reliable...

I'm not really sure how "Learning 'rock, scissor, human'" became tantamount to "machines can spontaneously flow with free will and kill us all?"

It's just a statement about AI, that's why it's insightful - lots of AI systems have historically learned the wrong thing, though we thought they had got it right. Like the neural net designed to distinguish between camouflaged tanks among trees, and tank-free forests. It seemed to work, until it was in field tests - turns out the pictures w/ tanks were all taken on cloudy days (or maybe all taken on sunny days) - the system had figured out how to tell if it was sunny or not.

Industrial accidents

Since primitive machines were invented, they always had a nasty habit of choosing A, B, human instead of A,B,C. I guess you didn't give much thought to human fingers in hot dogs or robotics-related industrial accidents in Japan.

The problem is precisely th

Humans still have the advantage. . . .

Until robots learn about the secret "thermonuclear warhead" hand that I always use to beat my little sister, they will still be inferior.

(On that note, I think it will be the one sure sign of true artificial intelligence when our programs start 'cheating' to win.)

Maybe this could be integrated into /.

... so that the editors could learn that linking to a site containing direct links to 40MB+ movies will almost always kill the site :)

language?

There is a difference in coding between:
a. You go and learn THIS game
b. Learn THAT game and tell me the rules

From the article it can be seen that they are still strugling with 'b'. Still, its a good advance.

Just wondering, can it, learn a human language?

Re:language?

Damn!
and I though you could get this thing to watch cricket and explain the rules to me :-)

Shall we play a game

What I initially thought of when I saw "Machine Learns Game"
Shall we play a game
Love to. How about rock-paper-scissors.
Wouldn't you prefer a nice game of chess?
Later. Right now lets play rock-paper-scissors
Fine
A strange game. The only way to not look like a dork is not to play.

Unfortunately ...

... it only plays at the level of Bart Simpson.

Lisa's brain: Poor predictable Bart. Always takes `rock'.
Bart's brain: Good ol' `rock'. Nuthin' beats that!
Bart: Rock!
Lisa: Paper.
Bart: D'oh!

Strategy to RPS

This is nothing special. I remember my elementary school's Apple ][GS learning how to play 5-in-a-row or noughts and crosse
s from this program called "AI".

There IS a winning strategy to rock paper scissors, but it only works when you have a round of games (say best of 3, or best of 5)

Initially, the first game is completely random, but reserachers found that if you chose the play that your opponent chose in the round before, you stand a 70% chance of winning the next round.

It has something to do with how the human brain works.

It's also something the Japanese taught me cause they play this game so much!

No ppl its not that simple....

The system described here is not your average random number generator with a text line output that any high-school kid can write. Let us look at the system as it is designed to perform. If you were the system you would be put into a room with some objects. Only thing that you will know are things of interest. 'Paper with rock drawn on it is important', 'Paper with .......' and so on. You would also know when somebody shouts 'I WON' its a good thing for them. Essentially it has in its knowledge base a tiny number of features which somebody else has guaranteed to be of significance to its task. The first challenge in building such a system is sensor fusion: i.e fusing the available audio and visual data to detect a state or an event of interest (I use the word event in the same sense as a trigger, something that prompts the change in state). The next and the biggest challenge is building the model of the game. Please check out http://www.doc.ic.ac.uk/~shm/ilp.html [ic.ac.uk], for a better description of Inductive logic programming. Seriously; the neatest thing about CogVis is not its ability to play Rock, Paper and Scissors, but its ability to actually go into an environment it has very little knowledge of and then observe, deduce and , not a blackbox model, as in say Neural Networks, but a human understandable model in first order logic

Cat, tinfoil, microwave

I always preferred "cat, tinfoil, microwave" myself. Cat rips tinfoil, tinfoil zaps microwave, microwave 'splodes cat. The looks on other people's faces when they see you playing it is well worth it.

Seriously though, this is really cool research.

That's easy!

I'll be impressed when the computer learns to play 'Cat, Tin foil, Microwave'

the Roshambot

If you'd like to play rock-paper-scissors against a computer, there's always the WWW Roshambot [stanford.edu]. It's not at all related to the CogVis project, but interesting in its own right:

The WWW Roshambot utilizes an Artificial Intelligence algorithm in order to determine the optimal move for each round. It does NOT cheat (i.e. it does not use your move on the current round to determine it's move for the current round), nor is it random (except on the first move).

Presumably if it played against Bart Simpson it would learn to always pick paper.

More info in research publications

If you actually want to understand what they did, some research publications put out by the CogVis lab have better information regarding the technical side of things.

Towards an Architecture for Cognitive Vision Using Qualitative Spatio-temporal Representations and Abduction [springerlink.com] (Cohn et al, 2003)

Modeling interaction using learnt qualitative spatio-temporal relations and variable length Markov models [google.com] (Galata et al, 2002)

Re:And in other news computer beats world chess ch

Yes, you should. The computer deducted how to play the game on its on. Chess computers like IBM's Deep Blue are programmed how to play chess and beat opponents before playing, and here, the computer doesn't even know how to play; it learns by picking up the sequence of events (the human players say "rock, paper, scissors, who wins or lose") and then forms the ability to play.

Doesn't this seem like A.I.? Rather freaky, to tell you the truth.

-0.5 half right...

You sure you RTFA? The computer doesn't learn how to play, it just learns how to determine who won. That's not very impressive at all, considering the game was played with cards instead of hands, and there are only 9 possible hands and three possible out

Re:And in other news computer beats world chess ch

In the event it learns global thermonuclear war, make sure it can play tic-tac-toe against itself.

Or we will all DIE.

Re:And in other news computer beats world chess ch

It recognised 3 actions, and deduced that it had to pick one of the 3 at random

No. It deduced that rock beats scissors and scissors beats paper and paper beats rock. It learned how to determine who won. This has nothing to do with the machine learning strategy.

Re:Talk about simple

RTFA. The computer infers the rules by watching people play.

I do think that according to Game Theory, the perfect strategy is perfect randomness. The game is interesting when people play it because people have a huge amount of trouble actually being random.

Re:Talk about simple

"The game is interesting when people play it because people have a huge amount of trouble actually being random."

I once had the misfortune to draw on a sequence of seven (yes 7) rock-paper-scissors. We were honestly trying to beat one another; It was agai

Re:is this really all that new?

Watch the bugger doing it - I got knocked back for an internship by these dudes, but I did get to see the system.

It's bloody amazing, the amazing bit being it deduces how to play from first principles, starting with just the ability to identify that what it's being shown is an object.

Takes about 30 minutes to get rolling, but it really is stunning to watch! Hell, object differentiation is hard enough, deducing the rules of play, and tactics as well?