Slashdot Mirror
After Go, Developers Are Now Building AI To Beat Us at Soccer (cnet.com)
After Google's AlphaGo artificial intelligence bested our best Go player, South Korea is now setting its sights on making AI that can play soccer. From a report: Hosted by the Korea Advanced Institute of Science & Technology (KAIST), the AI World Cup will see university students across South Korea developing AI programs to compete in a series of online games, reported The Korea Times. The prelims will begin in November. "The football matches will be conducted in a five on five tournament," a KAIST spokesperson told the publication on Tuesday. "Each of the five AI-programmed players in such positions as striker, defender and goalkeeper will compete with their counterparts."
Bender: Now Wireless Joe Jackson, there was a blern-hitting machine!
Leela: Exactly! He was a machine designed to hit blerns! I mean, come on, Wireless Joe was nothing but a programmable bat on wheels.
Bender: Oh, and I suppose Pitchomat 5000 was just a modified howitzer?
If you have to "develop" AI for a specific task (play games) or whatever then in my view it's not AI. I think AI should learn to do what ever task you throw at it.
low bar
Developers Are Now Building AI To Beat Us at Some Crappy 5-Person-Per-Team, Soccer-Themed Video Game
FTFY - no charge.
Football (sorry americans, thats what the rest of the world calls it) games have had fairly decent "AI" players for years. This is hardly raising the bar. Now if they created real robots that could beat a human at football, THAT would be something to behold. Though I don't suspect Boston Dynamics will be worrying about their share price anytime soon over that possibility.
The point of Go is not to get points, it's to be used as a simulation of real-life problems to increase your ability to reason abstractly.
Go isn't a game, it's a language.
Not too many people know that nowadays.
With how easy the walking robots fall over (we have all seen the videos), it should not be that hard.
Don't fight for your country, if your country does not fight for you.
AI is the most overhyped bullshit on the planet.
Here's the code:
if (in_posession_of_ball && opposition_near) {
take_a_dive_but_pretend_you_were_fouled();
}
Summation 2
"There is no rule that a dog can't coach a team of robots!"
To me, those are glorified tic-tac-toe programs. They're not "thinking" about how to win at the game, they are simply exhaustively searching the solution space to find moves whose branches give them the best chance of winning. Tic-tac-toe always ends up a tie with semi-competent players because its within the capability of humans to see the entire solution space and avoid moves which result in a loss. Computers are simply able to search a much larger solution space in the same amount of time as a human.
Humans don't play chess and go like that. The solution space in those games is too massive for them to thoroughly search, so they have to resort to a different strategy. They learn patterns and trends from repeatedly playing the game. Then they develop algorithms and heuristics based on those patterns. When they play a new game, they try to determine which pattern the current game board best approximates, and apply the appropriate algorithm to decide what their next move would be. Program a computer to learn how to play like that - without searching the solution space - then I will be impressed. e.g. What's the best chess program you can write while limiting it to, say, a thousand iterative loops for each move?
Has anyone read "Artificial Intelligence with Python" by Prateek Joshi? Is it any good?
That is not how Go and chess ai works. Chess ai does not exhaustively search the entire solution space, rather it usually uses heuristics in order to find out which parts of the move tree contains the best solutions.
Go AI is a bit different altogether, as it uses a neural network (at least AlphaGo does) in order to match patterns of play. It was first trained using games between professionals, and later by letting it play against itself.
None of these types exhaustively searches the solution space, which would imply running through EVERY SINGLE possible game. For chess that is somewhere around 10^120, for Go over 10^300 - way too large to compute efficiently.
how many ea coins will they get from that?
No, it's absolutely not about brute-force searching. The search space for Go is so massive that even the fastest computers really can't do exhaustive, brute-force searches for a solution. That's part of what made an AI winning at high-level Go such a milestone. To give you some context: the search space for Go is significantly larger than the estimated number of atoms in the universe.
I agree that it's a bit silly to call these algorithms "AI", but they're not nearly as simplistic as you're making them out to be. To be effective, the algorithms have to do a massive amount of heuristics-based culling before it can start searching for solutions, or else it would get bogged down in the math, no matter how fast it was.
Irony: Agile development has too much intertia to be abandoned now.
Now football is a real sport. Unlike American football
South Korea is now setting its sights on making AI that can play soccer
Playing soccer (aka football) is the easy bit: kick and miss.
But will the AI be able to convincingly roll around on the ground just inside the penalty box? And more importantly will it be able to deal with the paparazzi and make vacuous statements like "it was a game of two halves" and "if we'd only scored more, we could have won".
politicians are like babies' nappies: they should both be changed regularly and for the same reasons
Association football is the correct name for the sport you are trying to describe.
Football includes a very large number of related sports, there is no single sport that is the true football. Some nations have officially tried to redefine the word "football" to apply exclusively to "association football" but bureaucrats don't have the authority to redefine language to be less precise.
An AI that can beat the best human players at Madden?
were made by an octopus.
Slashdot, fix the reply notifications... You won't get away with it...
When you say "football" do you mean Gaelic football, Rugby football, or Association football (soccer is a shortened form of this). I assume you aren't referring to something obscure like Shrovetide Football.
Hey, be nice. Even straight people might feel threatened by AI. Now that Google can beat humans at purely intellectual challenges, like Go; they need AI to beat humans at physical challenges like soccer, and the locker room activities that follow. But no need to worry about the potential downsides of AI.
A guy worked at a coal mine. The coal mine closed.
So he got a job as an auto assembly line worker. His job was replaced by a robot that could do the job faster, more accurately and cheaper.
So he became a truck driver, because those trucks aren't going to drive themselves.
I'll see your senator, and I'll raise you two judges.
Python is 50 times slower than C++, but Python also allows to write in an easier way algorithms that are exponentially faster. O( n ) >> O( 50 log (n)) ..
Slashdot, fix the reply notifications... You won't get away with it...
I have no idea what KAIST is up to in November, but Robocup doesn't have anything to do with it.
In Robocup we've been playing robot-soccer for decades, also with real robots. Always completely autonomous.
The next World Championship is in a couple of weeks in Nagoya, Japan: https://robocup2017.org/eng/index.html
And if you're curious what 5 vs 5 with real humanoid robots looks like, check out our YouTube channel: https://youtube.com/htwk-robots
Now, turn those AIs into robots and you may have something.
But what the article is talking about is not AIs that can play soccer, it is AIs that can play a soccer-themed online video game.
We can't compete with robot soccer. http://imgur.com/1Yfuwrn
Well, the RoboCup has it's goals set way higher for decades now!
Id like to see Robo Baseball! Seems like a lower-hanging fruit.
"The first official RoboCup games and conference was held in 1997 with great success. Over 40 teams participated (real and simulation combined), and over 5,000 spectators attended." [http://www.robocup.org/a_brief_history_of_robocup]
By "real and simulation" they mean that AI has been playing soccer for more than 20 years, both in simulation (as in TFA) and in real, physical robots.
Welcome to the world of AI research, South Korean!
(To be fair, it's probably some reporter's snafu, rather than a researcher's.)
Talking about soccer, the football game played with feet: can't bear watching a game.
Overpaid players who miss everything they can, and who prefer to cheat, pulling shirts or worse, instead of fighting for the ball and the goal.
Disgusting.
Totof
... now that would be interesting. Give, they'd probably kick the ball at 300km/h, unstoppable for mere mortals, scoring goals from every position and across the field.
But it would be a neat step forward in engineering none-the-less.
We suffer more in our imagination than in reality. - Seneca
An AI will be terrible at playing soccer, because you can put 11 computers on a soccer field running whatever program you want, and no matter how good the AI they are running, they will just sit there.
WTF is "an" AI?
An AI is a program, running on a computer, that simulates intelligence and/or solves problems in the way an intelligence solves them.
.... I don't know what "an" AI is capable of because I don't know what "an" AI is....
Now you do.
You're welcome.
This type of research will really make a difference in people's quality of life. :/ Yet another example of how easy it is for the average team of engineers to thoroughly lose the plot. Then again, maybe playing games is all 'AI' is good for. I think both are likely.
...but that wouldn't be much of a challenge!
It's spelled F-O-O-T-B-A-L-L you fucking idiots.
I remember in 1998 I downloaded a new chess program and somebody asked if computers were the end of chess. My father said, "No, and airplanes didn't end running."
But for context, top chess engines added strong non-brute-force algorithms 20 years ago, and they've had pruning of bad lines for longer than that. So even when they're saying "brute force," it isn't really and doesn't add up to silly numbers like the atoms in the universe, which is just %$&^!*@.
The problem with Go was always, according to the chess programmers, a problem of programming hours. There were simply more people working on chess software. The newer results are exactly as predicted; it is a matter of algorithms, and Go is more like a self-driving car than it is like chess. But still, they're also similar. The problem space isn't a list of every possible legal move, but rather whatever the list of known strategies is. In Go, as in chess, they had help from top professionals in the sport to get useful algorithms into the software. Then the software just drives through the strategies, and brute-forces the things that it "knows" contextually. But that is what all software does anyways, so it isn't even AI. It is just a sort of semantic calculator.
https://www.amazon.com/gp/product/B000FC0SIM
https://www.amazon.com/gp/product/B0054RPY7I
https://www.amazon.com/gp/product/B008DQ7EMK
https://www.amazon.com/gp/product/B009VV7G60
https://www.amazon.com/gp/product/B01IRD0LBY
was of some dude curled up in a fetal position on the ground, screaming and holding his shin.