Slashdot Mirror
Google's AlphaGo AI Beats Lee Se-dol Again, Wins Go Series 4-1 (theverge.com)
An anonymous reader quotes an article at The Verge about Korean grandmaster's fifth and final game with Google's AlphaGo AI: After suffering its first defeat in the Google DeepMind Challenge Match on Sunday, the Go-playing AI AlphaGo has beaten world-class player Lee Se-dol for a fourth time to win the five-game series 4-1 overall. The final game proved to be a close one, with both sides fighting hard and going deep into overtime. The win came after a "bad mistake" made early in the game, according to DeepMind founder Demis Hassabis, leaving AlphaGo "trying hard to claw it back."
letting the human live after the first game.
I imagine the next version will go 5-0 as these kind of things tend to be iterative in nature.
and at that point go will be as bad as chess and it will be nigh impossible to find a fair game online.
Except this had never been done before without either player having a handicap at this high of a level. You might not find it interesting, but it is still breaking new ground.
Nobody is trying to make a general intelligence because nobody wants it. What is wanted is domain specific algorithms that are very good at what they do.
Although, it seems that the tech is quite general and learned to play multiple Atari games without having to be tuned for each.
Guess what - computers can run more than just AlphaGo.
Those who do not learn from commit history are doomed to regress it.
It's a huge surprise.
Your comment is a perfect example of the AI Effect:
https://en.wikipedia.org/wiki/AI_effect
was that really an "early mistake" or was it part of the plan? how do we know?
Tonight on Sore Losers, Se-dol makes a noodle soup which gweihir calls "Fantastic! Better than that robot swill they sell at Google".
I can assure you that the hardware Alpha Go runs on is well capable of handling other tasks. It is true that single programs will probably always tend to be specialized. It is better to keep the AI that excels at Go separate from the one that is a superior driver than a human, and from the one that does medical diagnosis better than a human. No reason they have to be, just better engineering.
Alpha Go is significant. The primary way it developed from a good Go player to one superior to humans was by studying the games of others and by experimentation, playing versions of itself to develop a superior knowledge of the best patterns. This is highly applicable to other applications that you would consider more useful.
At this point, there are still some things humans do better than AIs. Dealing with imperfect knowledge is a challenge slowly being overcome. Partnering with unpredictable humans is difficult (that is why self driving cars are tricky). The areas where humans beat AIs are steadily becoming fewer.
I get tired of hearing people say that Go is a game that required creativity to win. It doesn't and if anything, this result demonstrates that.
"AlphaGo's algorithm uses a combination of machine learning and tree search techniques, combined with extensive training, both from human and computer play."
It's a game, based on pre-defined rules.It's just more opague and vague than chess.
Sorry, but I'm a mathematician. Check my comment history, I'm the first to disparage any kind of "AI" (which just means human-programmed heuristic most of the time), especially that which just does brute-force search of possibilities. That's NOT AI. Almost every "game AI" isn't AI. Not even close.
However, in uni, one of my lecturers was studying Go as one of his prime areas of research, and I've seen - and checked - some of the numbers here.
You have no idea what this machine has just done. It's leapt forward some 10-20 years in terms of computer Go-playing capability in one fell swoop. The numbers involved in Go are so huge that brute-force search, even for a limited number of moves, is absolutely impossible in the times given.
And it isn't being given programmed hints, because Go is just too complex a game for that beyond amateur play. There's a handful of hard-and-fast rules of what's a stupid move and what's not and everything else interacts SO MUCH with the rest of the board and future plays that it's almost impossible to even tell who's winning most of the time!
As such, this system, no matter the power behind it, is doing something that dumb, brute-force, play-the-game AI written by world-experts in Go, AI, and game theory wasn't expected to be able to achieve within the next decade. And it primarily gets there because it learns from information fed to it.
At that point, although it's only limited to Go, the engine is proving itself capable of - almost - a kind of intuition, insight and "feel" for the positions rather than anything to do with numbers and scoring and weighting and pre-written rules. Now, that's a vastly overblown explanation, still. The computer isn't "feeling" anything. But whatever it's emulation and use of such, it's leaps-and-bounds ahead of its competitors.
This is why it makes BBC News, Slashdot and every other media outlet. It's not just winning by brute force. It's doing something else. It's spotting patterns in data it's never been exposed to before. It's able to hypothesise and learn from mistakes on board layouts that maybe NO HUMAN HAS EVER SEEN BEFORE OR WILL AGAIN (that's how large some of the numbers of possibilities get!).
Even a pack of cards, with 52! = 8x10^67 potential arrangements of a shuffled deck:
http://www.murderousmaths.co.u...
Pales in comparison to the number of possible Go positions (2x10^170) and the ways that you can move from one to another (~ 1 x 10^768). And that's just on a standard 19x19 board (something almost unplayable for a computer just a decade again).
This thing isn't calculating. It's gaining insight from historical observation and applying that to self-similar situations that nobody has ever been able to analyse, nor which it could ever analyse fully in the time given. That's the start of "true" AI. It's only a start, but it's quite seriously ground-breaking in that ability.
And once you start down that route, there's nothing stopping AlphaGo quickly learning every similar game, then dis-similar games, then other games, then other things entirely, using the same kinds of system underneath.
Honestly, there's a reason that game theorists and AI-experts are making a fuss about this.
If Google AI you can beat Lee Se-dol at Go, can it beat the IRS and Her Majesty's government at Tax Evasion? http://www.huffingtonpost.com/... http://www.theverge.com/2016/1... http://www.thelocal.it/2016021... http://www.bbc.com/news/magazi... http://news.yahoo.com/italy-cl...
This was a great proof of concept for some "intuition" in AI, one of the behavioral aspects people believed hard to reproduce.
Now I am really looking forward to see the real applications for this, and their consequences:
- smart AI assistants, "a Siri that actually works" and similar
- AI assisted science
- AI assisted healthcare
There is a great interview with Demis Hassabis about this. There is hope for noticeable progress in mass products within 3-5 years.
This new tech will help a lot of people directly, and the related mass unemployment threat should force us to adopt better social policies. I already start hearing about base income experiments and the like more often.
Yes, absolutely nobody is trying to make a general intelligence that could be given a copy of its own code, taught how it works, and allowed to make whatever changes it can think of. Nobody at all would find that a cool project to work on, or want, or could forsee it leading to interesting developments. Nobody.
Idiot.
I've been following the matches with the same expectation and anger I felt in 1997 during the Kasparov & Deep Blue rematch. The final result has been similar, and although it has been well reasoned that chess and go are pretty different games and Deep Blue and AlphaGo are pretty different machines, the bittersweet sensation is identical. I had a naive hope in the human superiority just for a little more time. I was pretty sad after the final game: Lee Sedol seemed really disappointed and sad himself. I can't imagine the pressure he's felt throughout the event, and his face -that's my impression- seemed to tell us "I've failed you all". He later told in the press conference that he felt he could have done more in the games -I'm sure he'd like to play more games to test himself again- and I wonder what could have happened if the matches would have been played without general knowledge. Feeling that kind of coverage must have been really stressful. If you ever read this, Mr. Sedol, thank you. And please, don't ever feel disappointed, you've done a fantastic job.
It won't turn out that Se-Dol has quite a few other skills. That's the problem. There's too much focus on brainpower to solve these highly restricted set of problems. That's the issue. What makes real creativity is not a mind like Lee Se Dol (with respect). It's the people that are capable of inveting somethng truly original. AI can't do that .. yet. That's true creativity, and it's not something Lee Se dol has, or something that you find very easily in Asia, generally.
How would Einstein do against Lee Se Dol? Not ver y good..
Who would you put your money on to contribute to meaningful, original science? Einstein .. every time.
Even just this last week, another of Einstein's theories has been verified - gravitational waves, as shown by the LIGO istrument. He has more "creativity" than the sum totoal of all these Go players.
Could you define your general competition? I'm sure Deepmind could do a bot which is crushing anyone at Go and Chess, for example.
Nobody is trying to make a general intelligence because nobody wants it. What is wanted is domain specific algorithms that are very good at what they do.
Well it's bits and pieces of it. Imagine you could start to combine Watson and AlphaGo, you tell it to "I'd like to play a round of Go with you" and Watson does the natural language parsing of the request and the rules, finds some games that would make good training material, spawns up an instance of AlphaGo that does the initial training and self-training to improve its play. Yes, the end result is to be able to solve domain specific tasks, but the goal is not to create domain specific solutions but more of a "solution factory". It's still not a generic intelligence that'll learn across the domain-specific ones but a collection of them might mimic it fairly well for a wide variety of tasks.
Live today, because you never know what tomorrow brings
You have no idea what this machine has just done. It's leapt forward some 10-20 years in terms of computer Go-playing capability in one fell swoop. The numbers involved in Go are so huge that brute-force search, even for a limited number of moves, is absolutely impossible in the times given.
And it isn't being given programmed hints, because Go is just too complex a game for that beyond amateur play. There's a handful of hard-and-fast rules of what's a stupid move and what's not and everything else interacts SO MUCH with the rest of the board and future plays that it's almost impossible to even tell who's winning most of the time!
As such, this system, no matter the power behind it, is doing something that dumb, brute-force, play-the-game AI written by world-experts in Go, AI, and game theory wasn't expected to be able to achieve within the next decade. And it primarily gets there because it learns from information fed to it.
For those who are more involved in AI research it is not so surprising. Similar general approaches to learning have been used in the "cognitive" branch of AI research for the last 15 years or so. The buzzword changed from "cognitive" to "deep learning" recently.
The key to success of AlphaGO is the position evaluation function that is learn from data. The surprise here is that learning from the game endings of internet GO players and somewhat informed computer vs computer games is enough to train an evaluation function with the predictive power to beat the world champion. In the old days of AI an expert-designed heuristic function would be used instead and a kind of smart position tree search would do the heavy lifting. But obviously this didn't work with GO due to combinatorial explosion and very difficult evaluation in the beginning and middle stages of the game.
Oh right. This is a leap forward in AI. That's why they only needed a mere 1,202 CPUs and 176 GPUs.
Did they program this thing in a smart way? Yes.
But "hypothesise", "true AI", even "intuition"? Sod off. They threw brains, statistics and a shitload of CPUs against the problem, and like any computational problem, it disappeared.
(same AC)
By the way, I'm a mathematician, working in a computer science department. Not that that should influence the validity of my argument, but you seem to care about those kinds of things.
I am not a mathematician, and I find this victory rather unimpressive and totally expected given the progress that have been made in machine learning in the last 20 years.
Go is rather simple compared to other problems like image recognition. The number of Go positions is dwarfed by the number of possible images (a 1M pixels color image leads to (3*255)^(10^6) possibilities - of course not all of them are valid and the manifold of relevant images is much smaller, but so is the manifold of relevant Go positions, I guess), and we've come up with pretty good results in those areas. Better than what humans can do in some areas.
The real question is: What is intuition? Is it something computable or not? If it is only some kind of statistical inference, then no wonder we are good at it: we have an inference engine which structure has been optimized by million years of evolution, and fed with bazillions of samples since our birth. But that doesn't mean it's impossible to build one as good as us. Sure the design of the model is trickier, but it's easier to feed the training samples given our technological ability to gather huge amounts of data.
I wonder when the term "true AI" will be ditched. To me, there is not "true AI" because there is no "AI" as opposed to "natural intelligence". The only difference is whether your computer is biological or electronic...
Video of some good progressive thrash music
For some reason you are assuming the "underneath system" can be made generic enough to be applied to dis-similar games and what-not.
The system conquered the "Go problem" and that by itself isn't necessarily a road for anything else.
Then you've got no excuse at all for not running the numbers.
1000 CPUs is nothing against 10^176. It's 10^3. You could have a billion times more CPUs running for billions of times more time and STILL not come close to evaulating a BILLIONTH of the possible moves even if they were all running at a BILLION GHz. NOT EVEN CLOSE.
1000 CPU's isn't even a rack. It's not even enough to handle a pittance of an Internet service. It's not even comparable to Deep Blue in terms of those numbers. Yet it's beat EVERYONE at a game several orders-of-magnitude numbers of orders-of-magnitude more difficult than Chess. And it can win against virtually everyone else with MUCH LESS than that stated number of processors.
Honestly, I'd expect better from anyone with a vague feel for numbers to look at the powers in the above post and realise what this means.
We have jumped some 10^50 - 10^100 times ahead of current machines (nobody could ever beat a grandmaster of that Dan on a full-size board without handicap EVER BEFORE, and weren't predicted to in our lifetimes even using Moore's Law-like extrapolations of much more powerful supercomputers than 1000 CPUs) by using a different TYPE of system entirely.
It won't turn out that Se-Dol has quite a few other skills. That's the problem. There's too much focus on brainpower to solve these highly restricted set of problems. That's the issue. What makes real creativity is not a mind like Lee Se Dol (with respect). It's the people that are capable of inveting somethng truly original. AI can't do that .. yet. That's true creativity, and it's not something Lee Se dol has, or something that you find very easily in Asia, generally.
How would Einstein do against Lee Se Dol? Not ver y good..
Who would you put your money on to contribute to meaningful, original science? Einstein .. every time.
Even just this last week, another of Einstein's theories has been verified - gravitational waves, as shown by the LIGO istrument. He has more "creativity" than the sum totoal of all these Go players.
Playing a game, and doing it well, requires real creativity. Arguably a lot more than science, actually. When you study science, all you're doing is discovering information already out there - water had its properties and was built by molecules long before it was classified as H2O, and nothing changed after. Doing well at Go can't be calculated cold and hard - much of it is subjective, and that's what makes this discovery so important. The computer didn't win by just repeating the same patterns or evaluations over and over, but actually learned from each game and was able to apply that to the future. That's the start of self learning AI.
Like a ton of people in the world (the majority most likely), you apply the no true scotsman argument to this debate. It's not real AI until it learns strategies not programmed into it? Oh wait, no, it's not true AI until it creates its own strategies? Oh wait, no, it's not a true AI until it can do this to something other than Go? What next, it has to socialize and disobey? The approach this machine used was incredible, and the insight was extremely important - being able to learn by studying a history of decisions, that's something that lays the groundwork for every future AI project from here on out.
This represents a massive step forward in artificial intelligence, by leaps and bounds, and the sad part is, you don't even know it.
"Set a man a fire, he'll be warm for the rest of the night. Set a man afire, he'll be warm for the rest of his life."
It is not because you are completely uninterested by a subject that advancements in that subject have less "creativity" than in other subjects. I could reverse your 'demonstration' by saying that Einstein did get lucky that some of his results were proved true. He has been proved wrong about quantum entanglement. Should I compare him to Leonardo Da Vinci who was a genius painter, a great engineer and an anatomist and made great advance and publication in these subjects but was also interested in invention, sculpting, architecture, science, music, mathematics, literature, geology, astronomy, botany, writing, history, and cartography.
AlphaGo in its matches against Lee Sedol showed 3 interesting moves who will certainly be studied and played by all the professional Go players around the world for the next years. It is not because you limit yourself to a certain set of problems with the objective to excel at it that you are a lesser being that someone else.
Why would you want to look at all possible moves? Humans don't do this, why would it be necessary for computers?
You can make a fair judgement without evaluating every individual option. You can be selective. So I don't know why you think 10^176 is the only number worth beating, because it isn't.
There is no reason to beat 10^176, because various (deterministic and well-understood) techniques help you severely reduce that number.
Also, the number we're working with here is not 10^3, it's 10^3, times 2 gigahertz (say) per processor, times (at least) 60 seconds of computation time per move, which is closer to 10^14.
Is this a respectable achievement? Yes. They must have applied some smart techniques to prune the search tree. But definitely a big chunk of the achievement was made using more kilograms of processing power.
If you prefer believing that these programmers found a way to write consciousness in C, I can't stop you. But I don't think that your viewpoint will get us anywhere.
(same AC)
By the way, I'm a mathematician, working in a computer science department. Not that that should influence the validity of my argument, but you seem to care about those kinds of things.
And when you're done grading your professor's CS 200 papers, there's a nice extra credit bonus for you.
It doesn't require creativity. It just requires a deep understanding of the game. The game itself has a very restrictive set of rules. It isn't creative at all. BTW, game playing isn't AI at all. Computers are good at playing games with a restrictive set of rules. In fact that is the one thing they are best at: computers LOVE rules and require them to perform any task.
He's likely to be remembered as the last human being to beat a Go AI on tournaments.
Move 78, in particular, was so good that his partners and commentators in China have already called it "the hand of God", but it really was one of those things which happens once in a blue moon, even for a player like Sedol.
He doesn't, and you never got the impression that he does.
Strawman arguments are lies.
They actually got poker before Go, about a year or two ago, at least in heads-up situations.
that the human was not a good player, then. I mean, if the computer wasn't perfect by losing once, surely the human must be crap at Go if it loses more than once, right?
Or is this editorialising against the meatbag and not allowed?
Come back to me when it can win at Roulette.
The team who made alphago deserve credit, but their approach (from a high level) isn't so revolutionary. Go AI devs moved away from solely using brute force tree pruning (like what deep blue used) a long time ago.
The first big change was to use pattern recognition (matching sub-sections of the game with already known patterns) to prune faster. The second (and far more revolutionary) change was to apply an upper confidence bound based on a monte carlo simulation. This is where computers gained the ability to bypass those billions of moves with a margin for error. The third was the use of neural nets as a way to balance between brute force and pattern matching while managing the confidence levels of the monte carlo simulations.
The biggest difference with Alphago is corporate backing. I don't know how many people Google hired for the job, but the paper lists 20 (so probably more than that). Buying and running supercomputers is extremely expensive as well. With the exception of darkforest (Facebook's go machine which, as expected, appears to use a similar design), most teams consist of a very few people on small budgets without someone willing to spend millions to buy and run supercomputers for them.
Go is rather simple compared to other problems like image recognition....
No, not really. We've had relatively strong image recognition algorithms for a good while now, and i'm not talking just about Google Image Search. Image sensors have been used for a long while in industrial automation settings, from anything for measuring to actively identify features in production lines. As a problem is way more accessible than Go is.
The real question is: What is intuition? Is it something computable or not? If it is only some kind of statistical inference, then no wonder we are good at it: we have an inference engine which structure has been optimized by million years of evolution, and fed with bazillions of samples since our birth.
Agreed. One could argue that the way AlphaGo picks up moves (adaptive neural network) is "intuitive"; we don't know really what drives after some training. A Google engineer today cannot really tell you why AlphaGo favored some moves over others.
... by saying "I Knew I shouldn't have had those two scotches before the game" and winking. Because when it comes to being a B*tard, humans can beat a computer every time!
This represents a massive step forward in artificial intelligence, by leaps and bounds, and the sad part is, you don't even know it.
That's wishful thinking from a bunch of singularity nuts.
There is nothing new here. The same "AI" techniques have been used countless times before. The primary difference here is that the game is Go and the computer is fancier.
This won't get us any closer to the AGI fantasy you've got in your head. We already know, and have known for decades, that computational approaches won't get us there. Car analogy: No matter how fast your new car zooms around that racetrack, it's not going to end up anywhere else.
Why would you want to look at all possible moves? Humans don't do this, why would it be necessary for computers?
Yes you do. You just use intuition to skip over moves that might not be worth your time, but you still consider them. AlphaGo does something similar with a neural network before brute-forcing into good possible moves.
Still, even if you don't want to consider 10^700 possible game trees on a clean Go board, the problem is still intractable. Go has, in average, 250 possible value moves to consider after each stone is placed. Chess has around 30.
100% correct.
Nobody is trying to make a general intelligence because nobody wants it
Nobody is trying to make a general intelligence because nobody has the foggiest fucking idea how to do that. We still argue over the definition of intelligence.
What is wanted is domain specific algorithms that are very good at what they do.
If I could be sure it wouldn't go all Skynet on me, you can bet your happy ass that I would like to have a general intelligence to write those domain-specific algorithms for me.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
WhatsApp is hardly my definition of a "pittance". 140 million concurrent connections on 800 servers, mostly dual-socket Ivy Bridge (40 threads each). That's 1600 CPUs and definitely more than twice a pittance. Contents of all that traffic is another story, a few precious needles embedded in even more of a pittance plague than our Slashdot exchange here.
For the quasi mathematicians among us, the size of the Go search space is completely irrelevant, as chess already exceeds direct search by repeatedly told (but generally uncomprehended) orders of magnitude. What makes chess different that Go is that a pruning gradient was far easier to construct, with material advantage acting as a New York phone book booster seat (they still had those then).
The search gradient in Go is far more subtle, but whatever it might be it didn't escape the neural networks for very long once they became any good. A really good article on this is The Believers by Paul Voosen, but it's moved behind a subscriber paywall since I last accessed it.
Here's a snippet for flavour:
From the dictionary:
Artificial Intelligence:
1
: a branch of computer science dealing with the simulation of intelligent behavior in computers
2
: the capability of a machine to imitate intelligent human behavior
Did you catch that? simulation. imitate.
A simple brute-force chess-playing algorithm absolutely qualifies by this definition. It is an enterprise of mimicry, not recreation!
People keep saying "this isn't AI, this still isn't AI" as if the engineers are claiming to have created life in a lab. That isn't what the goddamn word means.
This is artificial intelligence. It is simple, task-specific artificial intelligence, but that is covered by the definition.
I think it's obvious that computers will shortly be able to be any human player at virtually any kind of structured game. In fact, I have a hard time imaging a game where computers won't soon be able to beat a human.
Even unstructured games like Pictionary and Cards Against Humanity will eventually be able to be played well by computers (after enough training and live competition). Determining the "winner" of those games is subjective, but I've little doubt that computers will eventually be able to master them.
Just cruising through this digital world at 33 1/3 rpm...
Nobody is trying to make a general intelligence
This is already false.
"First they came for the slanderers and i said nothing."
This thing isn't calculating.
What exactly do you think it's doing with all those GPUs then? It's making a ton of calculations. Really. It's also doing tree searches, and it's also using a monte-carlo algorithm to prune the tree. On top of that, it used a neural network to fine-tune its position evaluation function.
It's actually rather incredible how much calculating power Google threw at this problem.
"First they came for the slanderers and i said nothing."
Going to a 1000 machines doesn't make that much difference - the distributed version only wins against the single-machine version 75% of the time.
Now let's imagine : let two AlphaGo machines play each other Go games. More games. More time allowed... Folks : it becomes IMO so abysmal. Where will it stops ? I literally shiver in awe. I believe this could be radically extreme disruptive technology. Keep in mind, AlphaGo invented moves it never observed before. Keep in mind, it can learn quite some different games, just by being exposed to samples. Wooooooaaaaaa. Impressed, concerned, exited, I am. Z.
Sorry, but I'm a mathematician.
I didn't realize that is something for which people apologize.
It doesn't require creativity. It just requires a deep understanding of the game. The game itself has a very restrictive set of rules. It isn't creative at all. BTW, game playing isn't AI at all. Computers are good at playing games with a restrictive set of rules. In fact that is the one thing they are best at: computers LOVE rules and require them to perform any task.
The comparison between games that have a restrictive set of rules and those that do not is the wrong comparison to be making.
The reason WHY computers tend to do well at game with restrictive sets of rules is because they're able to take those rules and fashion them into a set of all (or at least a significant portion of) possible positions that are going to come up in the game that they're playing.
That's not a valid solution to Go because the number of possible positions, even in the context of an individual game, is too vast to be able to use a brute force approach. Despite having a "restrictive rule set", just like chess does, Go does not allow for the same kind of "AI" (which isn't really AI at all) to solve it. The comparison here should be between games where brute force is possible and games where it isn't. The size of the rule book is meaningless.
So no, just because you can write the rules on one page doesn't mean that you can just unleash a supercomputer at it and suddenly it beats grandmasters. That happened in chess. What's happening here is much more profound.
We still argue over the definition of intelligence.
I predict people will still argue over it even after it's been duplicated on a computer.
Heads Up limit has been beat, heads up no limit probably not till the end of the year, maybe slightly longer.
Not really. When looked at naively, the problem space seems much larger in image recognition, but there are algorithms to drastically simplify things in IR, while such things are very scarce on the ground for Go.
Go is an interesting game for this approach. It is thin, which mean that the moves and pieces are the same and don't do a lot. It is wide so calculating everything from scratch is essentially not doable...
Chess got to be good enough by essentially matching GM search depth, by intelligently narrowing the search tree. And either capitalizing or avoiding tactical issues, within that depth, and if there are no tactical issues if there are a collection of moves that are left, make the ones that follow a distinct set of priorities. But total search depth and search thinning have proved by far to be the most valuable of contributions. With intelligently managing the priorities a distant but important second.
Tic Tac Toe is done by being able to search to the game ending. As well as connect four.
Checkers is solved by being able to connect opening books with endgame table bases making the search limited enough to be doable by current computing power..
Go is interesting because the nature of the game is one of knowing the right moves to make in given positions. And the pieces on the board don't move. There are going to be two kinds of moves tactical calculating moves which can be iterated (And not as deeply as chess), and those that are correct based on "knowledge" without iterative proof. The best human players are going to be the ones that can do both and not just one. The trick is to first split up the two types of moves. And with the knowledge moves the AI mechanism essentially stole the knowledge from games by human players. And then tested and retested that knowledge by playing against itself. There is also some expertise in manipulating the games to handle board rotation, and location of plays depending on position of the plays in relation to the edge of the board. There is also the reactive vs active moves and when the board requires one type or another.
I am entirely unconvinced that this methodology is going to be universally useful at unknown problems. It has a high level of specialization that must be known by the program in advance to even get to the big data stealing of correct moves (knowledge). What will be useful is that the technique will be added to our knowledge of how to attack certain types of problems, and will help in creation of certain expert systems. What they have actually solved is how to play go as well as humans can, and very little more than that. And not all problems are going to be solvable this way, and even if the can, it is not always going to be the easiest, most elegant, or provide truth.
What is still not likely to be convincingly known is "truth". Chess and Go even beating the best human players still doesn't know the truth in every situation. It is believe that in Chess that the truth always leads to a "draw". In Go, it is believed that Black should always win. We are a long long away from demonstrating, much less proving either case. And may be impossible as the iterative requirement may be simply too high.
What will be interesting is when we can develop AI that can understand and break down a problem into it's meta components and rules, and discover knowledge by itself. You know like some humans seem to be able to do.
You just use intuition to skip over moves that might not be worth your time, but you still consider them.
I don't think so. Out of a hundred+ moves, a good player may consider a dozen or so, but the majority isn't even looked at. The patterns of the stones already on the board, guides the brain directly to a bunch of candidate moves.
Exactly my point.
Lee Sedol is a loser, I like people who don't lose to supercomputers!
The surprise here is that learning from the game endings of internet GO players and somewhat informed computer vs computer games is enough to train an evaluation function with the predictive power to beat the world champion.
It is surprising. I'd go so far as to say "stunning". This kind of ML is really, really fallible, in exactly the areas that humans do well. I'm kinda baffled.
I darkly suspect that it means that humans aren't really very good at Go. The combinatorial explosion is so fast that the vast majority of moves don't get any consideration at all. Humans apply a well-trained intuition, but there's reason to think that good moves are completely ignored.
In chess, the best computers play slightly better than the best humans; it's as though we're near perfect play even if we can't actually work out all of the routes for all games. In Go, it sounds as if in the near future humans won't even be in consideration; you'd no more ask a human to play Go against a computer than you'd ask one to race an automobile.
Which is to say: perhaps the explanation is that the computer still isn't very good. It's just better than the extremely-fallible human.
I'd really like to see a rundown of what moves the program its self thought were clutch. And how it predicted the game would have played out had it acted differently.
Playing a game, and doing it well, requires real creativity. Arguably a lot more than science, actually. When you study science, all you're doing is discovering information already out there - water had its properties and was built by molecules long before it was classified as H2O, and nothing changed after.
This is a ridiculous position. There's a long chain of creative thought that led to our knowledge of chemistry. AI has a long, LONG, way to go before it's capable of replicating this achievement. Could you even begin to write a program that ponders about the nature of the physical world, performs experiments, and comes up with chemistry?
Doing well at Go can't be calculated cold and hard - much of it is subjective, and that's what makes this discovery so important. The computer didn't win by just repeating the same patterns or evaluations over and over, but actually learned from each game and was able to apply that to the future. That's the start of self learning AI.
What is objective about Go are the unambiguous (as used by computer Go) and simple rules and the binary win/loss results, along with the underlying game tree that has a theoretically perfect solution. While learning to play Go well with neural networks is impressive, it's still very far from general intelligence.
" Thou shalt not make a machine in the likeness of a man's mind"- Orange Catholic Bible- Dune Series "The target of the Jihad was a machine-attitude as much as the machines," Leto said. "Humans had set those machines to usurp our sense of beauty, our necessary selfdom out of which we make living judgments. Naturally, the machines were destroyed."[6]-God Emperor of Dune "Man may not be replaced."-The Butlerian Jihad a.k.a. The Great Revolt use of technology trains humans to think like machines. The problem is that machines are deterministic; thus, training people to be machines is self-limiting. Herbert seemed to think that to be human is to be essentially 'open-ended', capable of undiscovered, indeterminate evolution, both personally and as a species.- Heidegger's thesis.- Over to you human race.