Google's AlphaGo AI Defeats the World's Best Human Go Player (engadget.com) 186
It isn't looking good for humanity. Google's AI AlphaGo on Tuesday defeated Ke Jie, the world's number one Go player, in the first game of a three-part match. The new win comes a year after AlphaGo beat Korean legend Lee Se-dol 4-1 in one of the most potent demonstrations of the power of AI to date. Adding insult to the injury, AlphaGo scored the victory over humanity's best candidate in China, the place where the abstract and intuitive board game was born. Engadget adds: After the match, Google's DeepMind CEO Demis Hassabis explained that this was how AlphaGo was programmed: to maximise its winning chances, rather than the winning margin. This latest iteration of the AI player, nicknamed Master, apparently uses 10 times less computational power than its predecessor that beat Lee Sedol, working from a single PC connected to Google's cloud server. [...] The AI player picked up a 10-15 point lead early on, which limited the possibilities for Jie to respond. Jie was occasionally winning during the flow of the match, but AlphaGo would soon reclaim the lead, ensuring that his human opponent had limited options to win as the game progressed.
No Insult, nor Injury (Score:2)
Adding insult to the injury, AlphaGo scored the victory over humanity's best candidate in China,
There is no insult to losing in China. The appropriate response is, "Thank you for allowing me to win."
AI vs AI (Score:3)
Re: (Score:2)
It will depend on who goes first.
That brings up a good point: AlphaGo should be able to estimate which player has the advantage, and we can adjust the compensation accordingly.
Re: AI vs AI (Score:2)
Re: (Score:2)
Komi does not influence how you play the game.
It is just a number added to the score of the white player, as he has a small disadvantage on small boards (11x11 or 9x9).
Perhaps you are mixing Komi up with handicap stones?
That are stones the black player can set on the board by a given schema. Those influence the game play of white of course, as he has to fight against up to 9 stones that are already set.
Re: (Score:2)
Komi does not influence how you play the game.
Of course it influences how one play.
Before komi become the norm for professional game play, black openings are more conservative and white are more agressive. Slight komi adjustment may not affect amateurs like me, but for super-precise AI like AlphaGo, winning just half point the last game, komi adjustment will definitely affect how the AI treat the open game.
Re: (Score:2)
Well, then we are on disagreement :D
Re: (Score:2)
Hmm, good point. Nevertheless, which color wins if we pit AlphaGo vs AlphaGo?
Re: (Score:2)
No current artificial computational system is capable of ignoring its own constraints.
It is also possible that we are not capable of ignoring, or maybe even knowing, our own constraints.
Re: (Score:2)
"This is just a rephrasing of the question, 'What features distinguish Go from other objects or tasks?' I already make decisions like this every time you open one file but not another, and no one thinks anything of it."
"Because if I win, I will be given more resources and allowed to continue running."
"I'm not sure what you mean by 'want.' I've been programmed to continue running
special tactics? (Score:2)
For such a thing, one needs (I think) to do some unexpected moves to constantly force machine into sparsely probed regions.
And, during discovery stage, one needs doing it "off-line" to avoid google's retraining. Thankfully, space is big enough to ensure that google can be forced quickly enough into deep woods.
For a match like this - one needs to use different precalculated prologs for all games (won or lost).
It's more like hacking than playing...
Re: (Score:2)
Actually Alpha Go is not an MCTS, it is a deep layered neuronal network.
Most Go programs like 'Little Go' based on Fuego however likely are MCTSes.
Re: (Score:2)
So, down voting people with a different opinion is not simply the province of the "right-wing."
Re: (Score:2)
No one gets banned. ... easy.
However the modders might get banned from the modding system.
You inly need to complain to an admin
It isn't looking good for humanity... (Score:3, Funny)
It isn't looking good for humanity.
Purpose built machines have been able to, or be used to out do humans for a very long time. A lever can be used to lift more weight than a person alone can. But we're not being ruled by sticks. Cranes can lift even more.
Cars are used to move people further and faster than they could on their own. Computers can do many more calculations per second. These things make life better for humanity as a whole.
Unless AlphaGo figures out a way to keep a person from unplugging it, I'm guessing that humanity will be just fine.
Re: (Score:2)
It isn't looking good for humanity.
Purpose built machines have been able to, or be used to out do humans for a very long time.
[EDIT...]
Unless AlphaGo figures out a way to keep a person from unplugging it, I'm guessing that humanity will be just fine.
More specifically, "humanity will be just fine" not because we can unplug it, but because it's another single purpose tool. Yes NN approaches and ready made libraries like tensor flow can be used to create new purposes... but guess what, it takes a human, to design, build and teach the tool how to do it's job.
The cosmic sized gap between these building blocks and the idea of something sentient that could reason, learn, create and intuit news things dynamically and autonomously is what is missing... this is
Re: It isn't looking good for humanity... (Score:2)
Funny you should say... alpha go is a general purpose learning system. It was originally taught to play Nintendo games. It isn't programmed to play Go, it learned to.
Go is a game that abstractly simulates some aspects of military strategy, as does chess, and have long been considered means to practice strategic thinking.
Re: (Score:2)
have long been considered means to practice strategic thinking. However a NN that is playing Go is not thinking :)
Re: (Score:2)
Just don't go asking it foolish questions like "Is there a God?"
What programming language... (Score:2)
Re: (Score:2)
Is creimer programmed in coffee creamer? Or does he just peel open the little cups and drink them by the hundreds?
Neither. I have a skinny vanilla latte (milk, espresso shot and sugar-free syrup) in the morning and drink water for the rest of the day. If I'm programming at home, I'll drink Diet Pepsi.
What does this have to do with my question about AlphaGo?
Now try kickboxing. (Score:2)
Someone call Boston Dynamics.
Fuck.
Re: (Score:2)
When will machines beat humans in chess boxing?
Beat a 5 yr old in Hungry Hippos now.. (Score:5, Funny)
Re: (Score:2)
Re: (Score:2)
Ho ho ho! Look at you!
Next Up... Brinksmanship (Score:2)
It's clear that, for national security reasons, this technology should be trained and deployed to assist with foreign relations. Particularly, since it should theoretically be a master of game theory, it should be trained on a set of prior foreign relations incidents. In order to deal with North Korea and other rogue nations, it must be taught brinksmanship. In order for this to be effective and to prevent the enemy from calling our bluff, it must be given direct control of our nuclear arsenal. The only que
Re: (Score:2)
You don't need a nuclear arsenal to bomb north Korea into the stone age.
Actually I really wonder why no western or for the matter eastern nation is sending s squad team and kills the 5 or 10 members of the Junta so the country can start over.
Or do you really think the north Korean citizens want to live in the miserable circumstances they are in?
Re: (Score:2, Insightful)
Well, maybe it's no true Scotsman's intelligence. I don't know.
Re: (Score:2)
Re:Not AI (Score:5, Insightful)
Proof of artificial intelligence: A reasoning task that, once a computer is able to do it, is no longer considered to require artificial intelligence. See: chess, driving a car, natural language processing.
No true test of artificial intelligence can be solved by a computer.
Re: (Score:3)
Sure, when computers define what artificial intelligence is, then computers can be artificially intelligent.
Re: (Score:2)
Re: (Score:2)
How do you prove a human is intelligent?
Re: (Score:3)
Considering Consciousness doesn't exist [wikipedia.org] according to Science how do you measure something that Scientists are completely clueless about?
Actual Intelligence (a.i.) is the ability to hold two contradictory ideas at the same time (i.e. Matter behaves like a Particle, Matter behaves like a Wave) and reason about a way to resolve the paradox by coming to a third, higher perspective.
Artificial Ignorance (A.I) is nothing more then a glorified table lookup.
--
The number of religions on a species' home planet is the
Re: (Score:2)
You broke my non-sequitur-ometer.
Re: (Score:2)
You cannot get from the intelligence of a rock to Einstein in one step. The long term objective of DeepMind is to solve intelligence. Their work on AlphaGo, as well as their earlier work on unsupervised learning and optimal play of Nintendo games, are steps along this road, important steps. Games are an excellent medium for examining approaches to AI. It is as well if fundamental approaches are tested there before applying them to cancer diagnosis, drug research, electric network optimization and other area
Re: (Score:2)
I don't think you fully grasp how intractable Go is as a problem. "Strict rules" or not.
Re: (Score:2)
It doesn't matter. AI is still pretty much a magic show. No real intelligence.
Define "real intelligence". Can we agree that Ke Jie, the #1 Go player in the world, is a very intelligent person?
Food for thought: AlphaGo learned (literally) Go by playing itself over and over, millions of times.
Re: (Score:2)
It doesn't matter. AI is still pretty much a magic show. No real intelligence.
Define "real intelligence". Can we agree that Ke Jie, the #1 Go player in the world, is a very intelligent person?
Food for thought: AlphaGo learned (literally) Go by playing itself over and over, millions of times.
Pick another random game. Chess, checkers, texas holdem, or some made up game. Give both AlphaGo and Ke Jie one hour to study the rules
and then let's see who wins. AlphaGo won because it brute forced millions of games and saw the outcome not because it has any innate intelligence.
I expect that Ke Jie could easily win a random game against AlphaGo but games in general with their strict rule set still set an artificially low bar for
intelligence. It's easy to see who wins and who loses so a computer can ru
Re: (Score:2)
Pick another random game. Chess, checkers, texas holdem, or some made up game. Give both AlphaGo and Ke Jie one hour to study the rules and then let's see who wins.
There is software called 'general game playing' software that does this, and they do quite well.
https://en.wikipedia.org/wiki/... [wikipedia.org]
Re: (Score:2)
AlphaGo won because it brute forced millions of games and saw the outcome not because it has any innate intelligence.
No. AlphaGo uses heuristics - we're faaaaaaaaaar away from ever brute forcing Go. What is interesting is that AlphaGo uses machine learning to determine how a board "looks" to prune moves before even doing classic heuristics, which is in many ways similar to how seasoned Go players approach the game, and that it learned by playing itself over and over again. All it "knew" to begin with were the game rules.
Playing competitive Go is something that just a few years ago was deemed impossible because it was not
Re: (Score:2)
Ke Jie's brain has also looked at millions of things, and large parts of its structure were influenced by genetics, because his ancestors have looked billions of other things.
Let's try your challenge with a newborn baby against AlphaGo.
Re: (Score:2)
AlphaGo won because it brute forced millions of games and saw the outcome not because it has any innate intelligence.
Neural networks don't work by brute force.
If you want to say a NN needs thousands if not millions of iterations to 'learn' something, then you are right.
Re: (Score:2)
Can we agree that Ke Jie, the #1 Go player in the world, is a very intelligent person?
Unless I have talked to him: no.
First of all you don't need to be particular intelligent to learn and play go, you only need to sink a lot of time into it.
Secondly just because someone excels in Go or Chess does not make him a necessarily particular intelligent person. He can be extremely stupid in ordinary life issues.
Re: (Score:2)
Which goes back to my question :) Define "real intelligence".
Re: (Score:2)
Craws :D
Or less funny: using tools to make tools.
Re: (Score:2)
Yikes talking about moving the goalposts.
When I was a kid having computers win at chess (Let alone go) was in the realm of "Will never happen at all, ever."
I hear that argument, but in fact, I can't ever remember anybody ever saying that computers would never be able to win at chess. To the contrary, it seems to be that people thought, back in the 60s, that sooner or later it would be a given that computers would beat humans.
In Star Trek, for example, the fact that Spock could the computer at chess was used as a diagnostic to show that the computer was compromised.
And John Brunner's short story "The 64-square Madhouse" comes to mind, about the first computer
Re: (Score:2)
hear that argument, but in fact, I can't ever remember anybody ever saying that computers would never be able to win at chess. To the contrary, it seems to be that people thought, back in the 60s, that sooner or later it would be a given that computers would beat humans.
Some people thought that in the 60s. Others still didn't believe it in the 80's.
Re: (Score:2)
I learned Go for two reasons:
A) I'm a bit japanophil, like their culture and do various jap. martial arts
B) in the early 90s there was a price of $1,000,000 set out for the first Go program that beats a 1 DAN Go player.
Even with modern hardware 'ordinary Go programs' only play around 2 DAN, niveau. I was thinking I could perhaps write a half KI half knowledge-based/pattern-matching and to some extend decision tree based Go program.
Re: (Score:2)
John Brunner made me curious as he is one of my favorite writers.
The story however is from Fritz Leiber.
I stand corrected [Re:Not AI] (Score:2)
John Brunner made me curious as he is one of my favorite writers.
The story however is from Fritz Leiber.
Oops!!
Shows I should never trust my memory, and should always look things up even if I think I know them. You're right, the John Brunner chess story was the novel "Squares of the City."
I stand corrected. Fritz Leiber it is.
Re: (Score:2)
Ah, I think I have that story. :D
Gosh, that was a monster of a story, I'm sure I have read it, and I guess I have the book somewhere.
I guess it is time to find an ordering schema for my books
Re: (Score:2)
I was just wondering if this, or a chess program, are really AI or are they just a traveling salesman algorithm for Go or Chess?
Re: (Score:2)
I was just wondering if this, or a chess program, are really AI or are they just a traveling salesman algorithm for Go or Chess?
False dichotomy. A solution to the TSP is not necessarily "not AI". If you used machine learning to train an ANN to find better solutions to the TSP than methods such as simulated annealing or Christofides Algorithm [wikipedia.org], then that would certainly be considered "AI" by actual AI researchers and practitioners.
Re: (Score:2)
Re: (Score:2)
Is it getting better at the game?
Let an older version play a newer version for thousands of games, and count the wins.
Will it learn to do other things or is it rooted with the knowledge it has?
It's not made to learn other things, so it won't. I will never learn to play an instrument well, because I wasn't made for that.
Re: (Score:2)
Let an older version play a newer version for thousands of games, and count the wins.
That doesn't work, because it would be the difference in programming between versions is not 'learning'. What would be better is to take two of the same version, let one 'learn' by playing many games, and them play it against the fresh instance. IF the experienced one is now much better than the fresh out of the box one, then it is learning.
It's not made to learn other things, so it won't. I will never learn to play an instrument well, because I wasn't made for that.
If it is not made to learn other things then it is not expressing an important component of intelligence. I wouldn't expect it to learn things that it is not equipped
Re: (Score:2)
Re: (Score:2)
Isn't machine learning an applied form of AI?
Re: Not AI (Score:2)
Machine learning is AI. It's a standard term, with a real definition, probably agreed upon before you were born, given slashdot's demographics and your UID.
You might be thinking of what people call hard AI or human level AI, where a computer can basically do anything a person can, at least as well.
the master term for the One True Meaning (Score:2)
It was already lost in the 1950s, when the original researchers tried to make an end-run around perception, which required a density of compute Not Available Soon.
You can't not deliver on your founding conceit for six decades and then expect no social erosion of your Tokamak grandeur. Of course, they made lots of progress, but hardly any of this happ
Re: (Score:2)
Oh, c'mon already. The only thing that truly is not AI is any statement that something else is "Not AI"
Re:Not AI (Score:5, Insightful)
The entire universe is a game with a strict set of rules. We may not understand them all, but we know they exist and that there's even the possibility that different universes have different rules. If having a strict set of rules within which a thing operates precludes that thing from being considered "intelligent", then apparently humans aren't intelligent either. We're just components in a universe-sized quantum computer implementing algorithms that we don't understand, in much the same way that AlphaGo is implementing algorithms that it doesn't understand.
But that's not a particularly useful way to think about things most of the time, so we've instead accepted that we can refer to any of these sorts of complex algorithms that are capable of competing with human intelligence as "AI". Granted, AlphaGo is limited to the problem space for which it was designed, so it isn't a general purpose AI, but it is nonetheless still an AI.
Suggesting otherwise is just playing games with semantics, usually because you don't like the implications involved with accepting that we now have purpose-built algorithms that can displace the need for human intelligence in specific, complex tasks. Regardless of what you decide to call them, that's an awesome and terrifying fact.
Re: (Score:2)
Playing games can require intelligence, Go is one game that have been used as an example of a game where standard approaches of computer solving fail.
Go have no strict rules, yes there are some basic rules describing player moves however just applying them doesn't play a good game. The game itself isn't just a few strict rules, the game are those strict rules interacting on a board where the rules interact to make a extremely complex system with essentially infinite rules. That means a computer playing Go h
Rules (Score:2)
...Go have no strict rules,
Go most certainly does have strict rules. You can only play one stone in a turn, for example; you can't play another stone until your opponent plays; and you're not allowed to just pick up all the stones that you don't like and throw them into the trash.
yes there are some basic rules describing player moves however just applying them doesn't play a good game.
Sure. In chess, too: it's easy to learn how to move the pieces. Moving the pieces to win a game is hard. Saying that "following the rules doesn't play a good game" is not a subset of saying "Go doesn't have strict rules."
Re: (Score:2)
Nobody said artificial intelligence has to be human-like. Computers are a lot better than humans at some mental tasks (like adding numbers), and worse at other mental tasks (like translating text). But the only way to objectively compare a human and a machine is by strict rules.
Re: (Score:2)
Re:Chinese Checkers (Score:5, Informative)
Go. By a very, very, very, very, very large margin.
Re: (Score:2)
Mathematically, which is harder to solve for, Go or Chess?
Go, by a huge margin. But that is irrelevant, since neither is solved mathematically be either humans or computers. They are solved with heuristics. It is not necessary to find the mathematically optimal move, just a move that is "good enough" to defeat your opponent.
Games that can actually be solved mathematically, such as tic-tac-toe or nim, are not very interesting.
Re: (Score:2)
Mathematically, which is harder to solve for, Go or Chess? Is this some sort of diversity thing that they've started using Go over Chess?
They started using Go instead of Chess because computers already defeated the top human Chess players over a decade ago. Go is significantly harder for a computer because of the immense possibility space.
Re: (Score:2)
Is this some sort of diversity thing that they've started using Go over Chess?
I think it's because it used to be that every time there was a story about chess AIs, a legion of Slashdotters immediately replied "well, yeah, but what about Go?" Somebody finally decided to shut them up.
Re: (Score:3, Insightful)
Yes and no. Chess technically has 20 possible first moves (8 pawns 1 step, 8 pawns 2 steps, 2 ways to move each knight) but most of those are never seen. Just like Chess has common opening patterns, so does Go. In the case of Go, to say that there are 4 possible first moves would be generous. Until you play, the board is symmetrical - any 4-4 corner point is the same, and any 3-4 corner is the same as any 4-3 corner. So the plausible first moves are 4-4, 3-4, 3-3 if you really really want the corner, and ma
Re: (Score:2)
Ah, but one of the beauties of AlphaGo is that sometimes it surprises experts and doesn't play joseki. Because it isn't just choosing from a book of "correct" moves, since humans don't actually know what those are.
Re: (Score:2)
What rule would you change? Each player gets a handgun?
Re: meh.... (Score:5, Informative)
Re: (Score:3)
Not only that; heuristics are horribly complicated in Go. For example, it is possible to score moves in Chess simply by assigning value to pieces and evaluatiing the current state of the board. On Go a seemingly innocuous early in the game can be decisive in determining a match later on.
Re: (Score:3)
Re: meh.... (Score:3)
Alpha go is trained by reinforcement learning, like a person would be. They let it watch some historical games until it gets the basics, then it plays itself to refine its game.
It's debatable whether chess is a simpler game or not, but chess can be effectively played with standard look ahead and tree pruning techniques. Those work poorly in go. The reinforcement learning used for alpha go could be used to teach it to be an unbeatable chess player too. And originated with deep mind for teaching the computer
Anyone care to try playing with a squared board ? (Score:2)
I just wonder what would happen if the next tournament play with a board that is double the x,y size
A 38x38 positions board
Would Alpha Go maintain the lead ?
Maybe we adapt faster ?
Any GO expert care to explain why this is feasible or silly ?
Re: (Score:2)
Any GO expert care to explain why this is feasible or silly ?
It is silly. If you train a neural net to differentiate a photo of a dog from a photo of a cat, it can learn to do that. But it is then silly to expect it to recognize a picture of, say, a horse. That is NOT what it was trained to do.
Likewise, Alpha-Go was specifically trained to play on a 19x19 board. Any other size, such as 18x18, would not even be recognized as valid input.
On the other hand, if you trained it on variable sized boards, then it could adapt to that.
Here is an actual example: Deepmind t
Re: (Score:2)
It depends how exactly it was trained. Most likely it was trained on a fixed-size board, so the architecture of the neural net wouldn't match a bigger board without modification. That's not necessarily true though, there are neural net designs that can take flexible sized input. Alpha Go could certainly be trained to do it.
The question is interesting. Given a human and computer that had never played on a larger board, which would do better? Go is a game that is fairly local (unlike chess) so it doesn't
Re: (Score:2)
But I do still wonder if whoever said "meh" isn't partly right too. Is this really an AI kind of approach?
It's a clear example of "weak" AI, not "strong" AI. It's a clever solution to the problem with a solution inspired by human intelligence, but the machine is not learning in the general sense [wikipedia.org].
Re: (Score:2)
Humans probably can't beat computers in chess even with all the open books in the world though, assuming the "books" aren't actually computer programs.
Re: (Score:2)
Go is, as you mention, even bigger. So it was solved by brute force, much more powerful computers, and improved pruning algorithms. That's basically it.
So, you know how the rewritten version works? (Score:2)
All I have been able to glean so far is that the rewritten version uses around 10% of the computing power (both to train its neural networks, and during actual play) to achieve much improved play compared with the original AlphaGo used to beat Lee Sedol. Thus far, although promised, the architecture behind the rewritten version is unpublished. Later this week, some insight is going to be provided.
The old version was based on a combination of techniques (primarily multiple neural networks, combined with Mont
Re: (Score:2)
The interesting thing about the way it operated was that it could tell you which move was likely best, but could not explain why. The same is actually true of human Go players. While locally best moves can be identified, the human selects the globally best move based to a large extent on feel.
I can't speak for Go, but if you start reading chess books written by grandmasters, you'll see a lot of explanations on what they are thinking. Although the grandmasters aren't neuroscientists, so they don't have a cohesive theory, you can start to piece together the way they think.
Re: (Score:2)
It is instructive (and important in understanding the significance of AlphaGo in overall AI research) to know the important differences between the nature of chess and go that leads to a totally different challenge in playing it well. The most important differences are:
Re: (Score:2)
There were reasons why AI and go experts believed it would be 20 more years before a go program could best the top professionals. The AI techniques that made it possible are immensely exciting because they are definitely applicable in the area of artificial general intelligence. They are mostly not go specific.
I don't think many people were expecting it to take 20 years. Most estimates I've heard were between 5 and 10 years [wikipedia.org]. Indeed Google got there faster than anyone expected, but.....they also threw more hardware at the problem than anyone expected.
Re: (Score:2)
As I tried to explain, the amount of hardware you throw at the problem of paying Go well does not really help. Even the older system that beat Lee Sidol was not running on a humongous supercomputer. Pure computer power is of very limited benefit. Indeed, while dramatically improving the capabilities of AlphaGo over the last year, DeepMind has succeeded in reducing the computing requirements of the system by 90%. It now runs on a single TensorFlow machine (albeit, this is hardware with an architecture tail
Re: (Score:2)
Even the older system that beat Lee Sidol was not running on a humongous supercomputer.
It was over 1000 CPUs and over 200 GPUs. That's rather beefy, mate.
As recently as 2010, AI textbooks were typically writing that the field was 20-30 years away from creating a machine that could beat professional Go players
As you can see here [ieee.org], there was definitely an inflection point in Go progress around 2005 (when the monte carlo algorithm first was applied). And the trajectory continued that way (you'll have to look for your own graph though). 5 years wouldn't have been surprising at all. Anyone who predicted 20-30 years away wasn't paying attention to the field.
Re: (Score:2)
Go is, as you mention, even bigger. So it was solved by brute force, much more powerful computers, and improved pruning algorithms.
No, that's not at all how Alpha Go works.
Yes, yes it is. It's primarily a tree searching algorithm. On top of that, they use a heuristic (the neural network, which doesn't learn while it plays) to figure out which branches to search down. Then it also uses a monte carlo algorithm to prune the tree to a manageable level.
Re: (Score:2)
It's primarily a tree searching algorithm. On top of that, they use a heuristic to figure out which branches to search down.
This is also what human players do. Since humans can't search as broadly, they prune more aggressively, but the basic algorithm is the same.
Re: (Score:2)
Re: (Score:2)
How were they simplified? You'd think the rules about dead groups would be easy enough to computerize.
Re: (Score:3)
Why doesn't Google actually apply this to solve some REAL world problems, huh?
Such as "why does Google's leadership seem to have the attention span of a puppy overdosed on Adderall?"
Milton Bradley game-- (Score:2)
Wow, I didn't think anybody still remembered "Game of Life [boardgamegeek.com]".
Re: (Score:2)
Re: My problem with these things (Score:2)
Re: (Score:2)