Google Go-Playing A.I. Retires To Focus On Energy Conservation And Medicine (engadget.com) 127
After "narrowly" beating the world's top Go player, what's left for Google's AlphaGo AI? Engadget reports:
Now that it has nothing left to prove, the AI is hanging up its boots and leaving the world of competitive Go behind. AlphaGo's developers from Google-owned DeepMind will now focus on creating advanced general algorithms to help scientists find elusive cures for diseases, conjure up a way to dramatically reduce energy consumption and invent new revolutionary materials. Before they leave Go behind completely, though, they plan to publish one more paper later this year to reveal how they tweaked the AI to prepare it for the matches against Ke Jie. They're also developing a tool that would show how AlphaGo would respond to a particular situation on the Go board with help from the world's number one player. While you'll have to wait a while for those two, you'll soon be able to watch 50 games AlphaGo played against itself when it was training
The first ten games that AlphaGo played against itself are already online. Shi Yue, 9 Dan Professional and World Champion, described them as "Like nothing I've ever seen before -- they're how I imagine games from far in the future." Google announced that this week's competition "has been the highest possible pinnacle for AlphaGo as a competitive program. For that reason, the Future of Go Summit is our final match event with AlphaGo... We hope that the story of AlphaGo is just the beginning."
The first ten games that AlphaGo played against itself are already online. Shi Yue, 9 Dan Professional and World Champion, described them as "Like nothing I've ever seen before -- they're how I imagine games from far in the future." Google announced that this week's competition "has been the highest possible pinnacle for AlphaGo as a competitive program. For that reason, the Future of Go Summit is our final match event with AlphaGo... We hope that the story of AlphaGo is just the beginning."
Re:Not enought balls for a rematch? (Score:5, Informative)
At least IBM had the balls to go again.
I don't think Ke Jie has the balls to do it again. It would be utterly pointless too, because the AI will keep improving much quicker than human players.
“I feel like his game is more and more like the ‘Go god’. Really, it is brilliant,” he said.
Ke vowed never again to subject himself to the “horrible experience”.
https://www.theguardian.com/te... [theguardian.com]
Re:Not enought balls for a rematch? (Score:5, Insightful)
I agree about the gullibility of some people, just disagree about who we are talking about.
The best chess AI programs are already rated 200+ points above the top-ranked human players. No master today disputes that a machine would kick their ass, even though masters train *daily* against AI programs.
Bill Gates recently acknowledged that their are different kinds of intelligence. We are starting to see how machine intelligence is of a kind we don't understand and won't match. How long we can keep an edge in other types of intelligence is the only point left to debate.
Re: (Score:2)
I would like to see human versus AI in essay reading comprehension.I know humans will beat AI in writing essays for a hundred years .. but I think AI beating humans in reading comprehension is in the possibility range. Make it like the SAT, or actually give it the SAT comprehension test. It might even be able to critique or point out logical fallacies.
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Indeed. In chess we haven't had any serious man/machine matches since 2006 where Deep Fritz defeated GM Kramnik 4-2. And in the year before that, Hydra beat GM Michael Adams 5.5-0.5. Modern versions of Stockfish and Komodo would wipe the floor with these old programs, and would totally humiliate any human grandmaster.
DeepMind's approach to Go is still relatively immature, and others will surely adopt and improve their ideas and develop even stronger machines.
Re: (Score:1)
Can you define what the intelligence humans use to play chess is? How about go? What is "intuition"? Is it pattern matching? When players train against an AI, what skill set are they training? Their 'intelligence'?
Re: (Score:3)
Intelligence according to a dictionary definition is "the ability to acquire and apply knowledge and skills". Surely the machine is doing that, and so are Chess computers.
But Go is not anywhere near that point
Right now, it's already beyond the point where it can beat any human. And we've only just started. It's already stronger than Deep Blue was when it beat Kasparov. People also cried when IBM retired Deep Blue, but by today's standards, Deep Blue is a mediocre program. The same will happen with Go.
Re: (Score:1)
Intelligence according to a dictionary definition is "the ability to acquire and apply knowledge and skills". Surely the machine is doing that, and so are Chess computers.
Actually, it is not. This approach does not work with "knowledge" or "skill". It works in an automated way which involves neither. It could be printed in a book and executed purely mechanically. Or would you claim a book can be "intelligent"?
Re: (Score:3)
Or would you claim a book can be "intelligent"?
AlphaGo has encapsulated knowledge about good and bad Go positions, and can apply that knowledge, not only in identical situations, but also in completely novel situations that resemble similar patterns.
Can you point me to a book that can apply the knowledge contained in it ?
Re: (Score:2)
Actually, it is not. This approach does not work with "knowledge" or "skill". It works in an automated way which involves neither. It could be printed in a book and executed purely mechanically. Or would you claim a book can be "intelligent"?
And by book you mean Chinese room [wikipedia.org], right? And you do realize that AlphaGo doesn't play by using a deep set of lookup tables, right? Much like a human player, it has a way of scoring possible play that doesn't involve canned openings or evaluating every combination N moves ahead, but instead by evaluating what play seems weak or strong.
There is no useful definition of "knowledge" or "skill" that excludes what AlphaGo does, unless you assume your own conclusions about human intelligence. There's certainly
Re: (Score:2)
Intelligence according to a dictionary definition is "the ability to acquire and apply knowledge and skills". Surely the machine is doing that, and so are Chess computers.
No, that's a naive definition, good enough for a dictionary, I guess.
If you want to understand the issue, you need to understand the difference between weak and strong AI. At least read the Wikipedia article, it will give you a good overview of the topic [wikipedia.org]. AlphaGo is weak AI, not even the creators claim it is strong AI.
Re: (Score:3, Insightful)
f you want to understand the issue, you need to understand the difference between weak and strong AI.
I know the difference, but weak AI still has an "I" in it. Intelligence is a broad and fuzzy concept, with many different elements. The computer can now capture some of these elements, just like a chicken or a dog can capture some, and in a growing number of cases, the computer can do it better than us. Obviously, we're not even close to making a computer that can capture the full range, but I don't believe there's a fundamental gap, just like there's no fundamental gap between a chicken and a human brain.
Re: Not enought balls for a rematch? (Score:1)
Better. Exactly like a calculator. That's it.
Re: (Score:3)
I know the difference, but weak AI still has an "I" in it.
Obviously you didn't understand what you read. There is a fundamental gap between AlphaGo and Chicken brain, much greater than between a human and a cockroach.
I tried to give you clues, but you closed your mind and continue to spout nonsense because of your lack of understanding.
Re: (Score:3)
I tried to give you clues, but you closed your mind and continue to spout nonsense because of your lack of understanding.
I'm afraid you're a victim of the Dunning Kruger effect.
Re: (Score:2)
You want it to be intelligent, so you close your eyes when people point out its not.
Re: (Score:2)
A word can have many definitions. Just agree on one and avoid confusion and equivocation. What definition are you using?
Are you using the research definition of strong versus weak AI o
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
We are starting to see how machine intelligence is of a kind we don't understand and won't match. How long we can keep an edge in other types of intelligence is the only point left to debate.
AlphaGo is provably inferior to human intellect (the brief proof is that it's not a Turing machine). It is incapable of self-introspection: it will never understand that it was playing Go. It doesn't know who Ke Jie is, it doesn't even know what a Go board looks like.
For us to have general AI (AlphaGo is weak AI [wikipedia.org]), we are going to need fundamental breakthroughs. The style of thinking that AlphaGo does won't take us there.
Re: (Score:2)
AlphaGo is provably inferior to human intellect (the brief proof is that it's not a Turing machine). It is incapable of self-introspection: it will never understand that it was playing Go. It doesn't know who Ke Jie is, it doesn't even know what a Go board looks like.
Of course not. It was never trained for that. It was trained for recognizing and judging Go patterns. The part of the brain of a human Go player that is responsible for a similar task also don't have self-introspection. That's a responsibility of a different part of the brain.
And human brains aren't Turing machines either. We can't even do trivial problems like factorization of million digit numbers.
Re: (Score:2)
Re: (Score:2)
You do realize that you don't know what a Go board looks like either, right? Human consciousness is far removed from raw sensory data. You have a software abstraction that represents the board, just like AlphaGo. Sure, your derives from sensory data, but it would be trivial these days to have the AI use an actual Go board, a camera, and a robotic arm to play - those are all solved problems, not interesting ones.
Also, you want a different term than "AI" if you want to mean consciousness. The "artificial"
Re: (Score:2)
Also, you want a different term than "AI" if you want to mean consciousness.
No, I want the term strong AI.
Who do you imagine thinks it will? Who do you imagine wants it to? Certainly not the researchers.
Maybe you don't, the researchers definitely don't.......but plenty of people on this forum do think it will.
Re: (Score:2)
No, I want the term strong AI.
Right, a term coined by philosophers. It's a bad term, discordant with what AI researchers research. "Machine consciousness", maybe?
but plenty of people on this forum do think it will.
Every time one of these stories is posted I see 100 people rushing to prove they're smart by posting "but this isn't real AI, it's just a trick", or something along those lines, and approximately 0 people posting otherwise.
Re: (Score:2)
but this isn't real AI, it's just a trick
What they don't realize is that real intelligence is also a trick.
Re: (Score:2)
when the carefully crafted illusion of "intelligence" (a blatant lie) still holds.
It's just a calculating machine. It is weak AI [wikipedia.org]. I used to tell people that it was not AI at all, but they got upset and argued, so now I just say it's not strong AI. No one can argue with that. It's not general intelligence.
Re: (Score:2)
What is more likely is that other people will figure out how Google did it (especially since Google is going to release a paper saying how they did it), and soon make engines that are better than AlphaGo.
I expect that will happen. And when it happens we may see machines that can beat even well-prepared human players.
when the carefully crafted illusion of "intelligence" (a blatant lie) still holds.
It's just a calculating machine. It is weak AI [wikipedia.org]. I used to tell people that it was not AI at all, but they got upset and argued, so now I just say it's not strong AI. No one can argue with that. It's not general intelligence.
Indeed. "weak AI" is the AI with no "intelligence" in it. Personally, I like to call it "automation". But you are right, many people do get upset when faced with the reality of things, not only in this question.
Weak AI is still very useful, especially as it is the only AI we have and some things that we thought would require intelligence turn out to actually do not. That makes them accessible to
Re:Not enought balls for a rematch? (Score:4, Interesting)
It's just a calculating machine. It is weak AI. I used to tell people that it was not AI at all, but they got upset and argued, so now I just say it's not strong AI. No one can argue with that. It's not general intelligence.
That depends on how broad you define general intelligence, you can certainly generalize "weak AI" without giving it any "strong AI" properties of consciousness, self-awareness, introspection or defining its own goals, ethics and morality to MacGyver-like levels of intelligence where it understands the physical and chemical properties of objects and can combine them into Rube Goldberg-like contraptions to achieve some goal even with multiple layers of gathering the resources, creating the components, tools and equipment themselves. Basically a "game" where the rules are the laws of nature as we know them, with no other restraints on the "moves".
For example say the goal is to start a fire. You don't have to teach the weak AI any methods, just basic chemistry of combustion, friction, optics and so on, the nature of the environment and available resources (earth atmosphere, sunlight, tinder, wood, glass, flint etc.) and it could work out that a bow drill, flint and steel or a lens to concentrate sunlight all may be possible ways to start a fire. With volume maybe it'll want to build a match factory or maybe it'll find some entirely novel way. And you can keep going from there until it can build every modern commodity, basically everything humans have produced is within the realm of weak AI. It just won't understand what anything is or why it's doing it, only that it satisfies some goal parameter(s) someone has set for it.
But that's okay, I think we're generally better at describing the result we want than the process anyway. If I want (fried) bacon and eggs it doesn't mean I know anything about owning a pig farm or raising chickens. I describe what I want to a chef AI that decides he needs raw bacon and eggs to a shopping AI, who purchases it. A stocking AI decides they need more bacon and eggs, which sets off a purchase/transport AI to resupply the store, that sets off a slaughterhouse AI to demand more pigs, that demand triggers the pig production AI to build more pig houses, buy more pig feed and raise more piglets and so on. It's weak AI all the way down. None of these AIs need to really understand "bacon and eggs" to do anything useful.
Re: (Score:2)
Regardless, Alphago is firmly in the weak AI category, and by itself will never decide to play tic-tac-toe.
Re: (Score:2)
Alphago is firmly in the weak AI category, and by itself will never decide to play tic-tac-toe.
Funny how you keep repeating that over and over, even though nobody is disputing it, or has even hinted otherwise.
Re: (Score:2)
then you could then use that as a training data set for a neural net.
No, this shows a lack of understanding of the capabilities of neural networks.
Re: (Score:2)
Re: (Score:3)
Google makes a grave mistake by not making AlphaGo available on Go servers against which the public can train to get better.
It doesn't matter. The real contributions are the ideas and techniques, which will be described in a paper. Other people will take it from here.
Doubt about AlphaGo being a fluke can be debunked by making it available to the public.
If you're not already convinced by the 60-0 victory, and the 3-0 victory, you're not going to be convinced by further games.
Re: (Score:2)
Google has also released a set of 50 games where AlphaGo played against itself.
That should help analyse its weaknesses.
https://deepmind.com/research/... [deepmind.com]
"To mark the end of the Future of Go Summit in Wuzhen, China in May 2017, we wanted to give a special gift to fans of Go around the world. Since our match with Lee Sedol, AlphaGo has become its own teacher, playing millions of high level training games against itself to continually improve. We’re now publishing a special set of 50 AlphaGo vs AlphaGo ga
Re: (Score:2)
Re: (Score:2)
I guess the guys at Google didn't watch this 3dfx commercial [youtube.com].
What's in a name (Score:2)
You think it's called AlphaGo because it plays Go?
Alphabet + Google = AlphaGo
Dead Game (Score:2)
Re: (Score:1)
OK:
AI Lawyer "Ross" has been hired by its first official law firm. [futurism.com]
AI cancer diagnosis tool [newatlas.com]
Using AI to Predict Malicious Infrastructure Activity [infosecurityeurope.com]
All 50 self-playing games have been released (Score:4, Informative)
https://twitter.com/DeepMindAI... [twitter.com]
We decided to publish the remaining #AlphaGo self-play games in one go. We hope players around the world enjoy them!
https://deepmind.com/research/... [deepmind.com]
Re:No surprise, as it cannot perform anymore (Score:4, Informative)
Wow, way to talk without knowing a thing about it.
AlphaGo managed to beat one of the best player in recent history (Lee Sedol) last year, then went on a 60-0 strike against the highest ranked professionals. Now he won a tournament with the world champion without ever showing a weakness. The experts were pretty much anonymous that he never had a chance at this point.
DeepMind have already published the details of the algorithm in a Nature paper and will do the same with the recent improvements later this year. I guess if you're right nobody will be able to reproduce the results...
Re: (Score:2)
Your reaction is exactly why Google went for that stunt: Even somewhat smart people fall for it.
Re: (Score:3)
gweihir is Zap Brannigan
https://www.youtube.com/watch?... [youtube.com]
Re: (Score:1)
Re: (Score:1)
"anonymous"? The word you're looking for is "ambiguous". Or did you mean "ambidextrous"?
Re: (Score:1)
No, he clearly meant unanimous. How on earth could you perceive that either of your suggestions make any sense?
Re: (Score:2)
Yep, it was "unanimous" and I can't spell. :p
Re: (Score:2)
And my comment was supposed to be a joke :)
Re: (Score:2)
And mine was an apology and admission that I need some sleep :)
Re: (Score:1)
You should have posted unanimously.
Re: (Score:2)
Got any proof of that? From say chess?
Here's one example [youtube.com]. There are plenty of others.
Re: (Score:2)
These kind of examples are for chess engines before the "deep learning" neural net era which AlphaGo opened our eyes. They are also just "bugs" like human players sometimes also have blindspots and make low level mistakes obvious to bystanders. Totally fixable.
Also sounds like the grandparent post don't know it is AlphaGo itself that is introducing a lot of unconventional style and stir the Go-water. Ke Jie studied those games, learnt from them, and fail to beat it one year after Lee Sedol's matches, as th
Re: (Score:2)
Once people have a chance to play against these kinds of computers, they'll find the bugs.
Re: (Score:2)
Of course AlphaGo has bugs and imperfect evaluation. It loses against itself. And once in awhile, a human player may create a board position that will reveal one of the bugs. Of course, to reach that position without making any mistakes yourself is incredibly hard. If you let GM Nakamura play 20 games against Stockfish, he may exploit a bug/weakness once, and get crushed the other 19 times simply because he never gets a setup that leads to one of the weaknesses before he makes a mistake himself.
Re: (Score:2)
Yea! And it's not really A.I. any more either now that a computer has done it.
And the sun got in his eyes.
Let me join your "humans bitter against computers" club too!
If they're not going to work on it (Score:1)
How about donating the current snapshot to a computer museum or a go association? If the hardware costs too much, perhaps do a crowdfunding campaign.
I'd say "the A.I. that beat the best human player" has some cultural value. Granted, the possibility of Google going under any time soon is very low, but this piece of great engineering achievement deserves a backup place of safekeeping to ensure it is not lost to the times.
Re: (Score:2)
How about donating the current snapshot to a computer museum or a go association? If the hardware costs too much, perhaps do a crowdfunding campaign.
I'd say "the A.I. that beat the best human player" has some cultural value. Granted, the possibility of Google going under any time soon is very low, but this piece of great engineering achievement deserves a backup place of safekeeping to ensure it is not lost to the times.
The cost of the hardware is likely trivial. The problem is likely that alphago is not really a simple program you can install and run. It is a huge dynamic neural net being fed data from multiple humans. It's not as simple as "computer program beats human" but more like "computer program beats human with the help of other humans". It's a parlor trick. Even if the other humans aren't Go masters they are still indirectly giving some of their intelligence to the machine. A truly functional go playing pro
Re: (Score:2)
It is a huge dynamic neural net being fed data from multiple humans.
No, this new 'Master' version of AlphaGo was initially trained with the self-play data from the previous version, and then enhanced by further play against itself. There was very little human input.
Why do people even post on this site? (Score:1)
Why would anyone ever read the comments here? It's the same guys posting their knee-jerk reactiona to every story. They don't red or think things through. If a headline makes something seem bad, they're outraged. Whenever a new technology is proposed they're pessimistic.
This is bullshit! (Score:2)
Instead of such frivolous things like "science" it should be handling real problems like spanking people in games of Candyland, The Game of Life and maybe even Monopoly! ;)
Nice (Score:2)
"Google Go-Playing A.I. Retires To Focus On Energy Conservation And Medicine "
Non-artificial intelligences usually do it the other way 'round.
hmm (Score:1)
Re:hmm (Score:4)
You must not be familiar with the search space that Go has, compared to the processing speed of computers.
AES-256 is also a set of rules. Are you surprised that computers can't break it ?
Re:Energy conservation? (Score:4, Insightful)
Are you stupid? They used their AI to cut the energy consumption in their data centers by 40% (!!!). And that was almost a year ago. If it did something similar to the grid, it would save $124 BILLION in electricity each year in the US alone.
Re: (Score:3)
You think they will let that happen?
Who is "they"? If you mean the corps that profit from selling electricity, it is likely that they would be more profitable with greater efficiency. Most electricity is sold at fixed prices, and power companies run "peakers" at a loss. So if more intelligent energy use smooths out the peaks and troughs, the peaker plants can be idled, cheap base power will cover more of the load, and their profits will go up.
Re: (Score:2)