Google's AI Program Takes on Human Go Champ in Five-Game Tournament

Google's AI Program Takes on Human Go Champ in Five-Game Tournament

Lee Sedol vs. Google

Google DeepMind's AlphaGo Program is in the middle of unprecedented five-game challenge in Seoul, South Korea against legendary human Go player Lee Sedol. Sedol has been the top Go player in the world over the last decade, but now he's playing an AI program -- and he lost the first game!

AlphaGo beat Sedol in the first game on Tuesday when the human resigned with more than 28 minutes remaining on his clock after the 186-move game.

"They were neck-and-neck for its entirety, in a game filled with complex fighting. Lee Sedol made very aggressive moves but AlphaGo did not back down from the fights. AlphaGo took almost all of its time compared to Lee Sedol who had almost 30 minutes left on the clock," Google said.

At the time of this publication, AlphaGo and Sedol are in the midst of the second match. Nearly 100,000 viewers tuned in live for the first match and over 1.2 million have watched the recording on the DeepMind YouTube channel, where all the matches are live-streamed. The remaining matches will take play on March 12th, 13th and 15th at 1:00 PM local time (4:00 AM GMT).

The best-of-five tournament has a $1 million USD prize. If AlphaGo wins, the money will be donated to UNICEF, STEM and Go societies.

Go is a 3000-year-old game with over 24 million players around the world, though you'll find that most Westerners aren't familiar with the game. Players place stones in key positions to surround territory and capture enemies. The concept and rules are simple -- but it isn't easy to master.

After watching the AlphaGo vs. Sedol match-up, more and more people in the United States and around the world are becoming interested in Go. Would you be up for a Go game on Steam?

On February 27th, the Ancient Go greenlight campaign was launched. Ancient Go is in full 3D and uses Unreal Engine 4. It's designed to be beginner friendly and includes a tutorial, a simple AI, online quick-play and full server for those more advanced Go players. If you'd like to see Ancient Go available on Steam, be sure to give them a "Yes" vote.

Source: Google | Ancient Go

Permalink

Meanwhile...

http://www.theguardian.com/technology/2016/mar/09/google-self-driving-car-crash-video-accident-bus

The Guardian:
It was the first time in several years of testing on public roads that a Google self-driving car prototype caused a crash. Google has blamed other drivers for past collisions during testing, but, in a first, accepted at least partial responsibility for this crash.

Still a tremendous track record. I remember when I started playing Go a decade ago, and it was impossible to play against a computer, even theoretically. The breakthroughs in DNN machine learning in the last two years has just been tremendous. Problems that computers have traditionally failed to manage are being solved so quickly. The next few years, as this begins to mature and become available to more people, so much is going to change. For one, the language barrier is going to go away. Given how many people around the world have smartphones, and how many well in a decade, it might even drop all at once. The dream of mankind since ancient times is on the horizon, to speak our own languages without compromise, and still be perfectly understood, all of the time, by everyone.

What's it going to be like in the next few years, when on the fly machine translation means that "The internet is in English" suddenly stops being true? All of that Chinese, Indian, Russian, Japanese, etc content out there that English speakers don't see, is going to start to be legible. Personally, I'm most excited to be able to finally stop worrying about ever learning Kanji. Speaking Japanese isn't a problem, but reading it? Fuck.

The Youtube channel link is a dead link.

Richard Gozin-Yu:
For one, the language barrier is going to go away. Given how many people around the world have smartphones, and how many well in a decade, it might even drop all at once. The dream of mankind since ancient times is on the horizon, to speak our own languages without compromise, and still be perfectly understood, all of the time, by everyone.

What's it going to be like in the next few years, when on the fly machine translation means that "The internet is in English" suddenly stops being true? All of that Chinese, Indian, Russian, Japanese, etc content out there that English speakers don't see, is going to start to be legible. Personally, I'm most excited to be able to finally stop worrying about ever learning Kanji. Speaking Japanese isn't a problem, but reading it? Fuck.

Maybe, maybe not. The language barrier is much more than just being able to translate the words on the fly.

There's still the matter of tone and the way even a single choice of word can drastically change it. Or how some expressions and words in one language simply can't be translated to another or carry a very different connotation (for example, 'cunt' is a big slur in the US, but not so much in the UK or Ireland). Or slang, regional and cultural differences within the same language. These are all things that are notoriously difficult to pick up on and translate even by a trained translator with years of experience. And every language has its own set of rules and quirks beyond just the vocabulary and grammar. And doing that for written language is arguably still the easy part, it would be exponentially harder for spoken language.

Not saying that it won't happen, it probably will at some point. Just that we're still a ways off from having a Star Trekesque Universal Translator (which, btw, was shown to mess up the subtle nuances too at time). Until then, we're probably going to have to live with a lot of miscommunication.

Richard Gozin-Yu:
Meanwhile...

http://www.theguardian.com/technology/2016/mar/09/google-self-driving-car-crash-video-accident-bus

The Guardian:
It was the first time in several years of testing on public roads that a Google self-driving car prototype caused a crash. Google has blamed other drivers for past collisions during testing, but, in a first, accepted at least partial responsibility for this crash.

Still a tremendous track record. I remember when I started playing Go a decade ago, and it was impossible to play against a computer, even theoretically. The breakthroughs in DNN machine learning in the last two years has just been tremendous. Problems that computers have traditionally failed to manage are being solved so quickly. The next few years, as this begins to mature and become available to more people, so much is going to change. For one, the language barrier is going to go away. Given how many people around the world have smartphones, and how many well in a decade, it might even drop all at once. The dream of mankind since ancient times is on the horizon, to speak our own languages without compromise, and still be perfectly understood, all of the time, by everyone.

What's it going to be like in the next few years, when on the fly machine translation means that "The internet is in English" suddenly stops being true? All of that Chinese, Indian, Russian, Japanese, etc content out there that English speakers don't see, is going to start to be legible. Personally, I'm most excited to be able to finally stop worrying about ever learning Kanji. Speaking Japanese isn't a problem, but reading it? Fuck.

I wouldn't be so sure about breaking the language barrier that quickly. Algorithms that can navigate through complex logical problems cannot very easily work around the largest issues with translation, that being context and grammatical systems. At the very least, the most a current era system could do is be able to pass on simple phrases, but beyond that, the capability of having a naturally flowing conversation is beyond its grasp without taking liberties with both parties' speech patterns and mannerisms, not to mention the erasure of slang or other unique expressions that cannot be easily described without stopping the conversation.

As for that accident, the human driver thought the same thing as the AI navigator, so at the very least it's breaking even with a driver who is presumably better than many other drivers, and would provide an overall net benefit if they were replaced.

Chimpzy:

Richard Gozin-Yu:
For one, the language barrier is going to go away. Given how many people around the world have smartphones, and how many well in a decade, it might even drop all at once. The dream of mankind since ancient times is on the horizon, to speak our own languages without compromise, and still be perfectly understood, all of the time, by everyone.

What's it going to be like in the next few years, when on the fly machine translation means that "The internet is in English" suddenly stops being true? All of that Chinese, Indian, Russian, Japanese, etc content out there that English speakers don't see, is going to start to be legible. Personally, I'm most excited to be able to finally stop worrying about ever learning Kanji. Speaking Japanese isn't a problem, but reading it? Fuck.

Maybe, maybe not. The language barrier is much more than just being able to translate the words on the fly.

There's still the matter of tone and the way even a single choice of word can drastically change it. Or how some expressions and words in one language simply can't be translated to another or carry a very different connotation (for example, 'cunt' is a big slur in the US, but not so much in the UK or Ireland). Or slang, regional and cultural differences within the same language. These are all things that are notoriously difficult to pick up on and translate even by a trained translator with years of experience. And every language has its own set of rules and quirks beyond just the vocabulary and grammar. And doing that for written language is arguably still the easy part, it would be exponentially harder for spoken language.

Not saying that it won't happen, it probably will at some point. Just that we're still a ways off from having a Star Trekesque Universal Translator (which, btw, was shown to mess up the subtle nuances too at time). Until then, we're probably going to have to live with a lot of miscommunication.

I'm from England, UK and we consider cunt to be the worst swear word (outside of racist words) there is. But that kind of proves your point of it being more difficult that simply translating words to fully understand other cultures :)

There are times when computer won't realise your if speaking US or UK English. For example if An American says "Nice pants", they would be on about trousers/jeans. If an Uk resident said it they are on about your underwear. So even using the exact same 2 words have completely different interpretations/meanings.

Dominic Crossman:
I'm from England, UK and we consider cunt to be the worst swear word (outside of racist words) there is. But that kind of proves your point of it being more difficult that simply translating words to fully understand other cultures :)

I was not aware of that[1].

I was under the impression that in the UK cunt can be used both as an insulting swear and in a more playful, friendly manner, like when a friend pulls a funny prank on you and you go "Aw, you cheeky cunt!". That it's one of those words where the meaning takes on different forms depending on context. Whereas in the US it's exclusively a slur, to my knowledge.

Oh well, maybe that too is limited to certain regions and/or demographics.

[1] Que relevant clip from Wayne's World

Chimpzy:
I was under the impression that in the UK cunt can be used both as an insulting swear and in a more playful, friendly manner, like when a friend pulls a funny prank on you and you go "Aw, you cheeky cunt!". That it's one of those words where the meaning takes on different forms depending on context. Whereas in the US it's exclusively a slur, to my knowledge.

Oh well, maybe that too is limited to certain regions and/or demographics.

One's person may use the word "cunt" in a casual, playful manner towards one's casual friends, it is allowed.
One's person cannot generally greet with said word nor can one playfully utter it towards one's grandmother or one's boss without committing social faux pas. Not allowed. Not yet.

Watched the second game as it unfolded last night. Sedol was behind by about 4.5 points going into the endgame, despite some excellent play close to the end, and lost by resignation.

The AI is definitely at the level of the high-dan pros (roughly the equivalent of chess grandmasters), with some of the pros observing expressing doubt that Lee will be able to win a single game in the five game series.

I'm just glad to see the game of go being talked about outside of the play servers for once. XD

weirdee:

Richard Gozin-Yu:
Meanwhile...

http://www.theguardian.com/technology/2016/mar/09/google-self-driving-car-crash-video-accident-bus

The Guardian:
It was the first time in several years of testing on public roads that a Google self-driving car prototype caused a crash. Google has blamed other drivers for past collisions during testing, but, in a first, accepted at least partial responsibility for this crash.

Still a tremendous track record. I remember when I started playing Go a decade ago, and it was impossible to play against a computer, even theoretically. The breakthroughs in DNN machine learning in the last two years has just been tremendous. Problems that computers have traditionally failed to manage are being solved so quickly. The next few years, as this begins to mature and become available to more people, so much is going to change. For one, the language barrier is going to go away. Given how many people around the world have smartphones, and how many well in a decade, it might even drop all at once. The dream of mankind since ancient times is on the horizon, to speak our own languages without compromise, and still be perfectly understood, all of the time, by everyone.

What's it going to be like in the next few years, when on the fly machine translation means that "The internet is in English" suddenly stops being true? All of that Chinese, Indian, Russian, Japanese, etc content out there that English speakers don't see, is going to start to be legible. Personally, I'm most excited to be able to finally stop worrying about ever learning Kanji. Speaking Japanese isn't a problem, but reading it? Fuck.

I wouldn't be so sure about breaking the language barrier that quickly. Algorithms that can navigate through complex logical problems cannot very easily work around the largest issues with translation, that being context and grammatical systems. At the very least, the most a current era system could do is be able to pass on simple phrases, but beyond that, the capability of having a naturally flowing conversation is beyond its grasp without taking liberties with both parties' speech patterns and mannerisms, not to mention the erasure of slang or other unique expressions that cannot be easily described without stopping the conversation.

A year or two ago, I would have agreed with you. Frankly, that is changing almost too rapidly to keep up with. What you're describing are the kind of inherent limitations of the "perceptron" sub-unit until about 2 years ago. Again, that has changed.

weirdee:
As for that accident, the human driver thought the same thing as the AI navigator, so at the very least it's breaking even with a driver who is presumably better than many other drivers, and would provide an overall net benefit if they were replaced.

It's true, and I'm interested to see what handing off the task of driving is going to do to how we get around.

Chimpzy:

Richard Gozin-Yu:
For one, the language barrier is going to go away. Given how many people around the world have smartphones, and how many well in a decade, it might even drop all at once. The dream of mankind since ancient times is on the horizon, to speak our own languages without compromise, and still be perfectly understood, all of the time, by everyone.

What's it going to be like in the next few years, when on the fly machine translation means that "The internet is in English" suddenly stops being true? All of that Chinese, Indian, Russian, Japanese, etc content out there that English speakers don't see, is going to start to be legible. Personally, I'm most excited to be able to finally stop worrying about ever learning Kanji. Speaking Japanese isn't a problem, but reading it? Fuck.

Maybe, maybe not. The language barrier is much more than just being able to translate the words on the fly.

There's still the matter of tone and the way even a single choice of word can drastically change it. Or how some expressions and words in one language simply can't be translated to another or carry a very different connotation (for example, 'cunt' is a big slur in the US, but not so much in the UK or Ireland). Or slang, regional and cultural differences within the same language. These are all things that are notoriously difficult to pick up on and translate even by a trained translator with years of experience. And every language has its own set of rules and quirks beyond just the vocabulary and grammar. And doing that for written language is arguably still the easy part, it would be exponentially harder for spoken language.

Not saying that it won't happen, it probably will at some point. Just that we're still a ways off from having a Star Trekesque Universal Translator (which, btw, was shown to mess up the subtle nuances too at time). Until then, we're probably going to have to live with a lot of miscommunication.

I agree with your analysis of the challenges of translation, but the point is that the kind of AI breakthrough we're seeing here reasonably heralds that. That is of course, if the rate of growth in this particular field proves to be as astonishing as many others. The news about Go, DNN's taking on image processing and so on, I think there is reason to be optimistic in a way that wouldn't have been warranted just a year or two ago.

Richard Gozin-Yu:
A year or two ago, I would have agreed with you. Frankly, that is changing almost too rapidly to keep up with. What you're describing are the kind of inherent limitations of the "perceptron" sub-unit until about 2 years ago. Again, that has changed.

While I don't disagree with you on the technological factor, what I am trying to say here is that the main limitation is the HUMAN factor. It would be significantly faster to just figure out how to jam a universal communicator chip into a human brain and then just have that transmit and receive messages in the future than to try to work around several hundred varying forms of expression.

weirdee:

Richard Gozin-Yu:
A year or two ago, I would have agreed with you. Frankly, that is changing almost too rapidly to keep up with. What you're describing are the kind of inherent limitations of the "perceptron" sub-unit until about 2 years ago. Again, that has changed.

While I don't disagree with you on the technological factor, what I am trying to say here is that the main limitation is the HUMAN factor. It would be significantly faster to just figure out how to jam a universal communicator chip into a human brain and then just have that transmit and receive messages in the future than to try to work around several hundred varying forms of expression.

Just a year ago, I would have agreed with you. Now, I think that the computers are likely to be able to keep up, soon enough. Again, just a couple of years ago the idea that computers would start to excel at image recognition, or human expression recognition would be a joke. Now, it's looking like a process that will bear fruit in just a matter of years. In a decade?!

Game 3 just concluded. Another win for the program. In game one, Sedoul clearly underestimated it, making uncharacteristicly rash and aggressive moves in an attempt to confuse it, only to realize too late that it was far more sophisticated than past programs that have given go a try.

Thus we were led into game 2. Sedol had learned his lesson from the first game, playing much more solid and balanced moves. As a result, he took an early lead, but made a misstep in the middle game that let alphago close the gap, and ultimately came up behind.

Game 3 now saw him attempt a more measured version of his aggressive approach from game one. he used the high Chinese fuseki, an opening pattern designed to force the opponent to invade and fight. Alphago proved up to this task, and the resulting fight favored the computer so strongly that the pros commenting on the game wondered why Sedol didn't resign. He made a crafty final attack on a large group toward the end, but came up short once again.

The stream attracted large numbers of viewers who knew nothing about go, who spent most of the match making crass comments about the female professionals... Stay classy, Internet

 

Reply to Thread

Posting on this forum is disabled.