阿尔法狗十周年：人工智能的转折点 | 索雷·格拉佩尔与普什米特·科利

欢迎回到谷歌DeepMind播客，我是汉娜·弗里教授。

Welcome back to Google DeepMinder Podcast, I am Professor Hannah Fry.

想象一下这个场景。

Picture the scene.

那是2016年3月。

It's March 2016.

在韩国首尔的一家酒店套房内，两名选手正在对弈古老的围棋游戏。

Inside a hotel suite in Seoul, South Korea, two players are playing the ancient game of Go.

这是一种难以想象的复杂游戏，长期以来被认为机器不可能掌握。

A game of unimaginable complexity, long thought impossible for a machine to master.

一方是李世石，一位传奇的18次围棋世界冠军。

On one side is Lee Sedol, a legendary 18 time Go world champion.

另一方是AlphaGo，一个基于强化学习这一强大技术的神经网络人工智能系统。

On the other, AlphaGo, a neural network based AI system built on a powerful technique called reinforcement learning.

欢迎来到在韩国首尔现场直播的DeepMind挑战赛。

Welcome to the DeepMind Challenge live in Seoul, Korea.

这步棋非常出人意料。

That's a very surprising move.

没有一个人类玩家会选择第37手。

Not a single human player would have chosen move 37.

经过连续七天、长达数小时的激烈对弈

After hours of intense gameplay spread over seven days

是的。

Yeah.

这是一步令人兴奋的棋。

That's an exciting move.

李世石放下两颗棋子，表示认输。

Lee Sedol placed two stones on the board to signal his final resignation.

击掌庆祝。

High five.

转瞬之间，世界改变了。

And in the blink of an eye, the world changed.

最终比分是四比一。

The final result of four one.

祝贺AlphaGo和整个团队。

Congratulations to AlphaGo and to the entire team.

那正是十年前，自那以后，人工智能领域发生了难以想象的变化。

That was exactly one decade ago, and the field of AI has changed unimaginably since then.

我们见证了大型语言模型的兴起、AI代理的日益复杂化，以及像蛋白质折叠这样的科学重大挑战的解决。

We have seen the rise of large language models, the growing sophistication of AI agents, and the solving of scientific grand challenges like protein folding.

但在许多方面，现代人工智能革命 arguably 正是从韩国那张木制棋盘上开始的。

But in many ways, the modern AI revolution arguably began right there, on that wooden board in South Korea.

因此，在这一集中，我们希望回顾并展望，当初这个教机器下棋的大胆实验，如何成为今日AI突破的基石。

So in this episode, we wanted to look backwards and forwards to how a bold experiment in teaching machines to play games became the foundation stone for the AI breakthroughs of today.

和我一起的，正是讲述这个故事的完美嘉宾。

And with me are the perfect guests to tell that story.

托雷·格拉佩尔是谷歌DeepMind的杰出研究科学家，当时就在首尔，作为AlphaGo项目的关键架构师。

Thore Graepel is a distinguished research scientist at Google DeepMind who was right there in Seoul as a key architect of the AlphaGo project.

还有普什米特·科利，他领导着谷歌DeepMind的科学工作，能告诉我们那些在围棋中首创的早期技术如何应对当今的关键问题。

And Pushmeet Kohli, who leads Google DeepMind's science work and is the person to tell us how those early techniques pioneered in Go can tackle crucial problems today.

欢迎两位来到本播客。

Welcome to the podcast, both of you.

索尔，我知道你本人也是一位出色的围棋选手。

Thore, I know you're an accomplished Go player yourself.

请给我们解释一下，为什么围棋被视为对人工智能的一个良好挑战。

Just explain to us why Go was seen as a good challenge for AI.

是的。

Yes.

围棋之所以被视为人工智能的完美挑战，是因为它的规则非常简单，却能产生极其复杂的对弈，包含战术、策略和复杂的模式。

The the game of Go seemed like the perfect challenge for AI because the game has such simple rules, yet it leads to such complex gameplay with tactics and strategies and complex patterns.

一旦国际象棋被攻克——或者说，深蓝已经击败了世界冠军——围棋就成了一个未解的难题。

And once the game of chess had been solved as it were, or at least, you know, Deep Blue had won against the world champion, then Go was this open challenge.

它的复杂程度远超国际象棋好几个数量级，当时没人预料到它能被迅速攻克。

It's much more complex than chess by many orders of magnitude, and nobody was expecting it to be solved anytime soon.

然而，对于计算机科学家来说，它看起来如此优雅简洁，因此在当时是理想的攻关目标。

Yet it it it looks so elegant and simple for computer scientists, and so it was the perfect game to tackle, at the time.

我的意思是，没人认为它会很快被解决，这一点可谓一针见血。

I mean, the idea of nobody thinking it would be solved anytime soon, that's sort of hits the nail on the head.

对吧，普什米特？

Right, Pushmeet?

我知道你当时在微软工作。

I mean, I know you were working at Microsoft at the time.

但当时这个问题被认为有多复杂呢？

But just how complex was this problem considered to be?

我认为它被认为极其复杂。

I think it was considered extremely complex.

这不仅是因为搜索空间的广度，也就是你能走的步数，还因为深度——你需要推理多久，以及对局有多长。

And that is because not only because of the breadth of the search space, of the number of moves you can make, but also the depth, how long you have to reason and how long the games are.

在国际象棋中，你可能只需要考虑大约60到70步的推理。

In a game of chess, you might think about reasoning about 60 to 70 sort of moves.

在围棋中，对局要长得多。

In a game of Go, it's much, much longer.

这就带来了这个问题的挑战。

And that leads to the challenge of the problem.

索尔，我知道你刚到DeepMind时是个围棋选手，你第一天就和AlphaGo对弈过吗？

Thore, I know when you first started at DeepMind being Go player, didn't you play against AlphaGo on your first day?

是的。

Yeah.

对。

Yeah.

没错。

Exactly.

所以想象一下，我第一天到DeepMind上班，认识了几个同事，包括大卫·西尔弗，他问我：索尔，你是围棋选手，对吧？

So imagine, I come first day at work at DeepMind, I know a couple of people including David Silver and he asks me, Thore, you're a Go player, right?

你能不能帮个忙，测试一下我们当时那个还叫不上AlphaGo名字的初级版本？

Couldn't you do us a favor and test our baby version of of something that wasn't even called AlphaGo at the time, of course, you know.

这是一个实习项目，他们刚刚从互联网上收集了大约几千盘棋谱，训练了一个系统，也许是几十万盘棋。

It was an internship project and they had just about taken a few thousand games from the internet and had trained a system, or a few 100,000 games maybe.

我有幸成为最早与它对弈的人之一。

And I had the opportunity to be one of the first people to play against it.

但你可以想象，我既兴奋又紧张。

But you can imagine, I was excited but I was also nervous.

那是我上班的第一天，却被带到了一个中央位置的桌子旁。

It was my first day at work, and there I was being dragged to a centrally located table.

对面坐着的我想是黄士杰，他后来被称为AlphaGo的手，面无表情。

On the other side, I think it was Aja Huang, who would later be known as the hand of AlphaGo, with his poker face.

我有机会与这个AlphaGo的初级版本对弈。

And I got to play against this baby version of AlphaGo.

想必当时有很多人围观吧？

With people watching presumably?

有

With

周围有很多人围观。

a lot of people watching all around me.

你知道，后来德米斯出现了，当然戴维一直都在。

You know, there was no Later, Demis showed up, and of course David was there the whole time.

是的。

Yeah.

那么人们会怎么做呢？

And so what does one do?

下得保守一点，对吧？

Play conservatively, right?

所以我只是想，别犯错就行，这应该不会太难吧。

So I just thought, just don't make a mistake, surely this can't be so hard.

但当然，这正是那个版本程序擅长的地方。

But of course that was exactly what that version of the program was good at.

它是在人类职业棋手对局上训练的，所以它完全知道如何应对常规下法。

It was trained on human professional games, so it knew exactly what to do against conventional play.

随着这场小规模的对局进行，我的局势越来越糟，最终以微弱劣势输掉了比赛。

And so as this little test match proceeded, my position became worse and worse and I ended up losing by a small margin.

但我成为了官方记录中第一个在Staff Ago中输掉的人。

But I took the crown of the first person who officially lost in Staff Ago.

这真是一次独特的经历。

It was quite the experience.

当然，之后每个人都认识我了。

And of course afterwards everyone knew me.

这是一种很棒的自我介绍方式。

It was a wonderful way of introducing myself.

一种令人谦卑的方式。

A humbling way.

一种令人谦卑的方式。

A humbling way.

没错。

Exactly.

当然。

Absolutely.

普什米特，提醒我们一下，好吧。

Pushmeet, just remind us I mean, okay.

我知道，从那个还是实习阶段的早期开始，这个算法已经取得了巨大的进展。

So I I I know that the the algorithm advanced quite substantially from from that early point where it was an internship.

但总的来说，能否为我们解释一下它是如何运作的，特别是关于破解组合空间这个概念？

But just broadly, explain to us how it works and this this idea about cracking the kind of combinatorial spaces in particular.

是的。

Yeah.

如果你看一下围棋，任何时候你能下的步数是有限的。

So I think if you look at the game of Go, the number of moves that you can make, at any given time, there are a finite number of moves.

但如果你观察并思考整个棋局状态，它的复杂度是指数级的。

But if you look and reason about the the overall game state, it's exponential.

而你需要考虑的状态数量呈指数增长，这正是让围棋变得极其复杂的原因。

And that exponential growth in the number of states that you have to reason about is what makes the game extremely complicated.

那么他们是怎么破解这个问题的呢？

So how did they crack it then?

再给我们提醒一下，他们发现的解决方案是什么。

What's just remind us of of of the solution that they discovered.

AlphaGo的精妙之处在于它结合了快速思考和慢速思考这两种元素。

The beauty of AlphaGo was there is this element of thinking fast and thinking slow.

在某种意义上，AlphaGo完美地融合了快速思考与慢速思考的过程，以应对这个极其庞大的搜索空间。

And AlphaGo in some sense was the perfect combination of those thinking fast and thinking slow processes coming together to take on this extremely large search space.

所以这与人类下棋的方式非常吻合，我认为。

So it matches quite well to how humans play the game, I think.

嗯。

Mhmm.

你可以想象，人类下象棋或围棋时，我们也有能力快速判断一个局面对黑方或白方是否有利。

You know, if you imagine how how a human would play a game of chess or a game of Go, we also have the capacity to look at a position and pretty quickly appreciate if that's good for black or good for white.

而且我们也能从一个局面中直接看出一些有潜力的走法。

And we can also look at a position and already see moves that seem promising.

我们从不会考虑所有可能的走法，象棋中可能有二三十种，围棋中则有两三百种。

We never look at all the possible moves which would be maybe 20 or 30 in chess or or 200 or 300 in Go.

我们立刻会依赖某些看似优美、恰到好处的走法，这些走法由我们的直觉引导。

We immediately draw on to certain, maybe even aesthetically pleasing moves that seem like just the right ones guided by our intuition.

而这种能力通过规划得到补充，在规划中我们会明确地推演各种可能性。

And that element is complemented by planning where we explicitly reason through the possibilities.

如果我走这一步，对手可能会走那一步，然后我必须用这一步来应对。

If I make this move, my opponent might make that move and then I have to counter with this move.

这两种不同的思维方式共同构成了人类下棋的方式，也同样体现在AlphaGo的下法中。

And these two different ways of thinking come together in how humans play these games, and they also come together in in how AlphaGo plays.

直觉与计算，正如

The intuition and the calculation as it

那样。

were.

没错。

Exactly.

那么，这就是灵感的来源吗？

So was that the inspiration then?

你有没有思考过自己下棋的方式，以及其他围棋选手的下法，并从神经科学中直接汲取了灵感？

Did you sort of think about how you were playing the game, how other Go players were playing the game, and draw that direct inspiration from neuroscience effectively as it were?

是的，我认为这确实是其中一个方向，因为很多团队成员本身就是棋手，能够进行自我反思，观察我们是如何应对这盘棋的。

Yeah, I think that is definitely one direction because a lot of team members were actually game players who were able to introspect and see how we tackled the game.

当然，这还结合了深度学习——自2012年以来，深度学习作为一种方向迅速发展，并首次为我们提供了学习这些近似函数的工具，例如价值函数，它接收一个棋盘并判断当前局面对黑方或白方有多有利；或者策略网络，它接收一个棋盘并根据职业棋手可能采取这些走法的概率对可用的着法进行排序。

And then of course that comes together with deep learning that at the time, you know, since 2012 had grown as a direction and now for the first time gave us the tools to learn these approximate functions for example the value function that takes a board and tells us how good it is for either black or white, or the policy network that takes a board and effectively ranks the available moves according to how likely it would be that a professional player would take them.

因此，当时深度学习正处在成熟阶段，能够应对这个问题，并为我们实现快速思考提供了机会。

And so deep learning was just ripe at the time to tackle this problem and gave us the opportunity to implement the fast thinking.

慢思考与深蓝的情况并无不同，你知道的，那就是对博弈树的搜索，这早已为人所知，我们现在或许称之为传统人工智能。

The slow thinking is not unlike what happened in Deep Blue, you know, it's the search of the game tree that was already known and that we might now call good old fashioned AI.

明白了。

Okay.

我的意思是，你很早就输给了这个东西。

Well, I mean, you lost to this thing quite early.

但当它经过团队中许多人的测试后，我知道你们曾让职业围棋手范廷钰来办公室进行测试。

But once it had gone through a lot of the people on the team, let's say, I know that you tested it with a professional Go player because you had Fan Hui come into the office.

是的。

Yeah.

没错。

Exactly.

那时你们有多大的把握认为它会赢过他？

How confident were you at that point that it was gonna beat him?

是的。

Yeah.

我们的信心程度各有不同，这非常有趣。

We had different levels of confidence, which was really interesting.

我们非常幸运地找到了他，他是当时欧洲围棋冠军。

So we had been really lucky to find him, you know, he was the European GO champion at the time.

他住在波尔多，然后过来了。

He lived in Bordeaux and and came over.

我们诱使他来和我们下这盘棋，安排是他会与当时版本的AlphaGo进行10场测试对局。

We lured him into into playing this game with us, and the setup was that he would play 10 test games against the version of AlphaGo at that point.

我个人认为，AlphaGo不可能已经强大到能击败欧洲冠军这样的职业棋手。

And I personally thought that AlphaGo cannot possibly be at the point already that it beats the European champion, a professional player.

于是我和大卫·西尔弗打了个赌。

And so I had a bet with David Silver.

大卫·西尔弗很有信心。

David Silver was confident.

他说，我觉得AlphaGo会以10比0完胜。

He said, I think AlphaGo is going to nail it ten zero.

我说，不可能。

And I said, no.

我觉得AlphaGo至少会输一局。

I think AlphaGo will lose at least one game.

赌注是，输的人必须打扮成一位古代日本围棋大师，到办公室里穿着这身装扮待上一整天。

And the the bet was that whoever lost would have to show up at the office dressed as an ancient Japanese Go master and be in the office for one day with that.

那么，是谁真的那样打扮出现了？

Well, who showed up like that?

是我，因为是十比零。

It was me because 10 nil.

事实上确实是十比零。

It was in fact 10 nil.

是的。

Yeah.

但这给了我们信心，也让他们相信，我们将来能够应对更强大的对手。

But it did give us confidence and gave them confidence that we would be able to tackle even harder opponents in the near future.

当然，你们后来在2016年登上了飞往首尔的航班，去与李世石对弈。

Which you did, of course, on a plane you got in 2016 to to Seoul and Korea to play against Lee Sedol.

我的意思是，跟我们说说他到底有多厉害。

I mean, just tell us give us a sense of how phenomenal a player he actually is.

是的。

Yeah.

当时，李世石确实是顶尖棋手之一，甚至可能是最强的，拥有辉煌的赛事夺冠纪录。

So Lee Sedol was really one of the or maybe the best players at the time with an incredible track record of of winning tournaments.

当时，人们将他与罗杰·费德勒相提并论，称赞他的成功与智慧才华。

He was compared to Roger Federer at the time for his success and intellectual brilliance.

因此，他接受我们的挑战，愿意与我们对弈，这对我们来说是极大的荣耀。

And so for us, it was a tremendous honor that he accepted our challenge to play against him.

这同样是一个巨大的挑战，因为我们必须确定一个具体日期，对吧？

And it was a tremendous challenge because we had to set a date, right?

你不能只是说：‘我们准备好了再告诉你。’

You can't just say, you know, we'll tell you when we're ready.

日期定下来后，我们必须朝着这个目标努力，真正把AlphaGo提升到足够强大的水平。

A date was set and we had to work towards that date actually make AlphaGo strong enough.

而让这场对弈更添紧张与激动的是，李世石坚信自己会赢。

And what added tension and excitement to it was that Lee Sedol was convinced that he would win.

当时他认为AlphaGo获胜的可能性极低。

He thought it highly unlikely at the time that AlphaGo would win.

当然，他的评估是基于他所看到的与樊麾对弈的棋谱，他认为自己更强。

And of course, he was basing his assessment on the game records that he had seen against Fanhua, and he assessed that he was better.

但他不太了解的是，AlphaGo通过我们不断进行的训练和算法优化一直在持续进步。

But of course, what he wasn't so aware of is that AlphaGo was constantly improving through the training and the algorithmic refinements and so on that we made.

于是整个团队前往韩国，你无法想象那里的人们有多兴奋。

And so the entire team basically went to South Korea, and you wouldn't believe the excitement of people there.

你知道，在英国，围棋只是一项小众活动。

You know, the truth is in in England, Go is a bit of a niche activity.

对吧？

Right?

很少有人会下围棋，甚至根本不知道这项运动。

Very few people would be able to play it or even know about it.

但在韩国，人们都非常兴奋。

But in South Korea, people were so excited.

顶尖的围棋选手是名人，我们到那里时，有大批摄影师拍照，还有一支纪录片摄制组跟随我们，想象一下，一群典型的电脑极客突然成为全球瞩目的焦点，这场对决真是一场奇妙的经历。

The best Go players are celebrities, and, you know, we came there and there were hordes of photographers that took pictures, we had a documentary film crew with us and so imagine typical computer geeks as it were, suddenly in the limelight of the world for this match, that was quite the adventure.

是的。

Yeah.

我的意思是，你对AlphaGo的表现感到紧张吗？

I mean, you nervous about the performance of of AlphaGo?

是的。

Yes.

我们确实很紧张。

We were definitely nervous.

当然，我们有一个非常复杂的评估流程。

So of course we had a very sophisticated evaluation pipeline.

你可以测试你能够接触到的选手，比如范华，这非常有帮助。

You can test against players that you have access to like Fran Hui, that was super helpful.

你也可以测试程序的先前版本，并计算我们所谓的系统ELO分数，这基本上是根据你与其他版本（可能是你程序的早期版本）对弈的所有结果，来计算新版本的评分，这些指标可以很好地校准。

You can also test against previous versions of the programme and you can calculate what we call the ELO score of the system, which basically takes the outcomes of all the games that you play against other versions, maybe earlier versions of your program, and calculates what the rating of the new version is and you can calibrate these things quite well.

但当然，我们并不知道李世石在这个评分体系中处于什么位置。

But of course we didn't know where on that scale Lee Sedol would be.

当然，我们也希望留一些余地，你知道的。

And of course we wanted a cushion as well, you know.

如果能好得多，有更大的把握，那就更好了，毕竟这是世界舞台，对吧？

It would have been would be nice to be quite a bit better, to have some certainty because this is the world stage, right?

如果你输了，这对声誉会是不小的打击。

If you lose this, that's bit of a hit to the to the reputation.

所以是的，我们很紧张，一直工作到最后一刻，我们还需要确保系统非常稳定，你知道的，你不想在最后一刻做改动来让它好一点点，却冒让它变得不稳定的风险。

And so yeah, we there was we were nervous, we worked up to the last minute, we also needed to make sure that the system is really stable, you know, you don't want to make last minute changes to make it that little bit better but risk that it now becomes unstable.

但最终我们对它相当满意，于是我们进入了那个如今闻名的酒店房间，那里是所有行动发生的地方，媒体都在那里等候，然后我们开始了比赛。

But in the end we were quite happy with it and so we entered that now kind of famous hotel floor where all the action happened, where all the press was waiting and so on, and embarked on the match.

世界各地的人都在观看，包括普什米特，也包括你。

And people were watching from around the world, including Pushmeet, including you.

是的。

Yeah.

那么，当时你在哪里？

So, I mean, where were you at this point?

你当时也在观看。

You you were watching on.

是的。

Yeah.

我在西雅图。

I was in Seattle.

我的意思是，我在第一局进行到一半时才真正投入进去。

I I mean, I really started getting into it in the middle of the first game.

很明显，AlphaGo已经达到了这一特定里程碑。

It became so clear that AlphaGo had reached that specific milestone.

你甚至能从媒体、评论员和李世石本人的反应中看出来。

And you could even see the reaction from the press and the commentators and Lee Sedol himself.

有趣的是，你提到是在那局的中段，但在那局的早期阶段，谁能占上风有明显迹象吗？

Well, it's interesting that you said in the middle of that game because in the early stages of that game, was it clear who had the upper hand?

我认为，对于只是旁观的人来说，早期大家都觉得李世石肯定会赢。

I think from, like, from person who was just watching it, I felt that in the early stages, everyone felt quite confident that Lee Sedol would would win.

事实上，只有当比赛进行到接近最终结果时，人们才意识到，通过计算领地，AlphaGo占据了优势。

In fact, only as the game progressed and it became closer to the final outcome that they realized that as you count the the territory, AlphaGo had an an advantage.

事实上，这让所有人都感到惊讶。

In fact, it came as a surprise to people.

你怎么看？

What did you think?

是的。

Yeah.

我当时在现场和一位美国职业围棋选手有过一次有趣的互动，他坐在我旁边一起观赛。当时棋盘一角出现了一连串变化，他靠近我说：‘我总是告诉我的学生不要下AlphaGo刚刚下的那种愚蠢的棋，这么说吧，这简直没救了。’

So I had this interesting interaction on-site with a professional Go player, an American professional Go player, who was sitting next to me while we were watching, and there was some sequence unfolding in a corner, and he kind of approached me and said, you know, I always tell my students not to play that stupid move that AlphaGo just played, so I mean, it's pretty hopeless.

我当时想，我可不是那么专业的专家。

And I was like, I'm not as much as an expert.

我的反应就是：咱们就等着看吧。

I'm let let's just wait and see was my was my reaction.

那场第一局结束后，这位先生走过来对我说：这是我这辈子见过最非凡的事情。

And then after after that first game, this gentleman came to me and said, this is the most phenomenal thing I've ever experienced.

我非常感激能在这里见证一台机器能够达到如此高水平的围棋对弈，我们从中将学到非常多的东西，而他已经欣然接受了这一点。

I'm so grateful that I'm allowed to be here to witness that a machine can play go at this level, and there's gonna be so much we can learn from it, and he was already embracing this.

我的意思是，想象一下这些人将一生都奉献给了这项游戏的研究，他们往往从孩童时期就开始训练，直到如今只为精通这个游戏。

I mean, have to imagine these people dedicate their lives to the study of this game, and they've often trained from being young children to their current age just to master this game.

因此，当一台机器能够与人类围棋选手匹敌甚至超越时，他们当然会感到震惊。

And so, of course, it comes as a shock to them that a machine might match or even exceed a human Go player.

因为如果第一局是AlphaGo获胜，那么在第二局中，AlphaGo下出了一步真正让所有人震惊的棋。

Because if that was the first game when AlphaGo won, in the second game, AlphaGo did something that, I mean, really surprised everybody.

这是一步非常令人惊讶的棋。

That's a very surprising move.

专业解说员几乎一致认为，没有任何一位人类玩家会选择第37手。

Professional commentators almost unanimously said that not a single human player would have chosen Move 37.

AlphaGo表示，人类玩家下出第37手的概率仅为万分之一。

AlphaGo said there was a one in 10,000 that Move 37 would have been played by a human player.

请给我们解释一下，现在著名的第37手究竟发生了什么。

Just just explain to us what happened with the with the now famous Move 37.

是的

Yeah.

所以这是一个非凡的场景，我当时坐在国际英语解说室里，我们的美国解说员迈克尔·雷蒙德在墙上放了一个巨大的演示棋盘，他会把所有棋子摆上去，向观众展示正在下的棋局，并分析不同的变化。于是，他拿起对应第37手的棋子放在棋盘上，然后退后一步说：‘不，这肯定错了’，并把棋子拿了下来。

So this this was a remarkable scene, and I was sitting in the the international English speaking commentating room, and Michael Redmond, our American commentator, he had this big demo board on the wall and he would put all the stones up there on the board to show people what was being played and comment on different variations, and so he took the stone corresponding to move 37 on the board, and then he stepped back and said, no, this must be wrong, and he took it back.

但他又看了看屏幕，说：‘不，不，这确实是AlphaGo下的棋’，然后把棋子放了回去。

And then he looked at the screen again and said, no, no, that is actually what AlphaGo played, and he put it

放了回去。

back.

他感到困惑。

He was puzzled.

你可以明显看出，对人类棋手来说，这一步完全违背直觉。

You could you could see it, that that was such a counterintuitive move for a human player.

这是一步五线的肩冲，而人类棋手通常会避免这种下法。

It was a shoulder move on the fifth line, and this is typically something that human Go players avoid.

在围棋中，常常是在边线上进行推挤，一方沿着棋盘边缘构筑实地，另一方则向中央发展势力。

So often in Go, is some kind of pushing going on along the along the edges, and one of the players builds territory along the wall of the board, and the other side builds influence towards the center of the board.

如果这种情况发生在第三和第四线，通常被认为是大致均衡的。

And if that happens on the third and fourth line, this is considered to be roughly equitable.

你知道，双方都能从中获得一些东西。

You know, both sides get something out of it.

但AlphaGo实际上暗示的是，即使你在第五线下子，让对方获得更多的地盘，这仍然是有利可图的。

But what AlphaGo was effectively suggesting is that it's still profitable if you do it on the fifth line and you give that much more territory to the other party.

这正是让人们感到惊讶的地方——居然会存在这样的情况，这种下法竟然是正确的。

And that's what was so surprising to people that that could that there would be situations in which that would be correct.

因此，这不仅是一步非常特殊的棋，而且在某种程度上，它代表了一种新的方式，来权衡即时地盘与向中央扩展势力之间的关系。

And so not only was it a very special move, but it in a way, it represented a new way of of weighing these two factors of immediate territory versus influence towards the center of the board against each other.

这超出了人类棋手通常会做的范围，普什米特。

Something that went beyond what a human go pay would normally do, Pushmeet.

是的，绝对如此。

Yeah, absolutely.

我的意思是，正是在这样的时刻，你看到了人工智能系统的真正潜力——它在拓展人类的知识边界。在这个特定案例中，人们长期以来一直将围棋视为一个需要深入研究的领域。

I mean, are moments like this where you see the true potential of an AI system, expanding human knowledge, where people have regarded in this particular case, the game of Go as a thing to be studied for many, many years.

就在这一刻，这种知识得到了扩展。

And there comes this particular point where that knowledge is expanded.

人们最初持怀疑态度，游戏中的情况也是如此。

And people are at first skeptical, which was the case in the game as well.

当这步棋下出时，很长一段时间内都被认为是幻觉或错误，直到其意义逐渐明朗。

When the move was played, it was considered a hallucination or a mistake, right, for quite a bit of time before its implications became clear.

后来在比赛中。

Later on in the game.

没错。

Exactly.

因为它被证明是第二次获胜的关键。

Because it proved to be pivotal to the second win.

是的。

Yeah.

这不仅是在那场比赛中的一个时刻，我认为，它也是整个AI历史中的一个时刻——这个特定的瞬间向我们表明，未来总会有一些时候，这些系统会产生我们甚至无法判断是否正确或是否为惊人突破的洞见，但它们却会深刻影响我们以全新视角看待整个研究领域的方式。

It was not just a moment in that game, but it was also a moment, I think, in the whole sort of history of AI where that particular moment showed us that there will be times when these systems will produce insights which we might not even be able to discern whether they are the right things or amazing breakthroughs, but yet they will have a lot of influence in how we look at whole areas of study in a completely new light.

我也想谈谈第78手。

Well, I also want to talk about move 78.

这是李世石下的一手棋，让AlphaGo感到困惑，最终导致它认输。

This is a a move that was played by Lee Sedol that confused AlphaGo, causing it to resign the game.

李世石在这里想做什么？

What is Lee Sedol up to here?

他在这一步上已经花了七到八分钟了。

He just burned like seven or eight minutes just on this move already.

看这一手棋。

Oh, look at that move.

这是一步令人兴奋的棋。

That's an exciting move.

哦。

Oh.

说实话，我不太确定AlphaGo在这里想做什么。

You know, I'm not actually sure what AlphaGo is trying to do here.

到这个时候，AlphaGo已经连续赢了三局，而李世石下了一步让系统困惑的棋。

So by this point, AlphaGo has won three games in a row, and now Lee Sedol does a move that confuses the system.

这么说公平吗？

Is that fair to say?

是的，这么说完全公平。

Yeah, that's absolutely fair to say.

第78手是李世石下的一步不寻常的尖顶棋。

So move 78 was an unusual wedge move that Isidol played.

当时棋盘中央发生了一场非常有趣的较量，李世石找到了这步棋，这步棋也让人们感到惊讶，就像第37手一样。

There had been a very interesting battle, as it were, at the center of the board, and Isidol found this move, and it was also surprising to people, similar to Move 37.

从那时起，我们观察到AlphaGo对局势的把握不再清晰了。

And from then on we observed that AlphaGo didn't have a good grasp of the position anymore.

我们发现它下的棋在我们看来很不合理——当然，第37手也不太合理，但这些新棋步甚至连我们这样的业余爱好者都觉得奇怪，可见它确实被这步棋打乱了。

We saw that the moves that it made didn't really make sense to us in a bad way, you know, 37 also didn't make sense to us maybe, but these moves even to amateurs like us seemed strange, and so it had been confused by the move.

为了让你理解为什么这仍然如此重要，你可能会说，这是一场五局三胜的比赛，AlphaGo已经赢了前三局。

And just to zoom out to give you a sense of why this still mattered so much, so you might say, okay, it's a match of five games, and AlphaGo has won the first three.

还有什么需要证明的呢？

What more is there to prove?

但那时我们在想，如果李世石赢下最后两局，你会得出什么结论？

But then we were thinking, well, if now, Lee Sedol was to win the last two, what would you conclude?

他已参透了。

He's got it figured

明白了，他抓住了系统的脆弱性。

out, He's bound the fragility.

没错。

Exactly.

因此，这将是人类的胜利。

So the human it would have been the human triumph.

这就是为什么那场比赛和最后一场对我们来说依然非常激动人心。

And so that's why that game and the last one were still very exciting to us.

但我们也并非完全感到失望。

But it wasn't entirely the case that we were disappointed.

我们当然感到失望，但同时也对李世石充满敬佩——作为一个普通人，他竟然能下出这样的棋。

We were certainly disappointed, but also we had so much admiration for Lee Sedol to, you know, as a human, to be able to find this move.

你必须想象一下，这位毕生致力于围棋的大师，在这场必须面对机器完美表现、自己却苦苦寻找应对之法的战斗中，内心该有多艰难，对吧？

You just have to imagine this this master who has dedicated his life to playing this game in this battle that must have been so hard on him, right, to to see this machine play so perfectly and him struggling to find a way.

而在第四局中，他终于找到了突破口。

And then in game four, he finds a way.

正如他在新闻发布会上后来所说，他感到无比高兴和自豪，因为这可能是他最后一次，代表人类找到战胜机器的方法。

And as he put it in the press conference, I think later, he he said that he was so happy and proud that he was able, maybe for the last time, on behalf of humanity, to find a way to overcome the machine.

因为有些人称它为‘神之一手’，不是吗？

Because some people called it the divine move, didn't they?

是的。

Yeah.

对。

Yeah.

我认为，考虑到当时紧张的氛围，以及他那一刻超越自我的表现，最终下出这步棋，这个称呼确实非常贴切。

And I think given the the tension at that point in time and and him really outgrowing himself at that moment and finding that move, I think it's a it's a good name for it.

最终比分是阿尔法狗以四比一获胜。

Well, the final score was four one to AlphaGo in total.

围棋界对此有何反应？

What was the reaction from the Go community?

是的。

Yeah.

围棋界密切关注了这场比赛，当然，结果极具戏剧性，对许多人来说也出乎意料。

So the Go community followed the match very closely, and, of course, the outcome was dramatic, and for many people, unexpected.

因此，人们的反应各不相同。

And so people showed very different reactions, you know.

有些人对结果感到无比惊讶和震惊，有些人简直不敢相信。

Some people were absolutely amazed and surprised about outcome, some people couldn't believe it.

另一些人则认为，一个时代已经终结，因为最强的围棋选手可能不再是人类，而是一台机器。

Others of course also thought that some era had come to an end because now maybe the strongest Go player was no longer a human but a machine.

但总体而言，我们感到惊人的是，人们对围棋的兴趣反而增加了。

But overall what we found amazing is that there was an uptick in interest in the game of Go.

我认为现在下围棋的人比以前更多了，围棋界也真正接受了从AlphaGo身上学到的东西。

I think more people play Go now than did before, and the Go community really embraced the learning from AlphaGo.

现在有许多程序的工作方式与AlphaGo基本相同，人们用它们来进行教学。

So there are now many programs that work essentially the same way that AlphaGo does, and people use it for teaching purposes.

他们通过这些程序分析自己的对局。

They analyze their games through it.

总的来说，我认为这为整个围棋界带来了积极的推动。

And overall, I think it has provided a lift to the whole Go community.

我想问问你，AI界对这场比赛的反应如何？

Let me ask you about the reaction from the AI world to to this match.

当时有什么热议？

What was the buzz?

当时的讨论是怎样的？

What was the conversation like?

李世石对战AlphaGo的比赛是一个关键转折点，许多人在机器学习领域一直将这些模型和技术视为数学和应用项目，但从此开始看到了证据，表明这些系统能够自我学习并超越人类的知识。

The Lee Sedol match, the AlphaGo, Lee Sedol match was a key pivot point where a lot of people, especially in the machine learning community who have been sort of working on these models and techniques as a mathematical and applied project started to see evidence that these systems can self learn and go beyond human knowledge.

这是一个非常重要的观点，因为在机器学习中，你是用收集到的训练数据进行训练的。

And that is a very important sort of point because in machine learning, you train with training data which has been collected.

你自然的预期是模型会与这些数据的分布保持一致。

And your natural sort of expectation is that the model is going to just be consistent with that distribution.

但能够超越这种分布，并让这种洞见被世界所利用，我认为这是整个经历中得出的惊人洞察。

And to show that you can go beyond that distribution and that insight then can be utilized by the world, I think, is an amazing sort of insight that comes out of this whole experience.

这真正揭示了人工智能的潜力——不仅在围棋领域，还在对世界的理解、化学、生物学、数学和计算机科学中。

And it really points to what is possible with artificial intelligence in not just the game of Go, but in the understanding of world, in chemistry, in biology, in mathematics, in computer science.

这些系统将能够发现并向我们揭示哪些类似第37手的惊人类比？

What are these amazing analogs of Move 37 that these systems will be able to discover and and reveal to us.

你提到的超越人类智能这一点，真的非常引人入胜。

I think that point that you made there about going beyond human intelligence is is just so fascinating.

但让我觉得AlphaGo故事中最令人着迷的一点，即使在4:1获胜之后，是你后来开发了AlphaZero——你移除了所有人类数据，所有它曾用于训练的围棋对局，结果发现，一旦剔除人类智能，这个系统反而表现得更好，这让我感到震惊。

But one of the things that I find most intriguing about the the AlphaGo story, even after the victory of four one, is that you then built AlphaZero where you took away all of the human data, all of the the games of Go that it had been trained on, and discovered that once you take out the human intelligence, the thing actually improved, which is astonishing to me.

是的。

Yeah.

从科学的角度来看，有人会认为这甚至比最初的AlphaGo更进一步。

From a scientific perspective, one could argue that that is even an even bigger step than the original AlphaGo.

因为正如你所说，AlphaZero系统无法访问任何人类对局记录，不了解人类如何下棋，也没有利用关于围棋的先验知识，而仅仅依赖于游戏规则以及我们之前讨论过的策略网络和价值网络的表示与学习方法。

Because as you were saying, the AlphaZero system doesn't have access to any human game records, how humans play, didn't have access to prior knowledge about the game, how the game is played, but really only had access to the rules of the game and means of representing and learning these functions that we talked about, the policy net and the value net.

因此，它最初完全是随机落子，因为它对什么是好棋或坏棋没有任何概念，但它通过对局积累经验，逐渐学会哪些走法更可能带来胜利，哪些更可能导致失败，哪些局面有潜力，哪些局面没有希望，最终，它下出的棋步变得越来越好。

So basically it starts playing entirely randomly at the beginning because it has no notion of what good or bad moves are, but it gathers experience from playing these games and it learns what are moves that are more likely to lead to a win, what are moves that are more likely to lead to loss, what are positions that look promising, what are positions that are not promising, and eventually, it it starts playing better and better moves.

当然，它不再受限于人类的知识。

And now of course it's not limited by human knowledge.

它所发现的东西令人惊叹。

And what it discovered was amazing.

首先，它重新发现了人类的下法。

So first of all it rediscovered ways of how humans play.

这让人感到非常安心，你知道的。

That was totally reassuring, you know.

在围棋的角落里有一些我们称之为定式的模式，在国际象棋中也有一些固定的开局走法。

There are certain patterns in the corner in Go that we call joseki, or in chess there are certain opening moves.

这个系统现在更具通用性。

The system was now more general.

它可以下国际象棋、围棋和将棋，如果我们这样训练它，它还能玩其他许多棋类游戏。

It could play chess, Go, and shogi and could have played any number of other board games if we trained it that way.

所以一开始，它重新发现了人类的知识，让我们惊叹不已。

And so at first, it rediscovers human knowledge, and we think, wow.

这太酷了。

This is so cool.

它找到了相同的开局方式等等。

It it finds the same openings and so on.

然后我们观察一些这些开局，发现它不再使用它们了。

And then we look at some of these openings and it stops playing them.

我们想，这是怎么回事？

We think, what's going on?

它已经建立了一种声誉。

It has found a reputation.

所以它重新发现了人类的知识，然后又抛弃了它，因为它已经超越了这些知识，找到了更好的下法。

So it discovered, rediscovered human knowledge, and then it discards it because it has now gone beyond it and has found there's actually better ways of playing.

我不会再继续以这种人类的方式下棋了。

I'm not going to continue playing in this human way.

一些人类尚未有效发现的东西。

Stuff that humans haven't found yet effectively.

没错。

Exactly.

当AlphaZero下围棋时，它的下法在我看来非常陌生，这根本不是我从围棋老师那里学到的那种围棋，那种结构化的方式或许是为了让人类更容易理解。

For AlphaZero when it played Go, the way it played Go looked alien to me in So the this wasn't the kind of Go that I had learned from my Go teacher, you know, which is structured maybe in a way that enables humans to understand it.

这些落子在当时看起来非常自由，完全说不通。

These moves looked very free and didn't make much sense at the time.

但三十步之后，一切就都清晰了。

But 30 moves later, everything would fall into place.

你会恍然大悟：哦，原来如此，真厉害，这下说得通了，仿佛它具有远见——而它确实有。

Would see, oh, yeah, oh, wow, that makes sense now, and so on, as if it had the foresight, in a way which it did.

对吧？

Right?

所以，从无到达到这种水平的棋艺，这一发现非常令人印象深刻。

So that discovery from nothing to to that level of play was very impressive.

好的。

Okay.

所以，有一件事我想给你看，是你们在首尔时发生的事。

So there's there's something I I want to show you, something that happened actually when you guys were in Seoul.

因为正如你之前提到的，你们当时正在为这部关于AlphaGo的纪录片拍摄。

Because you as you mentioned before, you were being filmed for this documentary for for AlphaGo.

有一些镜头没有被用在电影里，但当时摄像机还在拍摄，而麦克风仍然开着。

And there's some footage that didn't make it into the film, but it was it was captured by the cameras as they were sort of packing up, but the microphones were still running.

我不知道你有没有听过这段小片段。

I don't know if you've you've heard this little clip.

我来放给你听。

Let me play it for you.

等一下。

Hold on.

这是德米斯和大卫在进行一次私下对话。

This is Demis and David having a sort of private conversation.

看到一个被认为不可能解决的问题如此迅速地发生变化，真是太惊人了，是的。

It's just amazing seeing how quickly a problem that is seen as being impossible Yeah.

可以转变为实际上是。

Can change to being And, technically,

我们研究蛋白质折叠。

we consult protein folding.

这简直太重要了。

That's like, I mean, it's just huge.

我相信我们能做到。

I'm I'm sure we can do that.

我之前就以为我们能做到。

I was I thought we could do that before.

是的。

Yeah.

但现在我们 definitely 可以做到。

And now but now we definitely can do it.

干得好。

Good job.

谢谢。

Thank you.

好的。

Okay.

太棒了。

Beautiful.

这难道不很棒吗？

Isn't that great?

是的。

Yeah.

索雷，你认为这捕捉到了当时的情绪吗？

Thore, do do you think that captured the mood at the time?

是的。

Yeah.

那正是阿尔法狗当时打开的那扇门。

That was the kind of door that AlphaGo opened at the time.

对吧？

Right?

如果我们能做到这一点，那我们还能做什么？

If we can do this, then what else could we do?

因为这是一款拥有10的170次方种不同局面的游戏。

Because this is a game with 10 to the power of 170 different positions.

这极其复杂。

This is super complex.

如果我们有系统的方法来应对这种组合搜索空间，那么我们似乎也能处理其他大规模的组合搜索空间。

And if we have principled ways of navigating that kind of combinatorial search space, then it seems plausible that we would also be able to handle other large combinatorial search spaces.