图奥马斯·桑德霍尔姆：扑克与博弈论

因此在现实世界中，找到纯粹的重复博弈实际上非常罕见。

So to find a purely repeated game is actually very rare in the world.

所以它们其实是对实际情况非常粗略的模型。

So they're really a very coarse model of what's going on.

然后，如果你从简单的重复矩阵博弈往上延伸——不是直接到扩展形式博弈，而是介于两者之间——还有随机博弈，你可以将其视为这些小矩阵博弈，当你和对手采取行动时，它们不仅决定下一步会进入哪个状态或哪个博弈，而是决定可能进入的下一个博弈的概率分布。

Then, if you move up from just repeated, simple repeated matrix games, not all the way to extensive form games, but in between, there's stochastic games, where you think about it like these little matrix games and when you take an action and your opponent takes an action, they determine not which next state I'm going to, next game I'm going to, but the distribution over next games where I might be going to.

这就是随机博弈。

So that's the stochastic game.

但这就像矩阵游戏的重复，随机博弈的扩展形式博弈。

But it's like matrix games repeated, stochastic games extensive form games.

这是从具体到一般的演变。

That is from less to more general.

扑克就是后者的一个例子。

Poker is an example of the last one.

所以它确实处于最一般的设定中。

So it's really in the most general setting.

扩展形式博弈。

Extensive form games.

这某种程度上正是AI社区一直在研究的内容，并以德州扑克单挑对局作为基准测试。

And that's kind of what the AI community has been working on and being benchmarked on with this heads up noblem with Texas Holden.

你能描述一下扩展形式博弈吗？

Can you describe extensive form games?

这里的核心思想是什么？

What's the motto here?

是的，如果你熟悉树形结构的话。

Yeah, so if you're familiar with the tree form.

所以它本质上是树形结构。

So it's really the tree form.

就像国际象棋中存在搜索树一样。

Like in chess, there's a search tree.

与矩阵形式相对。

Versus a matrix.

对阵矩阵，是的。

Versus a matrix, yeah.

这种矩阵被称为矩阵形式、双矩阵形式或标准型博弈。

The matrix is called the matrix form or bimatrix form or normal form game.

而这里你看到的是树形结构，在这种形式下可以进行某些类型的推理，但转换为标准型时会丢失信息。

And here you have the tree form, so you can actually do certain types of reasoning there that you lose the information when you go to normal form.

存在某种形式的等价性，比如从树形结构出发，如果把每个可能的应急方案都视为策略，那么实际上可以回归到标准型，但会因缺乏时序性而丢失部分信息。

There's a certain form of equivalence, like if you go from three form and you say every possible contingency plan is a strategy, then I can actually go back to the normal form, But I lose some information from the lack of sequentiality.

多人博弈与双人博弈的区分是个重要概念。

Then the multiplayer versus two player distinction is an important one.

因此零和博弈中的双人游戏在概念上和计算上都更简单。

So two player games in zero sum are conceptually easier and computationally easier.

虽然像这个例子规模仍然很大，但在概念和计算层面确实更简单。

They're still huge like this one, but they're conceptually easier and computationally easier.

从概念上说，当存在多个均衡时，你无需担心对方会选择哪个均衡策略。

That conceptually you don't have to worry about which equilibrium is the other guy going to play when there are multiple.

因为任何均衡策略都是对其他均衡策略的最佳回应。

Because any equilibrium strategy is the best response to any other equilibrium strategy.

所以我可以选择与你不同的均衡策略，我们仍能得到正确的博弈值。

So I can play a different equilibrium from you and we'll still get the right values of the game.

这个特性在双人非零和博弈中就会失效。

That falls apart even with two players when you have general sum games.

即使不考虑合作因素，也要说明这一点。

Even without cooperation, just to say.

即便没有合作。

Even without cooperation.

从两人零和博弈到两人非零和博弈，甚至到三人零和博弈，存在巨大差异。

So there's a big gap from two player zero sum to two player general sum or even to three player zero sum.

至少在理论上，这是个巨大的鸿沟。

That's big gap, at least in theory.

你能用非数学的方式直观解释为什么三人及以上玩家时体系就会崩溃吗？

Can you maybe non mathematically provide the intuition why it all falls apart with three or more players?

看起来应该仍能存在一个纳什均衡点，既有指导意义又能保持稳定。

It it seems like you should still be able to have a Nash equilibrium that, that's instructive, that holds.

好的。

Okay.

确实，所有有限博弈都存在纳什均衡点。

So it is true that all finite games have a Nash equilibrium.

这就是约翰·纳什实际证明的内容。

So this is what John Nash actually proved.

所以它们确实存在纳什均衡。

So, they do have a Nash equilibrium.

问题不在这里。

That's not the problem.

问题在于可能存在多个均衡点。

The problem is that there can be many.

于是就面临选择哪个均衡点的问题。

And then there's a question of which equilibrium to select.

如果你从不同的均衡中选择策略，而我选择我的策略，那意味着什么？

And if you select your strategy from a different equilibrium and I select mine, then what does that mean?

在这些非零和博弈中，我们可能会因为愚蠢而失去一些共同利益。

And in these non zero sum games, we may lose some joint benefit by being just simply stupid.

如果我们采取其他行动，实际上双方都能获得更好的结果。

We could actually both be better off if we did something else.

是的。

Yes.

在三人游戏中，还会出现诸如共谋等其他问题。

And in three player you get other problems also like collusion.

比如也许你我可以联合起来对付第三方玩家，通过共谋我们能取得显著优势。

Like maybe you and I can get up on a third player and we can do radically better by colluding.

因此这里会出现很多复杂问题。

So there are lots of issues that come up there.

与你合作的学生No Brown在Reddit的AMA中提到，我查阅了相关内容，他表示扑克玩家间的协作能力将改变游戏格局。

So No Brown, the student you worked with on this has mentioned, I looked through the AMA on Reddit, he mentioned that the ability of poker players to collaborate will make the game.

有人提问你们将如何设计扑克游戏，使其超越当前AI方法的可解范围？

He was asked the question of how would you make the game of poker, or both of you were asked the question, how would you make the game of poker beyond being solvable by current AI methods?

他回答说让扑克变得更难的方法不多，但玩家间的合作会使其变得极其困难。

And he said that there's not many ways of making poker more difficult, but collaboration or cooperation between players would make it extremely difficult.

你能解释一下这背后的原理吗？

So can you provide the intuition behind why that is?

如果你同意这个观点的话。

If you agree with that Yeah.

我在碰撞博弈方面做了大量研究，目前有一篇与我的学生Gabriela Farina及NIPS的其他合作者共同完成的论文。

So I've done a lot of work collisional games and we actually have a paper here with my other student, Gabriela Farina and some other collaborators at NIPS on that.

实际上我刚从展示这项研究的海报会议回来。

Actually just came back from the poster session where we presented this.

所以当出现共谋时，问题性质就不同了。

So, when you have a collusion, it's a different problem.

而且通常情况会变得更加棘手。

And it typically gets even harder then.

即便是博弈表示法，有些表示法确实不利于进行有效计算。

Even the game representations, some of the game representations don't really allow good computation.

为此我们专门引入了一种新的博弈表示方法。

So we actually introduced a new game representation for that.

这种合作行为是模型的一部分吗？

Is that kind of cooperation part of the model?

你们是否掌握其他玩家正在合作的信息？

Do you have information about the fact that other players are cooperating?

还是说这完全是个一无所知的混沌状态？

Or is it just this chaos where nothing is known?

有些信息是未知的。

Some things unknown.

你能

Can you

举个共谋类博弈的例子吗？或者这类情况通常

give an example of a collusion type game or is it usually

就像桥牌一样。

So like Bridge.

所以想想桥牌。

So think about Bridge.

就像我们同在一个团队时，我们的收益是相同的。

It's like when you and I are on a team, our payoffs are the same.

问题在于我们不能交流。

The problem is that we can't talk.

所以当我拿到牌时，我不能悄悄告诉你我有什么牌，那是不被允许的。

So when I get my cards, I can't whisper to you what my cards are, that would not be allowed.

因此我们必须提前协调策略，而且只能提前协调。

So we have to somehow coordinate our strategies ahead of time and only ahead of time.

然后我们可以讨论某些信号，但这些信号必须让对方团队也能理解。

And then there are certain signals we can talk about, but they have to be such that the other team also understands them.

这就是一个协调机制已经内置于游戏规则中的例子。

So, that's an example where the coordination is already built into the rules of the game.

但在许多其他情况下，比如拍卖、谈判、外交关系、扑克等，这种机制并没有内置，但对共谋者仍然非常有用。

But in many other situations like auctions or negotiations or diplomatic relationships, poker, it's not really built in, but it still can be very helpful for the colloders.

我在某处读到过你，

I've read you somewhere,

在谈判时，你会带着事先准备好的策略来到谈判桌前，比如你愿意做什么、不愿意做什么之类的。

when negotiations you come to the table with prior, like a strategy that you're willing to do and not willing to do, those kinds of things.

那么现在如何从扑克转向其他应用领域，比如谈判？如何开始将其应用到其他领域，甚至是你曾经工作过的现实世界领域？

So how do you start to now moving away from poker, moving beyond poker into other applications like negotiations, how do you start applying this to other domains, maybe even real world domains that you've worked on?

是的，我其实有两家初创公司专门做这个。

Yeah, I actually have two startup companies doing exactly that.

一家叫Strategic Machine，主要做商业应用、游戏、体育等各种相关领域。

One is called Strategic Machine and that's for kind of business applications, gaming, sports, all sorts of things like that.

这些技术在商业、体育、游戏以及金融、电力市场等各类领域都有应用。

Any applications of this to business and to sports and to gaming, to various types of things in finance, electricity markets and so on.

另一家叫Strategy Robot，我们将这些技术应用于军事安全、网络安全和情报领域。

And the other is called Strategy Robot where we are taking these to military security, cybersecurity and intelligence applications.

我想你之前做过一些...怎么说来着

I think you worked a little bit in, how do

就是广告推荐这类工作，对

you put it, advertisement sort of suggesting ads kind of thing, Yes,

那是另一家公司，叫optimised markets。

that's another company, optimised markets.

Optimised markets。

Optimised markets.

但那更多是关于组合市场和基于优化的技术。

But that's much more about a combinatorial market and optimisation based technology.

并没有使用这些博弈论推理技术。

That's not using these game theoretic reasoning technologies.

明白了。

I see.

好的，那么从高层次来看，你认为我们运用博弈论概念来建模人类行为的能力如何？

Okay, so what sort of high level do you think about our ability to use game theoretic concepts to model human behavior?

人类行为在扑克游戏之外也能适用这种建模方式吗？

Do human you behavior is amenable to this kind of modeling outside of the poker games?

你在工作中见过哪些成功应用案例？

And where have you seen it done successfully in your work?

我不确定。

I'm not sure.

真正的目标是模拟人类行为。

The goal really is modeling humans.

比如在零和游戏中，我其实不在乎对手是否遵循我的理性行为模型，因为如果他们不遵循，对我反而更有利。

Like, for example, if I'm playing a zero sum game, I don't really care that the opponent is actually following my model of rational behavior because if they're not, that's even better for me.

对吧。

Right.

所以在游戏中对阵时，前提是你需要将互动形式化

So see with the opponents in games, the prerequisite is that you formalize the

为某种可分析的方式。

interaction in some way that can be amenable to analysis.

你在机制设计方面做了出色工作，设计了能产生特定结果的游戏。

And you've done this amazing work with mechanism design, designing games that have certain outcomes.

那我举个自动驾驶领域的例子。

But so I'll tell you an example from my from my world of autonomous vehicles.

对吧？

Right?

我们研究行人，行人与车辆通过这种非语言交流进行协商。

We're studying pedestrians, and pedestrians and cars negotiate in this nonverbal communication.

这是一种奇怪的紧张游戏舞蹈，行人基本上在说，我相信你不会撞我。

There's this weird game dance of tension where pedestrians are basically saying, I trust that you won't kill me.

因此作为一个乱穿马路者，我会走上马路尽管这是违法的，而且存在这种紧张感。

And so as a jaywalker, I will step onto the road even though I'm breaking the law and there's this tension.

问题在于，在尝试模拟意图时，我们真的不知道如何很好地模拟这种情况。

And the question is, we really don't know how to model that well in in trying to model intent.

所以人们有时会提出博弈论等想法。

And so people sometimes bring up ideas of game theory and so on.

你认为人类行为的这种方面可以使用这类不完美信息方法进行建模吗？

Do you think that aspect of human behavior can use these kinds of imperfect information approaches, modeling?

当你甚至不知道如何设计游戏来描述这种情况以便解决它时，我们该如何开始解决这样的问题？

How do we how do you start to attack a problem like that when you don't even know how the game design the game to describe the situation in order to solve it?

好的。

Okay.

我其实没怎么思考过乱穿马路的问题，但我认为在自动驾驶车辆中一个很好的应用场景如下。

So I haven't really thought about jaywalking, but one thing that I think could be a good application in autonomous vehicles is the following.

假设你有不同公司运营的自动驾驶车队。

So let's say that you have fleets of autonomous cars operating by different companies.

比如这边是Waymo车队，那边是Uber车队。

So maybe here's the Waymo fleet and here's the Uber fleet.

如果你思考交通规则，它们定义了某些法律规则，但仍留下了巨大的策略空间。

If you think about the rules of the road, they define certain legal rules, but that still leaves a huge strategy space open.

举个简单例子，当车辆并道时，你知道人类如何并道——他们会减速、互相观察然后尝试并道。

As a simple example, when cars merge, you know, how humans merge, you know, they slow down and look at each other and try to merge.

如果这些情况能预先重新协商，让我们能全速合并，明确现状和操作方式，一切不就更快了吗？这样不是更好吗？

Wouldn't it be better if these situations would already be renegotiated so we can actually merge at full speed and we know that this is the situation, this is how we do it and it's all going to be faster.

但需要手动协商的情况实在太多了。

But there are way too many situations to negotiate manually.

所以你可以采用自动化协商。

So you could use automated negotiation.

至少这是个思路。

This is the idea at least.

你可以用自动化协商提前处理所有这些或大部分情况。

You could use automated negotiation to negotiate all of these situations or many of them in advance.

当然，可能有时候你不会总是让我先走。

And of course, it might be that hey, maybe you're not going to always let me go first.

也许你会说，好吧，在这些情况下我让你先走，但作为交换你要给我更多，或者在其他情况下让我先走。

Maybe you said okay, well, in these situations I'll let you go first, but in exchange you're going to give me too much, you're going to let me go first in this situation.

所以这是个庞大的组合式协商。

So, it's this huge combinatorial negotiation.

你认为在那个合并案例中，是否有空间将整个情况建模为不完全信息博弈，还是你更倾向于视为完全信息？

And do you think there's room in that example of merging to model this whole situation as an imperfect information game or do you really want to consider it to be a perfect?

不，这是个好问题。

No, that's a good question.

对，这是个好问题。

Yeah, that's a good question.

你是否要为假设自己并非全知而付出代价？

Do you pay the price of assuming that you don't know everything?

是啊，我也不清楚。

Yeah, I don't know.

这确实简单多了。

It's certainly much easier.

完全信息博弈要容易得多。

Games with perfect information are much easier.

所以如果可行的话，你应该这么做。

So if you can get away with it, you should.

但如果实际情况信息不完整，那你就得处理信息不完整的问题。

But if the real situation is of imperfect information, then you're going to have to deal with imperfect information.

太好了。

Great.

那么你从中吸取了哪些经验教训？

So what lessons have you learned?

年度计算机扑克大赛。

The annual computer poker competition.

人工智能的一项惊人成就。

An incredible accomplishment of AI.

你知道，回顾深蓝、AlphaGo的历史，这些时刻都是人工智能在工程与科学双重努力下超越人类顶尖选手的里程碑。

You know, you look at the history of Deep Blue, AlphaGo, these kind of moments when AI stepped up in an engineering effort and a scientific effort combined to to beat the best of human players.

那么从这次经历中你获得了什么？

So what what do you take away from this whole experience?

关于设计这类游戏的人工智能系统，你学到了什么？

What have you learned about designing AI systems that play these kinds of games?

这对人工智能整体而言，对未来AI发展意味着什么？

And what does that mean for sort of AI in general for the future of AI development?

是的，这是个很好的问题。

Yeah, so that's a good question.

关于这方面有很多可说的。

There's so much to say about it.

我确实喜欢这种以性能为导向的研究。

I do like this type of performance oriented research.

虽然在我的团队里，我们从构思到理论、实验、大型系统部署再到商业化全程参与。

Although in my group we go all the way from idea to theory, experiments, to big system fielding, to commercialisation.

所以我们覆盖了整个研究光谱。

So we span that spectrum.

我认为在AI的很多情况下，你必须先构建大型系统并进行规模化评估，才能真正知道哪些方法有效。

I think that in a lot of situations in AI, you really have to build the big systems and evaluate them at scale before you know what works and doesn't.

我们在计算博弈论领域就见过这种情况，很多技术在小规模时表现良好，但大规模时就失效了。

And we've seen that in the computational game theory community, that there are a lot of techniques that look good in the small, but then they cease to look good in the large.

我们还发现很多技术在理论上看起来更优越——我指的是像一阶方法这样的收敛速度。

And we've also seen that there are a lot of techniques that look superior in theory, and I really mean in terms of convergence rates, like first order methods.

更好的收敛速度，比如基于CFR的算法，但实际上基于CFR的算法才是最快的。

Better convergence rates, like the CFR based algorithms, yet the CFR based algorithms are the fastest in practice.

这确实告诉我必须在现实中测试这些方法。

So it really tells me that you have to test this in reality.

可以说现有理论还不够完善，无法准确判断哪种算法更优越。

The theory isn't tight enough, if you will, to tell you which algorithms are better than the others.

你必须从宏观角度看待这些问题，因为在这个领域，任何从小范围做出的预测都极可能产生误导。

And you have to look at these things in the large because any sort of projections you do from the small can at least in this domain be very misleading.

这是从科学与工程的角度来看。

So that's from a science and engineering perspective.

从个人角度而言，我们组织的首次扑克比赛——人脑对决AI的人机扑克大赛，真是一次疯狂的经历。

From a personal perspective, it's been just a wild experience with the first poker competition, the first brains versus AI, man machine poker competition that we organised.

顺便说一句，其他类型的扑克比赛之前有过，但这是首次举办单挑无限注的比赛。

Had been, by the way, for other poker games there had been previous competitions, but this was for heads up, no limit, this was the first.

而我可能因此成了扑克界最招人恨的家伙。

And I probably became the most hated person in the world of poker.

我本无意如此。

I didn't mean to.

为什么会这样？

Why is that?

为了某些目的破解游戏。

Cracking the game for something.

是啊。

Yeah.

很多人都觉得这对整个游戏、游戏的存在构成了真正的威胁。

A lot of people felt that it was a real threat to the whole game, the whole existence of the game.

如果AI变得比人类更强，人们会害怕玩扑克，因为那些超人般的AI四处游走，夺走他们的钱等等。

If AI becomes better than humans, people would be scared to play poker because there are these superhuman AIs running around taking their money and all of that.

所以，这确实非常激进。

So, it's just really aggressive.

那些评论超级激进。

The comments were super aggressive.

我收到了除死亡威胁之外的所有恶毒言论。

I got everything just short of death threats.

你觉得

Do you

国际象棋也是这样吗？

think the same was true for chess?

因为就在最近他们刚结束国际象棋世界锦标赛，人类开始刻意忽略现在有AI系统表现优于人类的事实，他们依然享受这个游戏，

Because right now they just completed the world championships in chess and humans just started ignoring the fact that there's AI systems now that outperform humans and they still enjoy the game,

这依然是个美妙的游戏。

it's still a beautiful game.

这就是我的看法。

That's what I think.

我认为扑克界也在发生同样的事。

And I think the same thing happens in poker.

所以我不认为自己会成为扼杀这个游戏的人，事实上我也没做到。

And so I didn't think of myself as somebody who was going to kill the game and I don't think I did.

我真的学会了热爱这个游戏。

I've really learned to love this game.

我以前不是扑克玩家，但从这些AI身上学到了许多精妙之处，顺便说一句，它们彻底改变了游戏的玩法。

I wasn't a poker player before, but learned so many nuances about it from these AIs and they've really changed how the game is played, by the way.

所以，它们用这些非常外星人式的打法玩扑克，而顶尖人类选手现在正把这些策略融入自己的玩法中。

So, they have these very Martian ways of playing poker and the top humans are now incorporating those types of strategies into their own play.

因此，在我看来，我们的工作反而让扑克对人类而言成为更丰富、更有趣的游戏，而不是让人们完全远离它。

So, if anything, to me, our work has made poker a richer, more interesting game for humans to play, Not something that is going to steer humans away from it entirely.

对你刚才提到的一点，我想简短评论一下——在学术界，这种情况有时确实相当罕见。

Just a quick comment on something you said, is, if I may say so in academia, is a little bit rare sometimes.

像你描述的那样将自己的想法付诸实践检验，这需要相当大的勇气。

It's pretty brave to put your ideas to the test in the way you described.

承认有些好点子在实际大规模应用时可能行不通。

Saying that sometimes good ideas don't work when you actually try to apply them at scale.

那么这种理念源自何处呢？

So where does that come from?

我的意思是，如果要给人们建议，是什么在驱动着你这样做？

I mean, if you could do advice for people, what drives you in that sense?

你一直都是这样的吗？

Were you always this way?

我想说的是，这确实需要勇气，去验证自己的想法，看它是否真能对抗人类顶尖选手等等。

I mean, it takes a brave person, I guess is what I'm saying, to test their ideas and to see if this thing actually works against human, top human players and so on.

我不确定是否勇敢，但这确实需要大量工作。

I don't know about Brave, but it takes a lot of work.

组织大型活动之类的事情需要耗费大量时间和精力。

It takes a lot of work and a lot of time to organize, make something big and to organize an event and stuff like that.

是什么驱使你付出这些努力？

And what drives you in that effort?

因为我认为，即使不做这些，你依然可以像2017年那样获得NIPS最佳论文奖。

Because you could still, I would argue, get a Best Paper Award at NIPS as you did in 2017 without doing this.

没错，是的。

That's right, yes.

所以总的来说，我认为在现实世界中大规模地做事非常重要。

So So, in general, I believe it's very important to do things in the real world and at scale.

这才是真正见真章的地方，如果你愿意这么理解的话。

And that's really where pudding, if you will.

事实胜于雄辩。

Proof is in the pudding.

关键就在这里。

That's where it is.

在这个特定案例中，多年来不同团队之间一直存在某种竞赛，看谁能率先在无限注德州扑克单挑中击败顶尖人类选手。

In this particular case, it was kind of a competition between different groups for many years as to who can be the first one to beat the top humans at Heads Up No Limit Texas Hold'em.

所以这演变成了一场看谁能率先达标的竞赛。

So it became kind of like a competition who can get there.

是啊。

Yeah.

看来良性的竞争确实能极大推动进步。

So a little friendly competition could do wonders for progress.

是的，完全正确。

Yes, absolutely.

机制设计这个话题非常有趣，对我来说也比较新颖——虽然我作为观察者接触过政治等领域，但你们论文中提出的自动化机制设计是我刚读到的。

So the topic of mechanism design, which is really interesting, also kind of new to me, except as an observer of, I don't know, politics and any I'm an observer of mechanisms, but you you write in your paper an automated mechanism design that that I quickly read.

所谓机制设计，就是通过制定游戏规则来达成特定理想结果。

So mechanism design is designing the rules of the game so you get a certain desirable outcome.

你正在以自动化的方式完成这项工作，而非通过精细调整。

And you have this work on doing so in an automatic fashion as opposed to fine tuning it.

那么你从这些努力中学到了什么？

So what have you learned from those efforts?

如果你观察，比如说，我不确定，像我们的政治体系这样的复杂系统。

If you look, say, I don't know, at complexes like our political system.

我们能否设计政治体系，使其以自动化方式产生我们期望的结果？

Can we design our political system to have in an automated fashion to have outcomes that we want?

我们能否设计像智能交通灯这样的东西，使其实现我们想要的结果？

Can we design something like traffic lights to be smart where it gets outcomes that we want?

你从这项工作中得出了哪些经验教训？

So what are the lessons that you draw from that work?

是的，所以我仍然非常坚信

Yeah, so I still very much believe in

自动化机制设计这个方向。

the automated mechanism design direction.

但它并非万能药。

But it's not a panacea.

机制设计领域存在不可能性结果，表明在C类中不存在能实现目标X的机制。

There are impossibility results in mechanism design, saying that there is no mechanism that accomplishes objective X in class C.

因此，在机制设计中，使用任何工具——无论是手动还是自动——都无法完成某些特定任务。

So there's no way using any mechanism design tools, manual or automated, to do certain things in mechanism design.

你能再描述一遍吗？

Can you describe that again?

所以意思是那是不可能实现的吗？

So meaning it's impossible to achieve that?

是的，这几乎是不可能的。

Yeah, there's certainly unlikely.

不可能。

Impossible.

所以这些并不是关于人类智慧可能想出聪明办法的陈述。

So these are not statements about human ingenuity who might come up with something smart.

这些证明表明，如果你想在类别C中实现属性X，任何机制都无法做到。

These are proofs that if you want to accomplish properties X in class C, that is not doable with any mechanism.

自动机制设计的好处在于，我们实际上并不是为整个类别设计的。

The good thing about automated mechanism design is that we're not really designing for a class.

我们每次都是针对特定情境进行设计。

We're designing for specific settings at a time.

因此，即使对整个类别存在不可能性结果，也不意味着该类别中的所有情况都不可能。

So even if there's an impossibility result for the whole class, it just doesn't mean that all of the cases in the class are impossible.

这只意味着其中某些情况是不可能的。

It just means that some of the cases are impossible.

所以我们实际上可以在这些已知不可能的类别中开辟出可能的岛屿。

So we can actually carve these islands of possibility within these known impossible classes.

而且我们确实做到了这一点。

And we've actually done that.

机制设计领域最著名的成果之一是罗杰·迈尔森和马克·萨特思韦特于1983年提出的迈尔森-萨特思韦特定理。

One of the famous results in mechanism design is the Myerton Sathethwaite theorem by Roger Myerton and Marc Sathethwaite from 1983.

在信息不完善的情况下，高效交易是不可能的。

It's an impossibility of efficient trade under imperfect information.

我们证明在许多情况下你仍可以避免这种情况，实现高效交易。

We show that you can in many settings avoid that and get efficient trade anyway.

这取决于你如何设计游戏规则。

Depending on how you design the game.

取决于

Depending

你如何设计游戏规则。

on how you design the game.

当然，这并不以任何方式否定不可能性结果。

And of course, it doesn't in any way contradict the impossibility result.

不可能性结果依然存在，但它只是在这个不可能类别中找到了某些特定点，在这些点上不存在不可能性。

The impossibility result is still there, but it just finds spots within this impossible class where in those spots you don't have the impossibility.

抱歉如果我有点哲学化，但正如我提到的，你认为这些对政治或人际互动有什么启示？比如设计机制不仅适用于交易、拍卖或纯粹形式化的游戏，还包括政治体系这样的人际互动。

Sorry if I'm going a bit philosophical, but what lessons do you draw towards, like I mentioned, politics or human interaction and designing mechanisms for outside of just these kinds of trading or auctioning or purely formal games or human interaction, like a political system.

你认为它能如何应用于政治、商业、谈判这类事务？设计能产生特定结果的规则。

What how can do you think it's applicable to, yeah, politics or to business, to negotiations, these kinds of things, designing rules that have certain outcomes.

是的。

Yeah.

是的，我确实这么认为。

Yeah, I do think so.

你见过成功实施的案例吗？

Have you seen that successfully done?

其实并没有。

It hasn't really.

哦，你是指机制设计还是自动化机制？

Oh, you mean mechanism design or automated mechanism?

自动化机制设计。

Automated mechanism design.

到目前为止，机制设计本身的成功相当有限。

So mechanism design itself has had fairly limited success so far.

虽然存在某些案例，但现实世界中的大多数情况从机制设计的角度来看实际上并不完善。

There are certain cases, but most of the real world situations are actually not sound from a mechanism design perspective.

即使在那些由精通机制设计的专家设计的案例中，人们通常也只是从理论中汲取一些见解并应用于现实，而非直接应用机制本身。

Even in those cases where they've been designed by very knowledgeable mechanism design people, the people are typically just taking some insights from the theory and applying those insights into the real world rather than applying the mechanisms directly.

一个著名的例子就是FCC频谱拍卖。

So, one famous example is the FCC spectrum auctions.

我也曾参与其中，扮演了一个小角色。

I've also had a small role in that.

许多精通博弈论的优秀经济学家一直在研究这个问题。

Very good economists have been working, excellent economists have been working on that who know game theory.

然而实际设计的规则却使得诚实竞价并非最佳策略。

Yet the rules that are designed in practice there, they're such that bidding truthfully is not the best strategy.

通常在机制设计中，我们试图让参与者更容易操作，因此诚实是最佳策略。

Usually in mechanism design we try to make things easy for the participants, so telling the truth is the best strategy.

但即便是在涉及数百亿美元频谱拍卖的高风险场景中，诚实仍然不是最佳策略。

But even in those very high stakes auctions where you have tens of billions of dollars worth of spectrum being auctioned, truth telling is not the best strategy.

顺便说一句，目前还没有人知道这些拍卖中的任何一种最优竞价策略。

And by the way, nobody knows even a single optimal bidding strategy for those auctions.

制定最优出价的挑战是什么？

What's the challenge of coming up with an optimal bid?

因为参与者众多且信息不完全

Because there's a lot of players and there's imperfectly

不是说很多

It's not to say a lot

参与者，而是拍卖品数量庞大。

of players, but many items for sale.

这些机制设计得即使只有两件或一件拍品，如实出价也并非最佳策略。

And these mechanisms are such that even with just two items or one item, bidding truthfully wouldn't be the best strategy.

纵观AI发展史，它是由一系列里程碑事件标记的。

If you look at the history of AI, it's marked by seminal events.

AlphaGo成为世界围棋冠军后，我认为LeBratis在无限注德州扑克单挑赛中获胜也是这类标志性事件之一。

And AlphaGo being a world champion, human Go player, I would put LeBratis winning the heads up no limit, hold them as one of such event.

谢谢。

Thank you.

那你认为

And what what do you think

下一个类似事件会是什么？

is the next such event?

无论是在你个人研究领域还是整个AI界，你觉得未来可能出现哪些让世界震惊的突破？

Whether it's in your life or in the broadly AI community that you think might be out there that would surprise the world?

这是个好问题，但我确实不知道答案。

So, that's a great question and I don't really know the answer.

在博弈求解领域，无限注德州扑克单挑模式确实是公认的最后一道基准测试。

In terms of game solving, Heads Up No Limit Texas Hold'em really was the one remaining widely agreed upon benchmark.

所以那是个重大里程碑。

So that was the big milestone.

那么，还有其他挑战吗？

Now, are there other things?

当然，肯定存在其他挑战。

Yes, certainly there are.

但学界尚未就某个特定问题达成共识。

But there is not one that the community has kind of focused on.

那么其他可能的方向是什么？

So what could be other things?

有团队在研究《星际争霸》。

There are groups working on StarCraft.

有团队在研究《DOTA2》。

There are groups working on DOTA two.

这些都是电子游戏。

These are video games.

或者像《外交博弈》《花火》这类桌游。

Or you could have Diplomacy or Hanabi, things like that.

这些都是娱乐性游戏，但尚未有哪个被公认为下一个核心挑战问题。

These are recreational games, but none of them are really acknowledged as kind of the main next challenge problem.

就像国际象棋、围棋或无上限德州扑克那样。

Like chess or go or heads up no limit Texas Hold'em was.

所以在游戏求解领域，我并不确定下一个标杆会是什么或何时出现。

So I don't really know in the game solving space what is or what will be the next benchmark.

我有点希望会有下一个标杆，因为过去十年里，不同团队研究同一问题确实极大地推动了这些与应用无关的技术快速发展。

I kind of hope that there will be a next benchmark cause really the different groups working on the same problem really drove these application independent techniques forward very quickly over ten years.

你认为是否存在一个让你兴奋的开放性问题，能让你从游戏转向现实世界的博弈，比如股票市场交易？

Do you think there's an open problem that excites you that you start moving away from games into real world games like say the stock market trading.

是的，这差不多就是我的风格。

Yeah, so that's kind of how I am.

所以我可能不会在这些娱乐性标杆上投入太多精力了。

So I am probably not going to work as hard on these recreational benchmarks.

我正在创办两家关于游戏求解技术的初创公司——战略机器和策略机器人。

I'm doing two startups on game solving technology, strategic machine and strategy robot.

我们真正感兴趣的是将这些技术投入实际应用。

We're really interested in pushing this stuff into practice.

你觉得

What do you

什么会是你认为真正强大且令人惊讶的成果？

think would be really you know, a powerful result that would be surprising?

比方说五年、十年后，从统计学角度看不太可能实现，但如果出现突破，可能会达成什么？

That would be, if you can say, mean, you know, five years, ten years from now, something that statistically you would say is not very likely, but if there's a breakthrough, what achieve?

是的，我认为总体而言，我们在博弈论领域所处的局面与机器学习领域截然不同。

Yeah, so I think that overall, we're in a very different situation in game theory than we are in, let's say, machine learning.

在机器学习领域，这是一项相当成熟的技术，应用广泛，并在现实世界中取得了显著成功。

So in machine learning, it's a fairly mature technology and it's very broadly applied and proven success in the real world.

在游戏求解方面，目前几乎还没有实际应用。

In game solving, there are almost no applications yet.

我们刚刚实现了超越人类的能力，而机器学习可以说在90年代甚至更早就已经做到了这一点。

We have just become superhuman, which machine learning you could argue happened in the 90s, if not earlier.

至少在特定复杂的监督学习应用上是如此。

At least on learning, certain complex supervised learning applications.

现在，我认为下一个挑战性问题——虽然你并非从这个角度提问，你问的是技术突破——

Now, I think the next challenge problem, I know you're not asking about it this way, you're asking about technology breakthrough.

但我认为重大突破在于能够证明，或许大部分军事规划或商业战略都将通过计算博弈论来策略性地完成。

But I think the big breakthrough is to be able to show that maybe most of, let's say, military planning or most of business strategy will actually be done strategically using computational game theory.

这是我期望在未来五到十年内实现的目标。

That's what I would like to see as a next five or ten year goal.

也许你可以再解释一下，请原谅如果这是个显而易见的问题：机器学习方法如神经网络存在不透明、不可解释的缺陷。

Maybe you can explain to me again, forgive me if this is an obvious question, but machine learning methods, neural networks suffer from not being transparent, not being explainable.

博弈论方法，比如纳什均衡，当你看到不同的解决方案时——比如在讨论军事行动时——那些战略方案是否合理？

Game theoretic methods, you know, Nash equilibria, do they generally when you see the different solutions, are they when when you talk about military operations, are they once you see the strategies, do they make sense?

它们是可解释的，还是存在与神经网络相同的问题？

Are they explainable or do they suffer from the same problems as neural networks do?

这是个很好的问题。

So that's a good question.

我的回答可以说是既肯定又否定。

I would say a little bit yes and no.

我的意思是，这些博弈论策略，比如纳什均衡，具有可证明的特性。

And what I mean by that is that these game theoretic strategies, let's say Nash equilibrium, it has provable properties.

所以它不像深度学习那样，你只能祈祷它能奏效。

So it's unlike, let's say, deep learning where you kind of cross your fingers, hopefully it'll work.

即使在事后获得权重参数后，你仍在祈祷它能正常工作。

And then after the fact, you have the weights, you're still crossing your fingers and I hope it will work.

而在这里，你可以确信解决方案的质量是有保障的。

Here, you know that the solution quality is there.

存在可证明的解决方案质量保证。

There's provable solution quality guarantees.

不过，这并不意味着这些策略是人类可以理解的。

Now, that doesn't necessarily mean that the strategies are human understandable.

那完全是另一个问题。

That's a whole other problem.

我认为深度学习和计算博弈论在这方面处境相同。

I think that deep learning and computational game theory are in the same boat in that sense.

两者都难以理解。

Both are difficult to understand.

但至少博弈论技术有这些质量保证。

But at least the game theoretic techniques, they have these guarantees of quality.

你认为未来的商业运营、战略运营甚至军事行动，是否会成为自动化系统提出的有力候选方案？

Do you see business operations, strategic operations, even military in the future being at least the strong candidates being proposed by automated systems?

你看到这种趋势了吗？

Do you see that?

是的，我有。

Yeah, I do.

我有。

I do.

但那更多

But that's more

是一种信念而非确凿的事实。

of a belief than a substantiated fact.

根据你持乐观还是悲观态度，对我来说那是个令人兴奋的未来，特别是在可证明的最优性方面。

Depending on where you land in optimism or pessimism, that's a really to me, that's an exciting future, especially if there's provable things in terms of optimality.

展望未来，有些人担心——特别是你看扑克游戏，这可能是最后一个被解决的游戏基准。

So looking into the future, there's a few folks worried about the especially you look at the game of poker, which is probably one of the last benchmarks in terms of games being solved.

他们担心未来和人工智能带来的生存威胁。

They they worry about the future and the existential threats of artificial intelligence.

无论以何种形式对社会产生的负面影响。

So the negative impact in whatever form on society.

这是否同样让你担忧，还是你对AI的积极影响更为乐观？

Is that something that concerns you as much or are you more optimistic about the positive impacts of AI?

我对积极影响要乐观得多。

I am much more optimistic about the positive impacts.

就我自己的工作而言，我们目前运营着全国肾脏交换计划。

So just in my own work what we've done so far, we run the nationwide kidney exchange.

如今有数百人因此得以存活。

Hundreds of people are walking around alive today.

谁会不愿意呢？

Who wouldn't be?

而且它还增加了就业。

And it's increased employment.

现在有很多人在运营肾脏交换项目，并在移植中心与肾脏交换系统互动。

You have a lot of people now running kidney exchanges and at transplant centers, interacting with the kidney exchange.

还有额外的外科医生、护士、麻醉师、医院等等。

You have extra surgeons, nurses, anesthesiologists, hospitals, all of that.

就业因此增加，世界也变得更美好。

Employment is increasing from that and the world is becoming a better place.

另一个例子是组合采购拍卖。

Another example is combinatorial sourcing auctions.

2001年至2010年间，在我之前创办的名为CombineNet的初创公司中，我们进行了800次大规模组合采购拍卖。

We did 800 large scale combinatorial sourcing auctions from 2001 to 2010 in a previous startup of mine called CombineNet.

我们在这600亿美元的采购中，将供应链效率提高了12.6%。

And we increased the supply chain efficiency on that $60,000,000,000 of spend by 12.6%.

因此，全球效率提升了超过60亿美元。

So that's over $6,000,000,000 of efficiency improvement in the world.

这不是将价值从一方转移到另一方，而是纯粹的效率提升。

This is not like shifting value from somebody to somebody else, just efficiency improvement.

比如在卡车运输中，减少了空驶。

Like in trucking, less empty driving.

这样浪费更少，碳足迹也更小等等。

So there's less waste, less carbon footprint and so on.

这在短期内会产生巨大的积极影响。

This is a huge positive impact in the near term.

但需要再坚持一段时间，因为我认为博弈论在这里能发挥作用。

But to stay in it for a little longer because I think game theory has a role to play here.

让我再回头谈谈这个问题。

Let me actually come back on that.

这是一方面。

That is one thing.

我认为人工智能还将使世界变得更加安全。

I think AI is also going to make the world much safer.

这是另一个经常被忽视的方面。

So that's another aspect that often gets overlooked.

嗯，让

Well, let

我问这个问题。

me ask this question.

也许你可以谈谈安全性。

Maybe you can speak to the the safer.

我曾与马克斯·泰格马克和斯图尔特·罗素交流过，他们非常担心人工智能的生存威胁，通常关注的是价值错位问题。

So I talked to Max Tegmark and Stuart Russell, who are very concerned about existential threats of AI and often the concern is about value misalignment.

即人工智能系统运作时追求的目标与人类文明、人类利益不一致。

So AI systems basically working, operating towards goals that are not the same as human civilization, human beings.

因此博弈论似乎可以在确保价值观与人类一致方面发挥作用。

So it seems like game theory has a role to play there to to make sure the values are aligned with human beings.

我不确定你是否这样看待这个问题。

I don't know if that's how you think about it.

如果不是的话，你认为AI可能如何帮助解决这个问题？

If not, how do you think AI might help with this problem?

你认为AI可能如何让世界变得更安全？

How do you think AI might make the world safer?

是的。

Yeah.

我认为这种价值观错位更多是理论层面的担忧，在实际应用中我并未真正遇到过，因为我主要从事实际应用开发。

I think this value misalignment is a fairly theoretical worry and I haven't really seen it in it because I do a lot of real applications.

我在任何地方都没见过这种情况。

I don't see it anywhere.

我见过最接近的情况其实是这样的思维实验：80年代末我们开发运输优化系统时，有人听说提高资产利用率是个好主意。

The closest I've seen it was the following type of mental exercise really, where I had this argument in the late 80s when we were building these transportation optimisation systems and somebody had heard that it's a good idea to have high utilisation of assets.

于是他们对我说，嘿，你为什么不把这个设为目标呢？

So they told me, hey, why don't you put that as an objective?

我们最终没把它设为目标，因为我向他证明：如果把这个设为目标，解决方案就会是把卡车装满然后绕圈行驶。

And we didn't even put it as an objective because I just showed him that if you had that as your objective, the solution would be to load your trucks full and drive in circles.

这样什么货物都送达不了。

Nothing would ever get delivered.

却能实现100%的利用率。

Would have 100% utilization.

所以是的，我了解这种现象。

So yeah, I know this phenomenon.

我了解这一点已有三十多年，但从未在现实中见过它真正成为问题。

I've known this for over thirty years but I've never seen it actually be a problem in reality.

确实，如果目标设定错误，人工智能会将其优化到极致，造成的伤害将超过人类半吊子的尝试。

And yes, if you have the wrong objective, the AI will optimize that to the hilt and it's going to hurt more than some human who's kind of trying to

用人类的一些见解半途而废地解决它。

solve it in a half baked way with some human insight too.

但我就是没见到这种情况在实践中发生。

But I just haven't seen that materialise in practice.

你刚才非常清晰地指出了理论与现实之间的这道鸿沟。

There's this gap that you've actually put your finger on very clearly just now between theory and reality.

我觉得这很难用语言表达清楚。

That's very difficult to put into words, I think.

这是你理论上能想象到的最坏情况，甚至更糟。

It's what you can theoretically imagine, the the worst possible case or even yeah.

我是说，糟糕的情况，以及现实中通常发生的状况。

I mean, bad cases, and what usually happens in reality.

举个例子，对我来说——也许你能就此发表看法——我是在苏联长大的。

So for example, to me, maybe it's something you can comment on, having grown up in I had grown up in the Soviet Union.

你知道，目前世界上有一万枚核武器。

You know, there's currently 10,000 nuclear weapons in the world.

几十年来，从理论上看核战争没有爆发让我感到惊讶。

And for many decades, it's, theoretically surprising to me that the nuclear war is not broken out.

你会从博弈论的角度思考这个问题吗？

Do you think about this aspect from a game theoretic perspective in general?

为什么这是真的？

Why is that true?

为什么理论上你能预见事情会变得很糟，但实际上却并未发生？

Why in theory you could see how things would go terribly wrong and somehow yet they have not?

你怎么看

How do you think

这个问题？

about that?

我确实经常思考这个问题。

So I do think about that a lot.

我认为人类面临的最大两个威胁，一是气候变化，二是核战争。

I think the biggest two threats that we're facing as mankind, one is climate change and the other is nuclear war.

这就是我最担心的两件事。

So those are my main two worries that I worry about.

我曾尝试为气候问题做些事情，两次考虑过为气候变化采取行动。

And I've tried to do something about climate I thought about trying to do something for climate change twice.

实际上，我为两家初创企业委托过相关研究，但都没找到理想切入点，不过我仍在持续关注。

Actually, for two of my startups I've actually commissioned studies of what we could do on those things and we didn't really find a sweet spot, but I'm still keeping an eye out on that.

如果我们能通过市场方案、优化方案或其他技术手段解决问题的话。

If there's something where we could actually provide a market solution or optimization solution or some other technology solution to problems.

比如当时我们正在研究污染信用市场。

Right now, for example, pollution credit markets was what we were looking at then.

更多是这些市场缺乏政治意愿导致效果不佳，而非市场设计本身的问题。

And it was much more the lack of political will by those markets were not so successful rather than bad market design.

我可以进去设计一个更好的市场机制，但如果没有政治意愿，这对世界的影响微乎其微。

I could go in and make a better market design, but that wouldn't really move the needle on the world very much if there's no political will.

而在美国，至少芝加哥市场刚刚被关闭等等。

And in The US, at least the Chicago market was just shut down and so on.

所以，无论你的市场设计有多出色都无济于事。

So then it doesn't really help how great your market design was.

至于核问题方面，情况更严重。

And then the nuclear side, it's more.

全球变暖是个日益紧迫的问题。

So, global warming is a more encroaching problem.

核武器一直存在于此。

Nuclear weapons have been here.

这是个显而易见却长期悬而未决的问题。

It's an obvious problem that's just been sitting there.

那么你认为是什么机制设计让一切看起来如此稳定？

So how do you think about what is the mechanism design there that just made everything seem stable?

你现在仍然极度担忧吗？

And are you still extremely worried?

我仍然极度担忧。

I am still extremely worried.

你可能知道'相互保证毁灭'的简单博弈论。

So, you probably know the simple game theory of MAD.

这就是相互保证毁灭机制，它不需要任何复杂计算。

So, this was a mutually assured destruction and it doesn't require any computation.

对于小型矩阵，你实际上可以确信这个游戏的性质是没有人愿意主动发起。

With small matrices you can actually convince yourself that the game is such that nobody wants to initiate.

是的，这是一种非常粗略的分析，在存在两个超级大国或少数几个超级大国的情况下确实适用。

Yeah, that's a very coarse grained analysis and it really works in a situation where you have two superpowers or a small number of superpowers.

现在情况已经大不相同了。

Now things are very different.

你们拥有更小型的核武器，因此启动的门槛更低，而且会有更多小国和非国家行为体可能获得核武器等等。

You have a smaller NUC, so the threshold of initiating is smaller and you have smaller countries and non nation actors who may get NUCs and so on.

所以我认为现在的风险比

So I think it's riskier now than

以往任何时候都要大。

it was maybe ever before.

关于AI的想法和应用，你刚才稍微提到了一些，但眼下最让你兴奋的是什么？

And what idea, application of AI, you've talked about it a little bit, but what is the most exciting to you right now?

我的意思是，你现在在NeurIPS大会上，手头有几项出色的工作成果。

I mean, you're here at NIPS, NeurIPS now, You have a few excellent pieces of work.

但你正在合作的几家公司，未来有什么规划？

But what are you thinking into the future with several companies you're doing?

最令人兴奋或其中一项令人兴奋的事情是什么？

What's the most exciting thing or one of the exciting things?

目前对我来说最重要的是开发这些可扩展的游戏解决方案技术，并将它们应用到现实世界中。

The number one thing for me right now is coming up with these scalable techniques for game solving and applying them into the real world.

我仍然对市场设计非常感兴趣，我们正在优化市场中实践这一点。

I'm still very interested in market design as well and we're doing that in the optimized markets.

但我最关心的是，当前的首要任务是否是战略机器、战略机器人，将这项技术推广出去，并且鉴于你们正在一线进行应用开发，实际需要填补哪些空白，还有哪些技术缺口需要填补。

But I'm most interested if number one right now is strategic machine, strategy robot, getting that technology out there and seeing as you are in the trenches doing applications, what needs to be actually filled, what technology gaps still need to be filled.

所以光是翘着脚空想需要做什么是非常困难的。

So it's so hard to just put your feet on the table and imagine what needs to be done.

但当你真正进行实际应用开发时，应用本身会告诉你需要做什么。

But when you're actually doing real applications, the applications tell you what needs to be done.

我真的很享受这种互动过程。

And I really enjoy that interaction.

将你们正在研发的尖端技术应用到实际中是否充满挑战

Is it a challenging process to apply

让工业界、军方或其他真正能从中受益的各方实际采用这些技术。

some of the state of the art techniques you're working on and having the various players in industry or the military or people who could really benefit from it actually use it.

以自动驾驶汽车为例，我们与汽车公司合作，他们在很多方面都有些保守。

What's that process like of, you know, in autonomous vehicles, we work with automotive companies and they're in many ways are a little bit old fashioned.

这很困难。

It's difficult.

他们确实想使用这项技术。

They really want to use this technology.

这显然会带来重大效益，但在数据、计算能力等各方面，现有系统还不足以轻松实现技术整合。

This clearly will have a significant benefit, but the systems aren't quite in place to easily have them integrated in terms of data, in terms of compute, in terms of all these kinds of things.

所以这是你们面临的主要挑战之一吗？你们是如何应对这个挑战的？

So is that one of the bigger challenges that you're facing and how do you tackle that challenge?

是的，我认为这始终是个挑战。

Yeah, I think that's always a challenge.

这实际上是一种缓慢和惰性，就是我们总是按老方法做事。

That's kind of slowness and inertia really of let's do things the way we've always done it.

你只需要找到客户内部的支持者，他们明白未来不能一成不变。

You just have to find the internal champions at the customer who understand that things can't be the same way in the future.

否则糟糕的事情就会发生。

Otherwise bad things are going to happen.

在自动驾驶领域非常有趣的是，传统汽车制造商正在这样做。

And in autonomous vehicles, it's actually very interesting that the car makers are doing that and they're very traditional.

但与此同时，像谷歌和百度这样与汽车或交通毫无关系的科技公司却在大力推动自动驾驶汽车。

But at the same time, have tech companies who have nothing to do with cars or transportation, like Google and Baidu really pushing on autonomous cars.

我觉得这很迷人。

I find that fascinating.

显然你对这些想法能在世界上产生影响感到非常兴奋

Clearly you're super excited about actually these ideas having an impact in

。

the world.

在技术和研究方向上，还有哪些领域让你感到兴奋？

In terms of the technology, in terms of ideas and research, are there directions that you're also excited about?

无论是你谈到的处理不完全信息博弈的方法，还是将深度学习应用于这些问题。

Whether that's on the some of the approaches you talked about for imperfect information games, whether it's applying deep learning to some of these problems.

在研究方面有什么让你兴奋的内容吗？

Is there something that you're excited in the research side of things?

是的，在游戏解决方案中有很多不同的方向。

Yeah, lots of different things in the game solving.

所以，解决更大的博弈问题，那些玩家行动中有更多隐藏动作的游戏。

So, solving even bigger games, games where you have more hidden action of the player actions as well.

扑克是一种机会行为真正被隐藏的游戏。

Poker is a game where really chance actions are hidden.

或者其中一些被隐藏，但玩家的行动是公开的。

Or some of them are hidden, but the player actions are public.

各种类型的多人游戏，共谋、对手剥削，甚至更长的博弈。

Multiplayer games of various sorts, collusion, opponent exploitation, and even longer games.

基本上是无限进行的游戏，但它们不是重复的。

Games that basically go forever, but they're not repeated.

因此，寻找那些无限进行的扩展形式博弈。

So, seek extensive form games that go forever.

那会是什么样子呢？

What would that even look like?

你如何表示这种情况？

How do you represent that?

你如何解决这个问题？

How do you solve that?

这类游戏有什么例子吗？

What's an example of a game like that?

这是你提到的一些随机博弈吗？

This is some of the stochastic games that you mentioned?

比如说商业策略。

Let's say business strategy.

不仅仅是模拟某个特定的互动，而是要从现在到永远地思考业务。

And not just modeling a particular interaction, but thinking about the business from here to eternity.

或者说军事策略。

Or let's say military strategy.

所以战争并不会消失。

So it's not like war is going to go away.

你认为军事策略会永远持续下去吗？

Do you think about military strategy that's going to go forever?

你该如何建模呢？

How do you even model that?

你如何判断某人采取的行动是否正确？

How do you know whether a move was good that somebody made?

诸如此类。

And so on.

这大致是一个方向。

So that's kind of one direction.

我也对学习更可扩展的整数规划技术非常感兴趣。

I'm also very interested in learning much more scalable techniques for integer programming.

今年夏天我们在ICML上发表了第一篇具有理论泛化保证的自动算法配置论文。

So we had an ICML paper this summer on that for the first automated algorithm configuration paper that has theoretical generalization guarantees.

如果我看到这么多训练样本，并以这种方式调整算法，它将在未见过的真实分布上表现良好。

So if I see these many training examples and I tool my algorithm in this way, it's going to have good performance on the real distribution, which I have not seen.

有趣的是，算法配置研究至少已经认真进行了十七年。

Which is kind of interesting that algorithm configuration has been going on now for at least seventeen years seriously.

在此之前从未有过任何泛化理论。

And there has not been any generalization theory before.

嗯，真的非常令人兴奋，能和你交谈是我的莫大荣幸。

Well, is really exciting and it's a huge honor to talk to you.

非常感谢你，托马斯。

Thank you so much, Tomas.

感谢你将Labratis带到这个世界，以及你所做的所有杰出工作。

Thank you for bringing Labratis to the world and all the great work you're doing.

嗯，非常感谢你。

Well, thank you very much.

这很有趣。

It's been fun.

问得好。

Good questions.