那些值得坚守的工作

人们通常不会把判断力或决策力当作一种技能。

People don't tend to think of either judgment or decision making as skills.

决策应该成为一种技术，一种可以重复使用的技术。

Decision making should become a technology, a reusable technology.

毕竟，我们的决策和行动正是我们影响周围世界的方式。

After all, our decisions, our actions are how we impact the world around us.

问题是，要采用谁的判断？

The problem is whose human judgment?

作为组织，我们需要开始认真思考这个问题。

We need to, as organizations, start getting serious about asking that question.

谁有权做出判断？

Who gets to apply the judgment?

人工智能的前景是真实的。

The promise of AI is real.

在你的组织中让AI发挥作用，同样也是一项挑战。

So is the challenge of making it work in your organization.

欢迎收听《未来之路》，我们在这里探讨人工智能时代的工作前景。欢迎回到《未来之路》。

Welcome to The Path Forward, where we explore the future of work in the age of Welcome back to The Path Forward.

我是主持人克里斯蒂安·波茨。

I'm your host, Christian Potts.

今天非常高兴请到了UiPath的首席执行官丹尼尔·戴恩斯，以及卡西·科济尔科夫。

I'm really pleased to be joined today by UiPath CEO, Daniel Dines, and Cassie Kozyrkov.

我觉得这真的会是一场非常精彩的讨论，那么我们就……

Really, I think this is gonna be really fascinating discussion, so let...

我们直接开始吧。

Let's just get right into it.

丹尼尔，首先恭喜你出了这本书。

So, Daniel, first off, congratulations on the book.

我拿过了。

I held it.

我亲手拿着那本书看了看。

I held the copy in my hand.

看到那本书，真的挺酷的。

It was it was pretty cool to see.

跟我聊聊这个书名，《工作永存》吧。

Talk to me about the title, The Work Remains.

我们在书里探讨过一个观点：AI负责推理，人类负责决策，机器人负责执行。

We have this idea we talked about that explored in the book, which is AI reasons, humans decide, robot does the work.

跟我讲讲，这些决策具体指什么？

Talk to me about sort of, like, what are the decisions?

当你思考AI的运作方式以及人类所做的决策时，这个想法最初是怎么来的？

What's the the the the sort of, like, genesis of that when you think about the way that AI works and and the decisions humans are making?

你看，

Look.

你知道，当我开始写这本书的时候，我就挺担心现在那种“AI要抢走所有工作”的说法。

As you know, when I started writing this book, I I was kind of concerned of the narrative out there that AI is gonna take all the jobs.

因为如果你真按字面理解那句“我们会在数据中心里拥有数百万个爱因斯坦”，说实话，那我只会雇爱因斯坦了。

Because if you literally take the statement, we will have millions of Einstein's in a data center, I would hire only Einstein's, honestly.

这真的很简单。

It's it's so simple.

然后我就想，那什么会留下来呢？

And then I I went, what's gonna remain?

那里有什么？

What's there?

这是真的吗？

Is it true?

不是真的吗？

Is it not true?

我应该从字面上理解，还是不该？

Should I take it literally or not?

然后随着我逐渐了解真相，其实我和AI聊了很多，关于最新的发展，实验室里正在进行的创新。

And then as, you know, I unfolded the truth, was actually talking a lot with AI about the latest development, what's innovations happening in the lab.

所以我发现了一些AI的、有趣的、长期存在的局限性，这些局限性……

So I got to some kind of interesting durable limitations of AI that...

同时，人类也有持久的特质。

And also durable treats that humans have.

我的结论是，在很长一段时间内，这都不会改变。

And my conclusion is that it's for a long time from now on.

我们仍然需要人类和AI一起合作。

We still need humans and AI working together.

我在书里建立了一个完整的框架，就是AI提出、人类决定、自动化执行。

And I've built an entire framework in the book, which is this AI proposes human decide and automation execute.

这个框架是围绕这个建立的……

And this framework is built around this...

人类和AI各自的优势和局限。

The strengths and limitations of both humans and AI.

所以很明显，我们正处于一个时刻，那……

So it's pretty clear we're at a moment here where the...

如果你允许我这么说的话，AI这个精灵已经出了神灯。

If you'll permit the phrase, the AI genie is out of the lamp.

已经不可能回头了。

There's no going back from here.

卡西，当你想到这些的时候，你谈到AI已经在这里、正在发生、正在发展、正在进步这些想法时，你通常会跟人聊什么？

When you think about that, Cassie, what's the conversation you find yourself in when you talk about this idea of, like, it's here.

它已经发生了。

It's happening.

就是现在。

It's now.

它在发展。

It's growing.

它在进步。

It's progressing.

比如，我们到底应该多大程度依赖AI，同时又多大程度保护人的部分？

Like, how much should we be relying on versus how much should we be protecting sort of, like, the human piece of this?

我觉得这个框架本身就是短视的。

I think even that framing is very short term.

我希望我们能把眼光放得更长远一些。

I want us to think a little bit longer term about this.

所以从技术的角度来看，非常有意思的一点是，通过自动化语言，我们第一次不需要提前知道客户的问题是什么，就能给他们提供解决方案。

So what's really fascinating from a technology point of view is that kind of for the first time by automating language, we don't need to know what the customer's problem is in advance of offering them a solution to.

这多棒啊，我……

Like, how fantastic that I...

他们不知道我要问什么，但能给我解决方案。

They don't know what I'm gonna ask for, but they can give me the solution.

我可以问任何问题。

And I can ask for anything.

也许这个解决方案是垃圾网络。

And maybe the solution's garbage networks.

也许它根本不管用。

Maybe it doesn't work at all.

但关键是，现在你可以突然提供这种通用性的东西。

But the point is now you can suddenly offer the general purpose thing.

这让我感觉特别像一个数字神灯。

And that feels to me very much like a digital genie.

而在实现这个目标的过程中，有一个很有意思的地方是，当你还没有这个神灯的时候，你怎么去把它造出来？

And in the journey to making that happen for humankind, there's obviously a fascination with when you don't have the genie yet, how do we build the genie?

所有的工程师和数据人员都在为此努力，也许他们自己都不确定这就是他们追求的那个愿景。

And all the engineers and data people are all running at this, maybe not even sure that that's the vision that they're going for.

但想想能造出这样的东西，是多么令人兴奋啊。

But how intoxicating it is to think about building something like that.

语言的自动化，人类协作通用界面的自动化。

The automation of language, the automation of the universal interface for human collaboration.

太酷了。

How cool.

而作为这个旅程的一部分，我不得不提醒自己，神灯本身并不是重点。

And having been part of this journey, I had to force myself to remember that the genie is not actually the point.

从来都不是。

It never was.

在故事里不是这样，在我们的技术征程中也不是这样。

Not in the stories and not in our technological journey.

只是当你还没有它的时候，就能理解为什么大家都想得到它。

It's just when you don't have it yet, you can see why everyone wants to go for it.

现在，如果我们看看硅谷为什么如此兴奋，以及大部分的讨论和资金都集中在什么上，那就是谁有哪个神灯？

Now, if we look at what Silicon Valley is so excited about and what most of the conversation and most of the funding is about, it is who's got which genie?

哪个神灯这周出来了，是 Opus 吗？

And which genie came out this week and, you know, is it Opus?

是 Fable 吗？

Is it Fable?

还是……

Is it...

你懂的。

You know?

如果这期播客发布的时候，中间又冒出一堆别的神灯，那这话可能就不太好说了。

And this isn't gonna age well if if this podcast comes out and there's a whole bunch of other genies in the meantime.

可能就一周之后的事儿了。

Which could be in a week's time.

甚至连一周都撑不过去。

It doesn't mean not even the wake of the week.

明天就出都有可能。

Be tomorrow.

但当我们还没有那玩意儿的时候。

But when we don't have that yet.

多么让人着迷啊。

How fascinating.

但是当我们开始拥有能力相当强的精灵，能在很多事情上干得不错时，就会意识到所有关于精灵的故事，其实真正讲的不是什么精灵，而是别的更重要的东西。

But when we start getting fairly capable genies that can execute reasonably well on a whole bunch of things, then we realize that all the genie stories are actually about something else as much more important than the genie.

我觉得有两个部分至少同样重要，甚至可能更重要。

And there are two components that I would say are at least as important, if not more important.

第一部分是神灯或者瓶子。对。

Component number one is the lamp or the bottle Mhmm.

就是控制精灵的那个东西。

The thing that controls the genie.

因为如果没有灯，精灵只是个包装更好的捣蛋鬼，你肯定不希望这种捣蛋鬼在你工作场所或生活里到处乱窜、制造混乱。

Because the genie without a lamp is just a poltergeist with better branding, and you don't want poltergeists running around causing chaos in your, workplace or in your life.

然后另一个部分就是许愿者。

And then the other piece is the wisher.

而这个许愿者，不只是故事里那个拿着灯的人。

And the wisher is not just the one person with the lamp like in the story.

许愿者其实是整个组织，它需要想清楚自己想要什么，希望世界变成什么样，希望业务怎么运作，还有那些……

The wisher is the entire organization that needs to now figure out what it wants, how it wants the world to be, how it wants business to work, what the...

流程怎么设计，以及未来是什么样子。

How the processes are gonna work, and what future looks like.

这需要很大的愿景，还需要一些技能，我怀疑大多数人根本没把这些当技能来培养。

Now that takes a lot of vision, and it also takes skills that that I'm not sure most people think about as skills to build.

比如，如果你家孩子跑过来跟你说：‘教我怎么去渴望’，你会怎么教他呢？

Like, you know, if your kid came up to you and said, teach me how to want, how would you teach them how to want?

我们有教人这些事吗？

Do we teach people these things?

在商业环境里，那什么呢……

Now in the business setting, what do...

怎么……

How...

人们是不是天生就具备这种能力，不仅知道自己想要什么，还知道自己团队想要什么，甚至知道整个组织想要什么，对吧？

Do people come with the skill not only, here's what I want, here's what my team wants, here's what my organization wants, right?

这种许愿的能力非常重要。

Those wisher skills are so important.

但投资却还是放在老精灵身上，灯神只有一点点，那什么时候我们才能更平衡地分配资源和投入呢？

And yet the investment has been old genie, tiny little bit of lamp, and then, well, when are we going to start balancing that stack out a little more?

我不是说不要对精灵着迷，但也要多想想如何控制它。

I'm not saying don't be fascinated by the genie, but also think a lot about control.

要多想想，如果会有这种充满概率的混乱，那么边界在哪里？

Think a lot about if you're going to have all this probabilistic chaos, what are the boundaries?

什么……

What...

你打算怎么检查？

How are you gonna check things?

有哪些明确的把关环节？

What are the deterministic gates?

但归根结底，这一切是为了什么？

But also, what is it all for?

由谁来决定？

Who decides?

由谁来统筹协调这一切？

And who aligns all that?

而这些信息在组织中又是如何流动的？

And how does that information travel in an organization?

所以我跟组织说，好，开始列愿望清单吧。

So then I say to organizations, okay, start forming wishlists.

开始弄清楚你想要什么。

Start figuring out what you want.

在这个过程中，就从某件事开始吧。

And as you do that, just start with something.

你会……

You're gonna...

这会迫使你去梳理那些你正在设想的流程。

It's gonna force you to map the processes that you...

你现在在许愿的那些事情。

You're now making wishes about.

你必须表达得很清楚。

You have to speak with clarity.

但随后你也会发现，你会和很多人争论，到底是谁有资格对这件事发表意见。

But then you'll also realize that you're gonna have a lot of arguments with who actually gets to make statements about this.

不可能是职位最高的人来决定整个组织所有事情的细节。

It's not gonna be the most senior person determining all the details of how things should be for the whole organization.

没有人有那个精力。

No one's got that that, bandwidth.

那么，我们实际上怎么在组织中传递这些信息，让它发挥作用呢？

So how do we how do we actually travel that information through an organization so it works?

我想谈谈你刚才提到的那个话题，我觉得那真的很有意思。

I wanna touch on something that you just mentioned, which I think is really fascinating.

听起来这几乎像是一场关于意愿、关于……的对话。

It almost sounds like it's a it's a conversation about will, about, like...

因为，你懂的，你……

Because you you you know, you...

就像，有些东西你可以向精灵许愿。

The idea of, like, that there are there are things that you can ask a genie for.

比如我想要一百万美元，或者我想要一辆新车，一栋新房子。

You know, I want a million dollars, or I want a new car, I want a new house.

但是我们不会向精灵许这个愿望，打个比方来说。

But we don't ask the genie to use the metaphor.

别向神灯许愿要“我怎么构建”这类的东西。

Don't ask the genie for things like, how do I build?

想法从哪里来？

Where do ideas come from?

我怎么才能产生最好的想法？

How do I generate the best idea?

这些东西从哪里来？

Where do those things come from?

其实，这些的根源是我们作为人类天生就有的东西，那就是创造的意志、驱动力和成功的渴望。

And really, where it comes from is it's like, it's this intrinsic thing that we all have as people, which is the will to create, the will to drive, the will to succeed.

丹尼尔，我想问你，因为你之前提到过，这是人工智能缺失的一个部分。

I want to actually ask you, Daniel, because you talked about this in your That's a missing component from AI.

你会怎么描述这种等式呢？就像，AI能做推理，但AI没有意志和良知来真正像我们这样去做事。

How do you describe this this sort of, like, equation of, like, AI AI does the the the reasoning, but AI doesn't have the will and the conscience to really do it the way that we do?

这就是我们之所以还能存在的论据吗？

Is that the argument for how we remain?

我觉得这就是问题的核心，你这么说我特别高兴。

I think this is the crux of the issue, and I'm so happy how you phrase it.

我觉得我们俩是各自独立想到这个概念的。

I think we independently got to this concept.

你管它叫“许愿者”。

You called it the wisher.

我讲的是意志和智能之间的分离。

I talk about separation between will and intelligence.

如果看看意志的本质，即使在哲学史甚至宗教史上，它们也完全把意志和智能分开了。

If we look at the nature of will, even throughout the history of philosophy and even religions, they separated completely will from intelligence.

我想很多人会说意志在智能之上，智能源自意志。

I think many people will say will is on the top of the intelligence and intelligence derives from the will.

如果你看看人类……

And if you look at the human...

这……

This...

我们物种的进化全都证实了这点。

Our species and evolution, I mean, it all confirms.

我觉得大家在目标优化上存在一个混淆。

I think there is a confusion between optimizing for a goal.

对。

Okay.

智能体可以追求一个无害的目标。

Agents can get the goal to do something that is harmless.

它们也可能产生有害的子目标。

They can come with sub objectives that can be harmful.

这倒是真的。

That's true.

但这并不意味着它们是源于

But it doesn't mean they originated

意愿。

The will.

在我的书里，我提到了一个非常有趣的观察，来自一位现代美国哲学家，法兰克福。

In my book, I point to a very interesting observation of a modern American philosopher, Frankfurt.

他提出了“二阶意志”这个概念，这个概念区分了人类和其他所有生物。

And he introduced the concept of second order of will that makes the difference between humans and any other sorts of creatures.

AI算是一种生物。

AI being a creature.

我们就这么说吧。

Let's call it this way.

他提到了一个有趣的观点，就是“想要去意愿某件事”的这种意志。

And he pointed to this interesting notion of the will to will something.

在某种意义上，我们是唯一一种能够“想要去想要”某种东西的生物。

We are the only type of creatures in a way that can want to want something.

我想要成为那种早上会去健身房的人。

I want to want to be a person that goes to the gym in the morning.

我想要成为那种思维更健康的人。

I want to want to be the type of person that thinks healthier.

这种想想要的意愿，跟单纯的目标优化差别太大了。

This want to want, such a huge distinction between, like, a goal optimization.

我认为当前的想法和智能水平，从底层是不会产生人类意义上的意志的。

I don't see will in the human sense emerging from below, the current thinking, the current intelligence.

另外，如果回到企业里工作是怎么完成的，我觉得这里头有……

And also now if we go back to how the work happens in an enterprise, I think there is...

工作是由一套微观的主动性系统构成的。

It's a system of micro initiatives that makes the work.

并不是我定了一个目标，然后所有人都围着它转。

It's not that I originated a goal and everybody is just within that.

人们都有各自的……

People have all...

人们会改变流程，改变方式。

People change the process, change the way.

他们有自己的直觉。

They have hunches.

他们……

They...

就是这种主动性。

It's this initiative.

在我看来，我在书里指出，这将是人类持久的特点之一。

And to me, I point in my book, this is gonna be one of the durable human traits.

我们需要更多地聘用有主动性的人，而不是只看重专业能力。

And we need to hire more people for initiative than for expertise.

你讲的这些我太赞同了。

So much in there to agree with.

这个结论，主动性和专业能力的对比，我完全同意你，尤其是因为一旦专业知识可以被数字化，那你就要转向那些还没被……对。

The the the conclusion, the initiative versus expertise, I'm 100% with you on, especially since once the expertise is something that you can digitise, then you need to move to, well, what haven't we Yeah.

数字化的东西。

Digitized yet?

我经常对工程师和数据科学家说，这些不是软技能。

I've always said, for example, to engineers and and data scientists, I said, you know, these are not soft skills.

这些恰恰是最难自动化的技能。

These are just the hardest skills to automate.

所以每当我们找到可以自动化的东西，就又往上一层，对吧。

And so every time we find something something we can automate, there is one more level Yeah.

那部分仍然留给我们人类。

That that that stays with us humans.

我还想说，我觉得我们此刻有机会发明第三层意愿，也就是你真正想要的愿望。

I also say, I think we have an opportunity in this moment to invent third order will, which is the wish you would want to make.

不是经典的那些，比如金钱、权力、爱情什么的。

Not not for for whether the classics money, power, love, whatever.

尤其是如果你能有无穷无尽的愿望，对吧，AI可不会让你只许三个。

Especially if you get infinite wishes, which, right, AI doesn't stop you at three.

就是这个。

That's the one.

最初的那个

The original

这就是那个经典的精灵漏洞嘛。

that's the original genie hack right there.

你说得对。

You're right.

你抓住了重点。

You got them.

他们只是花了很少的代价。

They they cast they cast a pittance in tokens.

好。

Okay.

你几乎有无穷的愿望，直到你的令牌上限。

You've got practically infinite wishes up to your token cap.

现在你想许什么愿呢？

Now what do you wish for?

第三层的愿望就是希望自己能更会许愿。

The third order will would be wish to be better at wishing.

而且这不是要把人排除在外。

And that that is not take the human out of the equation.

而是说，在你接触到的所有资源和动物，包括AI的背景下，你如何利用这一切，为你想优化的人——无论是你自己、你的家人、你的团队，还是你的世界——发挥最大的作用。

That is given the context of all the resources and animals, including AI, that you have access to, how can you now take all of this and do the most with it for whoever you are optimizing for, whether it is yourself, your family, your team, your world.

我觉得我们在那之前就卡住了几级。

And I feel like we are stuck a few levels before that.

当我们思考决策、意志、愿望这些事时，感觉还停留在上个千年，而我们应该意识到，我们正被非常复杂的技术武装到牙齿，每个人都能利用它，但很难预测那个充满复杂技术的未来会是什么样子。

When we think about decision making, when we think about will, wishing, all these things, it feels very last millennium when we should appreciate just how incredibly augmented we're all becoming by very complex technology and the way in which each of us can now use it, it it it will be very hard for us to predict what that future with all that complex stuff is gonna look like.

而我们人类需要保持领先一步，或者至少能驾驭它。

And we humans will need to stay one step ahead of or on top of it.

我不是说在那种……

And I don't mean in the...

那种我们跟不上、像跑步机一样不停转的方式。

In a kind of treadmill that we can't keep up with way.

当你面对“你可以要任何东西”，而且得到好东西的概率越来越高时，你就会开始要更多更小众、更奇怪的东西。

I mean, when we're faced with you can ask for anything and and the likelihood of getting something good back goes up, you're gonna start asking for more and more niche and weird stuff.

而这些又会成为其他人继续发展的基础。

And that's going to come into the pile of things that other people will build on top of.

你永远无法自动化掉那种人的因素。

You'll never automate out that that human element.

我对这个想法很着迷，因为它有点呼应了我们经常讨论的那个……

I'm I'm fascinated by this idea because it sort of plays into this...

我们一直在说的，丹尼尔，那就是并不是每个流程都值得配备一个智能代理。

What we talk about all the time, Daniel, which is not every not every process deserves an agent.

确定性流程。

Deterministic.

有些事情本质上就是概率性的。

There are things that are going to be inherently probabilistic.

关键，也就是人的部分，在于分辨哪些是确定性的，哪些是概率性的，然后在你构建这个系统的过程中，始终保持这种判断。

And the trick, the human piece of it, is knowing which is which, and then being able to, like, carry that through as you build this thing.

而这基本上就是同一个观点。

And it's it's essentially sort of the same idea.

就是说，当你可以拥有一切的时候，你真的需要一切吗？

It's like, when you can have everything, do you need everything?

是不是每件事都需要...

Does everything need to be...

比如说，你是不是需要把每件事都当成一个AI转型的机会？

Like, do you need to treat everything as an AI transformation opportunity?

我...

I...

这真的是一个很有意思的想法。

It's it's a really fascinating idea.

所以我想深入探讨一下关于判断力这个话题。

So I I wanna explore this idea of judgment.

这其实就是...

This is...

说到底，这真的就是关于判断力的问题。

I mean, this is really what it does come down as a question of judgment.

当你决定哪里应该放什么，以及你应该要求多少的时候，这个对话从哪里开始呢？

When you're deciding what should be where and how much you should be asking for, where does that conversation start?

谁来主导这个对话？

Who leads that conversation?

是流程负责人吗？

Is it the process owner?

是业务负责人吗？

Is it the business owner?

是首席执行官吗？

Is it the the CEO?

这个对话从哪里开始，又是如何逐渐落实到真正产生影响的层面呢？

Where does that conversation start, and how does it make its way down to a place where it actually starts to make a difference?

你是在问实际操作中是怎么做的，还是说如果我们已经实现了那个愿望，我们会怎么处理事情？

Are you asking how it gets done in practice or in a world where we might have made that wish

我们可能会怎么去做。

How we might approach things.

我可以两个都问吗？

Can I ask both?

两个都很有意思，因为一个意味着人类有能力继续学习，另一个意味着我们好像全知全能，觉得我们一旦做了决定，就永远不会出错。

It's equally fascinating for both because one implies the ability of humans to continue to learn, and one implies that somehow we're, like, we're omniscient, that we...

一旦我们做出那个决定，它就不会出错了。

Once we make that decision, it's never gonna go wrong.

我读研究生的时候，学的是决策心理学和神经科学，这个领域自称是“判断与决策”，对语言挑剔的人来说，这听起来可能像是判断和决策是两回事。

When I when I went to grad school for the psychology and neuroscience of decision making, the the field called itself judgment and decision making, which to the linguistic pedant would sound like, maybe judgment and decision making are two different things.

那为什么要用两个不同的词呢？

So so why would we use two different words?

嗯，决策是有正确答案的那部分，而判断是处理错误的部分，但可能有无穷多个正确答案。

Well, decision making is the piece that has a right answer after judgment has dealt with the wrong, but with the potentially endless right answers.

判断是选择元启发法的过程。

Where judgment is the selection of metaheuristics.

那是我们决定如何做决定的地方。

It's where we decide how to decide.

而且，做到这一点的方式并不只有一种。

And there is not just one option to do that.

一旦我们理清了想怎么处理事情，接下来就是决策环节，这时候我们的统计系统、AI系统可以帮我们一把。

Once we have sorted out how we want to approach things, then there is decision making and things for for our statistical systems, our AI systems to help us out with.

但当我们面对无限可能，真的没有唯一正确答案的时候，这时候就是我们作为人去站出来，说，好，这就是我要走的方向。而且，我觉得人们经常有点困惑，因为你可以有无数个正确答案，也可能有无数个错误答案。

But when we're faced with endless possibility, and there literally is no one right answer, that is when we step up as people and say, okay, this is the direction I'm And often, think folks get get a little bit confused because you can have endless right answers and also situations with endless wrong answers.

但过去一个世纪，我们习惯处理的情况、我们追求的目标，通常是那种只有一个正确答案、很多错误答案的。

But what we've been used to dealing with and how metrics look from last century is typically what we would go after is where there is one right answer and many wrong answers.

预测今天的温度。

Predict the temperature today.

答案只有一个。

There is one right answer.

负无穷肯定是错的。

Minus infinity is a wrong answer.

在这两者之间我们还能找到很多答案。

We can we can find many between that.

有些……

Some...

这里可能引用了《安娜·卡列尼娜》里关于幸福和不幸福指标的内容，这是给热爱文学的朋友们的。

There's some Anna Karenina reference maybe about happy and unhappy metrics for for for the literary people.

每一个

Every

智能体转型都是一样的。

agentic transformation is, you know, is is the same.

你懂吗？

You know?

它们都以各自独特的方式相同。

They're all they're all they're all the same in their own unique way.

每一个智能体转型都是不幸福的。

Every agentic transformation is unhappy.

就是

It's

全都错了。

all wrong.

没错。

Exactly.

正是如此。

Exactly.

有了生成式AI，领导者现在面临的挑战是，不再只有一个正确答案，我们得到答案了吗？

With generative AI, what leaders now have to struggle with is now instead of one right answer, did we get it?

我们没得到答案吗？

Did we not get it?

现在答案是无穷无尽的。

Now it's endless right answers.

现在必须有人来做判断，比如，界限划在哪里，或者怎么为这个制定衡量标准。

Some person has to now be making judgment calls about, well, where do we draw the lines, or how do we make metrics around this?

举个具体的例子，比如，帮我写一封邮件，然后自动发送，不需要我审核。

And to give a concrete example, right, write an email for me and agentically send it without me checking.

哦，不。

Oh, no.

什么叫好的邮件？

What is a good email?

那是什么意思？

What does that mean?

有很多……

There are many...

正确答案有很多种。

There are endless right answers.

所以有人得做出一些判断。

And so someone's gonna have to make some judgment calls.

回到我们最开始的问题……

And our first problem getting back to...

你看，我没忘记你问我的事。

See, I haven't forgotten what you've asked me.

我们把问题拉回到核心上，首先就是要认识到，人们往往不把判断或者决策当成一种技能。

Our first problem getting back to the core of how this is done is that people don't tend to think of either judgment or decision making as skills.

决策好像就是自然而然发生的事，或者就是开个会就完了。

Decisions are just things that happen or decisions are meetings that we do.

在我看来，决策应该变成一种技术，一种可以重复使用的技术，就像其他任何可复用的技术一样。

In my view of all this, that decision making should become a technology, a reusable technology, like any other reusable technology.

毕竟，我们的决策和行动，就是我们影响周围世界的方式。

After all, our decisions, our actions are how we impact the world around us.

这才是关键。

That's that's the important bit.

所以我们需要把它变成一种可复用的技术，而人类判断始终是这种技术的输入。

So we need to make this a reusable technology, and that reusable technology will always have human judgment as its input.

问题在于，我们用的是谁的判断。

The problem is whose human judgment we have.

作为组织，我们必须开始认真对待这个问题。

We need to, as organizations, start getting serious about asking that question.

谁来应用这个判断？

Who gets to apply the judgment?

在我看来，这种判断和决策过程恰恰暴露了AI相对于人类的另一个弱点。

In my opinion, that entire judgment and decision making points to another weakness of AI versus humans.

AI不承担自己决策带来的后果。

AI doesn't bear the consequences of their decisions.

你不可能把后果和决策过程分开。

You cannot separate consequences by decision making process.

我认为在推理方面……

I think in reasoning...

所以，如果我把判断和决策简单拆解一下，你先有数据输入，然后有逻辑推理。

So if if I'm dissecting simplistically judgments and decisions, you have you have data input, you will have a logical reasoning.

你得得出结论，然后评估后果。

You reach a conclusion, and then you evaluate the consequences.

现在在推理方面，AI能比大多数人类做得更好。

AI can do better job than most humans on reasoning right now.

我觉得我们应该明白这一点，而且我们往往想当然地认为，但人工智能确实无法评估后果。

I think we should conceive this and we take it for granted, but AI cannot evaluate the consequences.

你看。

And look.

刚才你提到托尔斯泰和《安娜·卡列尼娜》的时候，我想起另一个关于冷战时期的真实有趣故事。

While you were talking about what Tolstoy and Anna Karenina, I I remember another interesting real story about what happened during Cold War.

那是……

So it was...

这是个真实的故事。

It's a true story.

有一艘俄罗斯潜艇收到了某种信息，让他们……

It was a Russian submarine that got some input that they...

基本上，系统认为遭到了核攻击。

Basically, the systems believe there is a nuclear attack.

潜艇失联了，因为这个……

The submarine was disconnected because this...

整个系统与莫斯科的中央指挥断了联系。

The entire thing was disconnected from central command in Moscow.

所以他们不得不自己决定是否反击。

So they have to make themselves a decision to counter attack.

最后是三个人来决定这件事。

So it was three people to make the decision.

其中两个人，相信我，决定对美国发动核攻击。

Two, believe me, made the decision to launch a nuclear attack on US.

是第三个人阻止了这件事，因为他们需要三票。

It was the third guy that stop it because they need three votes

对。

Yep.

才能做决定。

To make the decision.

说真的，就是这样避免了核战争。

Seriously, this is how we prevented the nuclear war.

所以我觉得这一点很能说明问题，关于我们是否应该让AI来做决策。

So I think this is very telling in where we keep AI to make decisions.

对我来说，这就是底线。

To me, this is the limit.

这是后果的问题。

It's the consequence.

如果后果不大，而且可以逆转，让AI做决定、采取行动，那是可以的。

If the consequence is small and you can reverse it, let AI make decisions, take actions, it's fine.

当然，得把它限制在可控范围内，避免出偏差。

Put put it in the bottle, of course, so they don't go astray.

但那没问题。

But that's fine.

但只要决策对组织有重大影响，你就必须让人参与。

But as long as decisions are consequential for an organization, you have to put people.

你需要一个能说‘不行’的人。

You need someone that you go...

你得负责任。

You are accountable.

而且，如果你……这种判断的应用确实会变得很别扭。

And the the application of judgment does become quite awkward if you...

比如，如果我可以对你的生活指手画脚，做各种判断，但我自己却不用承担任何后果。

If if I get to make all kinds of judgment calls, let's say, about how your life has to work, and then I don't face any consequences of of anything that I've done.

但就算从那种学术化的判断和决策理论角度看，选一个元启发式规则，说“可逆、风险低，就自动化，别操心”，这本身就是一个判断。

But even from the the judgment and decision making nerdy academic framing, to to pick the metaheuristic that says reversible, low stakes, automate it and don't worry about it.

这本身就是一个判断。

That that right there is a judgment call.

而且这个判断是在说：我们的时间比那点小事更值钱。

And that is a judgment call where I'm saying our time is worth more than whatever the small thing is.

设定这个门槛，怎么才算“不值得花人的时间和注意力”？这也是一个判断。

Picking the threshold, what does it mean for something to be insufficiently interesting for, the investment of human time and attention?

这也是一个判断。

That is also a judgment call.

但在我看来，真正稀缺的东西，以及为什么我们必须始终把人放在第一位，是因为人是我们最宝贵的资源。

But as I as I look to what's limited and why we we have to keep thinking about the human piece first is our most precious resource.

不是时间。

It's not time.

而是高质量的、人的注意力。

It is high quality human attention.

在正确的时间，让对的人把注意力放在问题上。

The right person's attention on the problem in the right moment.

就拿潜艇那个情况来说，第三个人的投票，晚了两个小时，那也没什么用。

The the third vote, for the submarine situation, two hours too late, wouldn't have been very good.

随便一个农民，给他看一些数字，能怎样呢？

Some random farmer being shown some numbers.

对。

Yes.

不对。

No.

而且，那可能是一件非常糟糕的事情，如果当时有干扰、注意力不集中、无法运用那种判断力的话，也会是非常糟糕的。

Also, probably would have been a very bad thing, and it would have been a very bad thing if there was distraction, lack of attention, lack of ability to apply that judgment call.

所以，无论我们的系统变得多复杂，我们都需要从人类那里分配这种非常非常宝贵的东西。

So no matter how complex our systems become, we will be asked to allocate this very, very precious thing from humans.

无论情况需要什么样的专家，就是注意力。

Expert for whatever the situation calls for, attention.

我喜欢这个想法，因为你们两位都为丹尼尔你在书里描述的东西——所谓的“轨道”——提出了非常有力的论证。轨道就是那些你知道要用来经营业务、运行流程的不可妥协的准则，它们规范整个系统，确保它不会偏离轨道，不管是由于监管问题、盈亏问题还是其他什么原因。

I like this idea because you've you've you've both made a very compelling case for what you described in your book, Daniel, is the rails, which is what are the things that you know you wanna run the business on, you wanna run the process on that are nonnegotiable, that govern the entire system, that make sure that it doesn't go off the trail, whether it's because of regulatory issues, it's because of p and l issues, whatever it is.

这些就是你们定义的：我们不能也不会违反这些东西。

These are the things that you've defined as we we cannot and will not violate these.

这些是源于判断力的。

Those are things that are born out of judgment.

这些是由领导者做出的决策，无论是盈亏负责人还是业务负责人。

Those are decisions that are made by a leader, whether it's a P and L leader or a leader of the business.

而且监管者也在运用判断力。

And the regulator is applying judgment too.