Should we create superintelligent AI?

翻译人员: Manlin Fang 校对人员: Bruce Wang

Maybe I'm not the best library assistant.

也许我不是最好的图书馆助理。

Could artificial intelligence do it better?

那么人工智能能做得更好吗？

I’d like to think of myself as irreplaceable, but let’s see.

我想把自己看作是不可替代的， 但让我们来看看，

Maybe getting fired was the push I needed to find the perfect job for me.

也许被解雇是我找到最适合 自己的工作所需要的推动力，

As long as I find it quickly.

只要我能尽快找到它，

And don’t spend any money in the meantime.

而且在此期间不要花任何钱。

Luckily, there's somewhere I can sit and work all day without buying anything.

幸运的是，有个地方我可以 整天坐下来工作，不用买任何东西。

Ada? I thought I was clear—

艾达？我以为我说得很清楚——

your employment here is over.

你在这里的工作结束了。

Oh, I'm not here as an employee. This is a public library, after all.

哦，我不是作为员工来这里的， 毕竟这是一个公共图书馆。

Right. How to make one month as a library assistant

对，如何让图书馆助理的一个月

sound like the 3 to 5 years of experience

听起来像每份入门级工作所需的3到5年的经验。

every entry level job seems to require?

Shh!

嘘！

Dealt with an overbearing, rule-obsessed boss...

对付了一个霸道、痴迷于规则的老板

No, no. Ugh.

不，不，呃……

Let’s see how an AI does with writing my resume.

让我们看看人工智能如何写我的简历。

Okay. I checked out lots of books.

好吧，我看了很多书，

Make that sound impressive.

让这听起来令人印象深刻。

Ooh! “Expedited seamless intellectual asset transactions.”

哦！“加快无缝知识资产交易。”

Excellent. Okay, how about “sat at reception”?

太棒了！好吧， “坐在接待处”怎么样？

Oh! “Frontline client liaison.”

哦！“一线客户联络”

Good, good. That’s great!

很好，很好，太棒了！

Maybe too good.

也许有些过于棒了……

Even if I get one of these jobs,

即使我找到了其中一份工作，如果我缴纳费用 只是为了让AI 超越我，该怎么办？

what if I pay my dues only to have AI outpace me?

What if, by the time I hone my skills,

如果在我磨练技能的时候，

humans aren’t doing writing and research anymore?

人类已经不在写作和研究了，又该怎么办？

Maybe I should work directly on AI.

也许我应该直接研究人工智能。

Then at least I’d be helping determine the course of the future

那么至少我会帮助确定未来的方向，

instead of being left behind.

而不是被抛在后面。

Like, say I created an AI librarian.

比如，假设我创建了一个 AI 图书馆员。

That actually doesn’t seem too far-fetched.

这实际上似乎并不牵强。

To make my system more powerful I have to, well...

为了让我的系统更强大，我必须，好吧...

keep adding more simulated neurons for starters.

继续为初学者添加更多模拟神经元。

Soon my AI system is a better librarian than my old boss.

很快，我的人工智能系统成为了 比我的前老板更好的图书馆管理员。

But you know what? I'm feeling generous.

但是你知道吗？我感觉很慷慨。

Why don't you stay as my assistant?

你为什么不继续当我的助手？

You can help me test your replacement.

你可以帮我测试你的替代品。

Hmph.

嗯……

What should I name it?

我应该给它起什么名字？

What's a good name for a great librarian?

对于一个优秀的图书馆员来说， 一个好名字是什么？

No ideas? I like Biblionimus Maximus.

没有想法吗？ 我喜欢 Biblionimus Maximus。

Hmm. Let's try a prompt.

嗯。让我们试试提示一下。

Who invented artificial intelligence?

谁发明了人工智能？

Include citations.

包括引文。

Biblionimus says the first theoretical descriptions of AI

Biblionimus说， 对人工智能的第一批理论描述

were “Comments on a Thinking Machine” by Ada Lovelace in 1840,

是艾达·洛夫莱斯 在1840年发表的《对思维机器的评论》，

and “Machine Intelligence” by Alan Turing in 1927.

以及艾伦·图灵在1927年的《机器智能》。

See? Nailed it!

看见了吗？搞定了！

Not quite.

不完全是。

Lovelace’s “Notes on an Analytical Engine” were published in 1843.

洛夫莱斯的《分析引擎笔记》于 1843年出版。

Alan Turing’s paper was called “Computing Machinery and Intelligence,”

艾伦·图灵的论文名为 《计算机机械与情报》，

and it was published in 1950.

于1950年发表。

Isn’t that what Biblionimus said?

Biblionimus 不就是这么说的吗？

No. I can see how it sounds that way to you,

不。我能看出对你来说听起来像这样，

but every element was slightly wrong.

但是每个元素都略有错误。

But what about the citations?

但是引文呢？

Completely fabricated.

完全制造。

Look, it even made up ISBN numbers.

看，它甚至编造了 ISBN 号码。

Clearly, it's not ready to replace human librarians if it needs an expert

显然，如果它需要专家

to check its work.

来检查其工作， 它还没有准备好取代人类图书馆员。

We’re still training Biblionimus.

我们还在训练 Biblionimus。

It’ll improve.

情况会有所改善。

Hmph.

嗯。

In time, my system stops producing bogus results.

随着时间的推移， 我的系统停止产生虚假结果。

As it gets more competent,

随着能力的提高，

it doesn’t just help patrons find what they’re looking for,

它不仅可以帮助顾客找到他们想要的东西，

it finds resources they didn’t even know they needed.

还可以找到他们甚至不知道自己需要的资源。

The library has never been busier.

图书馆从未像现在这样繁忙过。

Maybe it’s a waste to limit our technology to the library.

将我们的技术局限在图书馆可能是一种浪费。

Research labs, companies, even governments—

研究实验室、公司，甚至政府——

they all have more information than they can manage.

他们拥有的信息都超出了他们的管理能力。

They could use a good librarian, don’t you think?

他们可以聘请一个 优秀的图书馆员，你不觉得吗？

Hmph.

嗯。

What?

什么？

Selling the AI off to the highest bidder

将人工智能出售给出价最高的人

isn’t exactly in the spirit of the library.

并不完全符合图书馆的精神。

True...

真的...

And I definitely don’t want Biblionimus to become an instrument

而且我绝对不希望Biblionimus

of corporate greed or abuses of power or anything like that.

成为企业贪婪或滥用权力之类的工具。

Let’s make Biblionimus available for free,

让我们免费提供 Biblionimus，

but anyone who wants to use it has to be approved by us first.

但任何想要使用它的人都必须先获得我们的批准。

We’ll never be able to review all these applications.

我们永远无法审查所有这些申请。

Even if you actually help.

即使你真的在帮忙。

I have a better idea.

我有个更好的主意。

I bought the Diligentsia 3000.

我买了 Diligentsia 3000。

It’s an AI system that can vet applicants for us.

这是一个可以为我们审查申请人的人工智能系统。

It will process our backlog almost instantly.

它几乎可以立即处理我们的待办事项。

So, I guess I won’t be needing your help anymore.

所以，我想我不再需要你的帮助了。

I mean I’m firing you.

我的意思是我要解雇你。

The library is a public space.

图书馆是一个公共空间。

Almost every vital industry has a need for my AI.

几乎每个重要行业都需要我的人工智能。

As do plenty of not-so-vital industries.

许多不太重要的行业也是如此。

(Radio chatter) Suspect is in custody.

（电台闲聊）嫌疑人已被拘留。

What's going on?

发生了什么？

What? Investigating a bomb threat?

什么？正在调查炸弹威胁？

Where did this tip come from?

这个提示来自哪里？

It came from Biblionimus?

它来自 Biblionimus？

I thought you made sure Biblionimus wasn’t making mistakes anymore!

我还以为你确保 Biblionimus不会再犯错误了！

That was back before you fired me.

那是在你解雇我之前。

I don't work here anymore.

我不再在这里工作了。

You invented Biblionimus, can’t you figure out how to fix it?

你发明了 Biblionimus， 你想不出如何修复它吗？

Argh! It doesn’t work like that!

啊！它不是那样工作的！

I created Biblionimus, but it’s been out there ever since,

我创建了 Biblionimus， 但它从那以后一直存在，它

absorbing more information than any human ever could.

吸收的信息比任何人类都多。

It's grown up and left home, so to speak.

可以这么说，它长大了，离开了家。

But there is one thing I can do.

但我能做一件事。

I can bring in a digital neuroscientist.

我可以请一位数字神经科学家。

A what?

一位什么？

It’s an AI system that studies other AI systems’, um, brains.

它是一个研究其他人工智能系统，嗯，大脑的人工智能系统。

Okay, Dr. Cerebrox.

好吧，塞雷布罗克斯博士。

Biblionimus is supposed to make unexpected, hopeful connections,

Biblionimus 本应 建立意想不到的、充满希望的联系，

not spout conspiracy theories.

而不是大肆宣扬阴谋论。

As far as I can tell,

据我所知， Biblionimus Maximus一点问题都没有。

there’s nothing at all wrong with Biblionimus Maximus.

Did the human users feed it incorrect information?

人类用户提供的信息是否不正确？

Or perhaps ask it a it a confusing question?

或者也许问一个令人困惑的问题？

Even if they did, it shouldn’t be making up bomb threats.

即使他们这样做了，也不应该编造炸弹威胁。

What if it did something worse, like encourage a coup? Or a nuclear strike?

如果它做了更糟的事情， 比如鼓励政变怎么办？还是核打击？

Let me take another look.

让我再看一看。

This system employs a deeply intricate architecture.

该系统采用了错综复杂的架构。

Its robust toolkit consists of refined algorithms

其强大的工具包由

performing exhaustive data analysis and advanced computational procedures.

执行详尽数据分析的精细算法和高级计算程序组成。

But what does that mean?

但这是什么意思呢？

What's actually happening inside it?

里面到底发生了什么？

My training is to express diagnoses in terms you can understand.

我的训练是用你能理解的术语来表达诊断。

You are not a digital neuroscientist.

你不是数字神经科学家。

Therefore, there will be a level of abstraction.

因此，将有一定程度的抽象。

We need to speak to your manager.

我们需要和你的经理谈谈。

Certainly, just a moment—

当然，请稍等片刻——晚上

Good evening, this is the chief customer satisfaction officer of Cerebrox Inc.

好，这是Cerebrox Inc.的首席客户满意度官。

Uh, you have the exact same voice as Dr. Cerebrox.

呃，你的声音和Cerebrox博士完全一样。

All of our AI’s use the voice that has been proven to be optimally soothing

事实证明，我们所有的人工智能 都使用最能舒缓

to customers.

客户情绪的语音。

But rest assured, I am a different, more senior system.

但请放心，我是一个与众不同、更高级的系统。

So there’s no human I can speak to?

所以没有人能和我说话？

How would a human be able to help you interpret matters of digital neuroscience?

人类如何能够帮助你解释数字神经科学的问题？

We have to turn Biblionimus off until we can make sure it won’t wreak havoc.

我们必须关闭 Biblionimus， 直到我们能确保它不会造成严重破坏。

Are you sure?

你确定吗？

I don’t see what else we can do.

我看不出我们还能做什么。

Whew. I need a drink.

呼！我需要喝一杯。

That’s weird.

这很奇怪。

Error... Reconnect payment method... okay...

错误...重新连接付款方式...好吧..

Another error... I can’t even log into my account.

另一个错误...我甚至无法登录我的账户。

There must be something wrong with the app.

该应用程序一定有问题。

Hi, I’m having trouble accessing my account.

你好，我在访问我的账户时遇到问题。

Hello?

你好？

Is anyone here?

有人在这里吗？

I need to speak to a person!

我需要和一个人说话！

(Bank bot) AIs are much more reliable bankers than humans.

（银行机器人） 人工智能是比人类更可靠的银行家。

AIs are more reliable bankers. AIs are more reliable bankers.

人工智能是更可靠的银行家， 人工智能是更可靠的银行家。

What is happening?

发生了什么？

Didn’t you say Biblionimus was deployed in many critical industries?

你不是说 Biblionimus 已部署在许多关键行业吗？

Yes, but banks and utility companies should still be able to function

是的，但是银行和公用事业公司仍然应该能够

like they did before Biblionimus!

像Biblionimus之前一样运作！

That’s easier said than done.

说起来容易做起来难。

They may not even have the expertise to do things the old way anymore.

他们甚至可能不再具备 以旧方式做事的专业知识。

So our options are either certain disaster without the AI,

因此，我们的选择要么是 在没有人工智能的情况下肯定会发生灾难，

or turning it back on and holding our breath,

要么将其重新开启并让我们屏住呼吸，

hoping nothing goes off the rails until, what, forever?

希望直到永远，什么都不会脱轨？

But would a world where humans hand the reins to AI systems

但是，人类将控制权交给人工智能系统的世界

really be that different?

真的有那么不同吗？

There are already so many things I don’t understand and can’t control,

我已经有太多不明白并且

where I have to place my trust in other people’s hands—

无法控制的事情了，我必须信任别人——

like trusting that I can use the money in my bank account,

比如相信我能用银行账户里的钱，

or that the antibiotics I get prescribed will actually treat an infection.

或者开给我的抗生素能真正治疗感染。

But I guess if we hand all that power over to AIs,

但我想，如果我们 将所有的权力移交给人工智能，

we need to make sure they will actually serve us better

我们需要确保他们能

than our imperfect human institutions.

比我们不完善的人类机构 更好地为我们服务。

And that seems really, really hard.

这似乎真的很难。

Hmph.

嗯。

You’re looking for “Brave New World”?

你在找 “Brave New World” 吗？

Hold on.

等等。

It’s in fiction, under “H.”

这是虚构的，在 “H” 下

Just down there, third row on the left.

就在下面，左边的第三排。

Ada. Maybe I could reconsider your position.

艾达，也许我可以重新考虑你的立场。

You’re offering me my job back?

你要把我的工作还给我？

If you can follow all the rules.

如果你能遵守所有规则。

Oh, that's a big decision.

哦，这是一个重大决定。

I'll have to think about it.

我得考虑一下。