未登录,请登录后再发表信息
最新评论 (0)
播放视频

人工智能会反过来控制人类吗?

Can we build AI without losing control over it? | Sam Harris

我想谈谈一种直觉上的失误
I’m going to talk about a failure of intuition
很多人都经历过
that many of us suffer from.
它会让人无法察觉特定危险的存在
It’s really a failure to detect a certain kind of danger.
我要向大家描述一个情景
I’m going to describe a scenario
一个我觉得既令人害怕
that I think is both terrifying
却又很可能发生的情景
and likely to occur,
这样两种情况的结合
and that’s not a good combination,
显然不是一个好的征兆
as it turns out.
不过 在座大多数都会觉得
And yet rather than be scared, most of you will feel
我要谈论的东西其实挺酷的
that what I’m talking about is kind of cool.
我将要说明人工智能
I’m going to describe how the gains we make
给我们带来的好处
in artificial intelligence
将怎样彻底地毁灭我们
could ultimately destroy us.
实际上 不想看到人工智能最终不摧毁我们
And in fact, I think it’s very difficult to see how they won’t destroy us
或者驱使我们自我毁灭 是很难的
or inspire us to destroy ourselves.
如果你和我一样
And yet if you’re anything like me,
你会发现思考这些问题相当有趣
you’ll find that it’s fun to think about these things.
而这种反应就是问题的一部分
And that response is part of the problem.
因为这种想法应该使你感到担忧
OK? That response should worry you.
假如我想在这个演讲中让你们相信
And if I were to convince you in this talk
我们因为气候变化或者其它灾难
that we were likely to suffer a global famine,
很可能会遭受全球性的饥荒
either because of climate change or some other catastrophe,
同时 你们的子孙后辈
and that your grandchildren, or their grandchildren,
都可能在这样的饥荒中挣扎求生
are very likely to live like this,
你们就不会觉得
you wouldn’t think,
“真有趣
“Interesting.
我喜欢这个TED演讲 ”
I like this TED Talk.”
饥荒一点都不有趣
Famine isn’t fun.
不过 科幻小说中的死亡 往往引人入胜
Death by science fiction, on the other hand, is fun,
现在我人工智能的发展最担心的是
and one of the things that worries me most about the development of AI at this point
人们对其将带来的危险似乎并未
is that we seem unable to marshal an appropriate emotional response
形成一个正确的认识
to the dangers that lie ahead.
我也同样如此 所以我想和大家一起探讨
I am unable to marshal this response, and I’m giving this talk.
我们就像站在了两扇门前
It’s as though we stand before two doors.
在第一扇门后面
Behind door number one,
由于某些原因 我们停下打造智能机器的脚步
we stop making progress in building intelligent machines.
也停止了 对电脑软硬件的升级
Our computer hardware and software just stops getting better for some reason.
现在让我们想一下为什么会这样
Now take a moment to consider why this might happen.
我是说 当我们认识到 智能和自动化不可估量的价值时
I mean, given how valuable intelligence and automation are,
我们总会竭尽所能地改善这些科技
we will continue to improve our technology if we are at all able to.
那么 什么会使我们停下脚步呢?
What could stop us from doing this?
一场大规模的核战争?
A full-scale nuclear war?
一次全球性的瘟疫?
A global pandemic?
一个小行星撞击了地球?
An asteroid impact?
或是贾斯汀·比伯成了美国总统?
Justin Bieber becoming president of the United States?
[笑声]
[Laughter]
重点是 总有一个事物会摧毁人类现有文明
The point is, something would have to destroy civilization as we know it.
你要想想这个灾难究竟有多恐怖
You have to imagine how bad it would have to be
才会永久性地阻止我们
to prevent us from making improvements in our technology
发展科技
permanently,
永久性地
generation after generation.
光想想它 就觉得这将是人类史上
Almost by definition, this is the worst thing
能发生的最坏的事了
that’s ever happened in human history.
那么 我们仅剩的选择
So the only alternative,
就藏在第二扇门的后面
and this is what lies behind door number two,
那就是我们持续改进智能机器
is that we continue to improve our intelligent machines
永不停歇
year after year after year.
在将来某一天 我们会造出比我们更聪明的机器
At a certain point, we will build machines that are smarter than we are,
一旦我们有了 比我们更聪明的机器
and once we have machines that are smarter than we are,
它们将进行自我改进
they will begin to improve themselves.
然后我们就得冒着数学家IJ Good
And then we risk what the mathematician IJ Good called
所说的“智能爆炸”的风险
an “intelligence explosion,”
科技发展进程将不再受我们控制
that the process could get away from us.
现在我们时常看到像这样一些讽刺漫画
Now, this is often caricatured, as I have here,
我们总会担心受到邪恶的
as a fear that armies of malicious robots
机器人军队的攻击
will attack us.
但这不是最可能出现的场景
But that isn’t the most likely scenario.
我们的机器不会自动变得邪恶
It’s not that our machines will become spontaneously malevolent.
真正的顾虑是 我们将会打造
The concern is really that we will build machines
比人类自己更有竞争力的机器
that are so much more competent than we are
而一旦我们和它们的目标有分毫不一致
that the slightest divergence between their goals and our own
我们将会被摧毁
could destroy us.
想想我们与蚂蚁的关系吧
Just think about how we relate to ants.
我们不讨厌它们
We don’t hate them.
我们不会主动伤害它们
We don’t go out of our way to harm them.
实际上 我们经常尽量避免伤害蚂蚁
In fact, sometimes we take pains not to harm them.
选择尽量绕开它们
We step over them on the sidewalk.
但只要它们存在
But whenever their presence
妨碍我们达成目标
seriously conflicts with one of our goals,
比如 当我们建造这样一个建筑时
let’s say when constructing a building like this one,
我们会毫不手软地杀掉它们
we annihilate them without a qualm.
令人担忧的是 终有一天 我们打造的机器
The concern is that we will one day build machines
不管它们是否有意 它们终将会以
that, whether they’re conscious or not,
我们对待蚂蚁的方式 来对待我们
could treat us with similar disregard.
我想很多人会说这很遥远
Now, I suspect this seems far-fetched to many of you.
我打赌你们中有人还会怀疑超级人工智能实现的可能
I bet there are those of you who doubt that superintelligent AI is possible,
认为我小题大做
much less inevitable.
但是你很快会发现 这些假设中的某一个将会是有问题的
But then you must find something wrong with one of the following assumptions.
下面是仅有的三种假设:
And there are only three of them.
第一 智慧可以被看做物理系统中的信息处理过程
Intelligence is a matter of information processing in physical systems.
事实上 这不仅仅是一个假设
Actually, this is a little bit more than an assumption.
我们已经在一些机器中嵌入了智能系统
We have already built narrow intelligence into our machines,
这些机器中很多已经
and many of these machines perform
有着超人的智能
at a level of superhuman intelligence already.
我们也知道 任何一个细节
And we know that mere matter
都可能引发所谓的“通用智能”
can give rise to what is called “general intelligence,”
这是一种能在不同领域间灵活思考的能力
an ability to think flexibly across multiple domains,
我们的大脑已经成功做到了这些 对吧?
because our brains have managed it. Right?
我的意思是 大脑里其实都是原子
I mean, there’s just atoms in here,
只要我们继续建造这些原子体系
and as long as we continue to build systems of atoms
我们就能实现越来越多的智慧行为
that display more and more intelligent behavior,
我们最终将能 当然除非遇到干扰
we will eventually, unless we are interrupted,
我们最终将能给我们的机器赋予
we will eventually build general intelligence
广泛意义上的智能
into our machines.
要知道这个进程的速度并不重要
It’s crucial to realize that the rate of progress doesn’t matter,
因为任何进程都足以让我们走进死胡同
because any progress is enough to get us into the end zone.
甚至无需考虑摩尔定律 也无需指数函数
We don’t need Moore’s law to continue. We don’t need exponential progress.
这一切就能顺其自然发生
We just need to keep going.
第二个假设 我们会一直创新
The second assumption is that we will keep going.
去继续改进我们的智能机器
We will continue to improve our intelligent machines.
由于智慧的价值就是——
And given the value of intelligence —
是我们所有珍爱的事物的来源
I mean, intelligence is either the source of everything we value
或是用于保护我们所珍视的一切
or we need it to safeguard everything we value.
智慧就是我们最有价值的资源
It is our most valuable resource.
所以我们想继续革新它
So we want to do this.
因为我们有很多迫切需要解决的问题
We have problems that we desperately need to solve.
我们想要治愈像阿兹海默症和癌症这类疾病
We want to cure diseases like Alzheimer’s and cancer.
我们想要了解经济系统 想要改善气候科学
We want to understand economic systems. We want to improve our climate science.
只要可能 我们就会将革新继续下去
So we will do this, if we can.
而且革新的列车早已驶出 车上却没有刹车
The train is already out of the station, and there’s no brake to pull.
第三种假设 人类没有登上智慧的巅峰
Finally, we don’t stand on a peak of intelligence,
甚至连接近可能都谈不上
or anywhere near it, likely.
这个想法十分关键
And this really is the crucial insight.
这就是为何我们所处的环境很危险
This is what makes our situation so precarious,
这也是为何我们对风险的直觉是不可靠的
and this is what makes our intuitions about risk so unreliable.
现在 请大家想一下 谁是世上最聪明的人
Now, just consider the smartest person who has ever lived.
几乎每个人的候选名单里都会 有约翰·冯·诺伊曼
On almost everyone’s shortlist here is John von Neumann.
冯·诺伊曼留给周围人的印象
I mean, the impression that von Neumann made on the people around him,
是他是那个时代最杰出的数学家和物理学家
and this included the greatest mathematicians and physicists of his time,
这都是有据可查的
is fairly well-documented.
即使他的故事里有一半是假的
If only half the stories about him are half true,
也没有人会质疑
there’s no question
他仍然是世界上最聪明的人之一
he’s one of the smartest people who has ever lived.
那么 让我们来看看智慧谱吧
So consider the spectrum of intelligence.
现在我们有了约翰·冯·诺伊曼
Here we have John von Neumann.
有我们大家
And then we have you and me.
另外还有一只鸡
And then we have a chicken.
[笑声]
[Laughter]
抱歉 母鸡的位置应该在这
Sorry, a chicken.
[笑声]
[Laughter]
话题已经够严肃了 开个玩笑轻松一下
There’s no reason for me to make this talk more depressing than it needs to be.
[笑声]
[Laughter]
然而 很可能的情况是 智慧谱上的内容
It seems overwhelmingly likely, however, that the spectrum of intelligence
已远远超出了我们的认知
extends much further than we currently conceive,
如果我们建造了比自身更聪明的机器
and if we build machines that are more intelligent than we are,
它们将非常可能以超乎想象的方式
they will very likely explore this spectrum
延展这个谱线
in ways that we can’t imagine,
最终超越人类
and exceed us in ways that we can’t imagine.
仅从速度方面考虑 我们就能够认识这一点
And it’s important to recognize that this is true by virtue of speed alone.
对么?现在让我们想象一下 我们刚建好一个超级人工智能
Right? So imagine if we just built a superintelligent AI
大概和斯坦福 或麻省理工学院研究员的
that was no smarter than your average team of researchers
平均水平差不多吧
at Stanford or MIT.
但是 电路板要比生物系统
Well, electronic circuits function about a million times faster
运行速度快一百万倍
than biochemical ones,
所以 这个机器思考起来会比它的创造者
so this machine should think about a million times faster
快一百万倍
than the minds that built it.
当你让它运行一周后
So you set it running for a week,
它将能给出相当于人类智慧20000年发展出的水平
and it will perform 20,000 years of human-level intellectual work,
而这个过程将周而复始
week after week after week.
那么 我们又怎么能理解 更不用说去制约
How could we even understand, much less constrain,
一个以如此速度运行的机器呢?
a mind making this sort of progress?
坦白讲 另一件令人担心的事就是
The other thing that’s worrying, frankly,
我们考虑一下最理想的情景
is that, imagine the best case scenario.
想象我们做出了 一个无任何安全隐患的
So imagine we hit upon a design of superintelligent AI
超级人工智能
that has no safety concerns.
我们有了一个前所未有的完美设计
We have the perfect design the first time around.
就好像我们被赐予了一件神物
It’s as though we’ve been handed an oracle
它能够准确的执行目标动作
that behaves exactly as intended.
这个机器将完美的节省人力工作
Well, this machine would be the perfect labor-saving device.
它设计出的机器 能够再生产其它机器
It can design the machine that can build the machine
去完成所有的人力工作
that can do any physical work,
由太阳能供电
powered by sunlight,
而成本的多少仅取决于原材料
more or less for the cost of raw materials.
那么 我们正在谈论的 就是人力劳动的终结
So we’re talking about the end of human drudgery.
也关乎脑力劳动的终结
We’re also talking about the end of most intellectual work.
那么 像我们这种”大猩猩”还能有什么用呢?
So what would apes like ourselves do in this circumstance?
我们可以悠闲地玩飞盘 给彼此做按摩
Well, we’d be free to play Frisbee and give each other massages.
服用一些迷药 穿一些奇装异服
Add some LSD and some questionable wardrobe choices,
整个世界都沉浸在狂欢中
and the whole world could be like Burning Man.
[笑声]
[Laughter]
那听起来可能很棒
Now, that might sound pretty good,
不过让我们扪心自问
but ask yourself what would happen
在现有的经济政治体制下 这意味着什么?
under our current economic and political order?
我们很可能会目睹
It seems likely that we would witness
前所未有的贫富差距
a level of wealth inequality and unemployment
和失业率
that we have never seen before.
钱人不愿意马上把这笔新的财富
Absent a willingness to immediately put this new wealth
贡献出来服务社会
to the service of all humanity,
这时一些千万富翁能够优雅地登上商业杂志封面
a few trillionaires could grace the covers of our business magazines
而剩下的人可能都在挨饿
while the rest of the world would be free to starve.
如果听说硅谷里的公司
And what would the Russians or the Chinese do
即将造出超级人工智能
if they heard that some company in Silicon Valley
俄国人和中国人 会采取怎样的行动呢?
was about to deploy a superintelligent AI?
那个机器将能以一种前所未有的能力
This machine would be capable of waging war,
去开展由领土问题和
whether terrestrial or cyber,
网络问题引发的战争
with unprecedented power.
这是一个胜者为王的世界
This is a winner-take-all scenario.
机器世界中的半年
To be six months ahead of the competition here
至少相当于现实世界
is to be 500,000 years ahead,
50万年
at a minimum.
所以仅仅是这种科技突破的传闻
So it seems that even mere rumors of this kind of breakthrough
就足以让我们的种族丧失理智
could cause our species to go berserk.
在我的观念里
Now, one of the most frightening things,
当前最可怕的东西
in my view, at this moment,
正是人工智能的研究人员
are the kinds of things that AI researchers say
安慰我们的那些话
when they want to be reassuring.
最常见的理由就是关于时间
And the most common reason we’re told not to worry is time.
他们会说 现在担心还为时尚早
This is all a long way off, don’t you know.
这很可能是50年或100年之后的事
This is probably 50 or 100 years away.
一个研究员曾说过
One researcher has said,
“担心人工智能的安全性
“Worrying about AI safety
就好比担心火星上人口过多一样”
is like worrying about overpopulation on Mars.”
这就是硅谷版本的
This is the Silicon Valley version
“不要杞人忧天 ”
of “don’t worry your pretty little head about it.”
[笑声]
[Laughter]
似乎没人注意
No one seems to notice
以时间作为参考系
that referencing the time horizon
是得不出合理的结论的
is a total non sequitur.
如果说智慧只包括信息处理
If intelligence is just a matter of information processing,
然后我们继续改善这些机器
and we continue to improve our machines,
那么我们终将生产出超级智能
we will produce some form of superintelligence.
但我们无法预估所需时间
And we have no idea how long it will take us
从而为安全发展创造条件
to create the conditions to do that safely.
我再重复一遍
Let me say that again.
我们无法预估所需时间
We have no idea how long it will take us
从而为安全发展创造条件
to create the conditions to do that safely.
你们可能没注意到 50年的概念已今非昔比
And if you haven’t noticed, 50 years is not what it used to be.
这是用月来衡量50年的样子
This is 50 years in months.
红点代表苹果手机发布的时间
This is how long we’ve had the iPhone.
这是《辛普森一家》在电视上的播出时间
This is how long “The Simpsons” has been on television.
50年时间并不是很长
Fifty years is not that much time
要做好准备面对 人类历史上前所未有的挑战
to meet one of the greatest challenges our species will ever face.
就像刚才说的 我们对确定会来之事
Once again, we seem to be failing to have an appropriate emotional response
做出了不合理的回应
to what we have every reason to believe is coming.
计算机科学家斯图尔特·罗素给出了一个极好的类比
The computer scientist Stuart Russell has a nice analogy here.
他说 想象我们从外太空接收到一条讯息
He said, imagine that we received a message from an alien civilization,
上面写着:
which read:
“地球上的人类
“People of Earth,
我们将在五十年后到达你们的星球
we will arrive on your planet in 50 years.
做好准备吧”
Get ready.”
于是我们就开始倒计时 直到它们的“母舰”着陆吗?
And now we’re just counting down the months until the mothership lands?
在这种情况下我们会感到更紧迫
We would feel a little more urgency than we do.
另外一个试图安慰我们的理由是
Another reason we’re told not to worry
那些机器必须拥有和我们一样的价值观
is that these machines can’t help but share our values
因为它们将会是我们自身的扩展
because they will be literally extensions of ourselves.
它们将会被嫁接到我们的大脑上
They’ll be grafted onto our brains,
我们将会成为它们的边缘系统
and we’ll essentially become their limbic systems.
现在我们再思考一下
Now take a moment to consider
最安全的 也是唯一经慎重考虑后
that the safest and only prudent path forward,
的建议
recommended,
是将这项技术直接植入我们大脑
is to implant this technology directly into our brains.
这也许确实是最安全的 也是唯一慎重的发展方向
Now, this may in fact be the safest and only prudent path forward,
但通常在把它塞进脑袋之前
but usually one’s safety concerns about a technology
我们不得不充分考虑这项技术的安全性
have to be pretty much worked out before you stick it inside your head.
[笑声]
[Laughter]
更深一层的问题是:仅仅制造出超级人工智能
The deeper problem is that building superintelligent AI on its own
可能要比
seems likely to be easier
既制造超级人工智能
than building superintelligent AI
又要让其能与我们的思想无缝对接的
and having the completed neuroscience
完整的神经系统要简单很多
that allows us to seamlessly integrate our minds with it.
而做这些研究的公司或政府
And given that the companies and governments doing this work
很可能将彼此视作竞争对手
are likely to perceive themselves as being in a race against all others,
因为赢得了比赛就意味着称霸了世界
given that to win this race is to win the world,
前提是不在刚成功后就将其销毁
provided you don’t destroy it in the next moment,
所以结论是:简单的选项
then it seems likely that whatever is easier to do
一定会被先实现
will get done first.
但很遗憾 除了建议更多人去思考这个问题
Now, unfortunately, I don’t have a solution to this problem,
我对此并无解决方案
apart from recommending that more of us think about it.
我觉得在人工智能问题上
I think we need something like a Manhattan Project
我们需要一个“曼哈顿计划”
on the topic of artificial intelligence.
不是讨论如何制造人工智能 因为我们一定会这么做
Not to build it, because I think we’ll inevitably do that,
而是去避免军备竞赛
but to understand how to avoid an arms race
最终以一种有利于我们的方式去打造它
and to build it in a way that is aligned with our interests.
当你在谈论一个可以自我改造的
When you’re talking about superintelligent AI
超级人工智能时
that can make changes to itself,
我们似乎只有一次正确搭建初始系统的机会
it seems that we only have one chance to get the initial conditions right,
而这个正确的初始系统
and even then we will need to absorb
需要我们在经济政治上做出很大的努力
the economic and political consequences of getting them right.
但是当我们承认
But the moment we admit
信息处理是智慧的源头
that information processing is the source of intelligence,
承认一些电脑系统是智能的基础
that some appropriate computational system is what the basis of intelligence is,
承认我们会不断改善这些系统
and we admit that we will improve these systems continuously,
承认我们现有的认知远没有达到极限
and we admit that the horizon of cognition very likely far exceeds
将很可能被超越
what we currently know,
我们又必须同时承认
then we have to admit
某种意义上 我们正在创造一个新的“上帝”
that we are in the process of building some sort of god.
现在正是思考人类是否
Now would be a good time
能与这个“上帝”和睦相处的最佳时机
to make sure it’s a god we can live with.
非常感谢
Thank you very much.
[掌声]
[Applause]

发表评论

译制信息
视频概述
听录译者

收集自网络

翻译译者

B11101001

审核员

瞌睡虫儿

视频来源

https://www.youtube.com/watch?v=8nt3edWLgIg

相关推荐