人工智能的真相（2023）（3）

So, first of all, what we've done is
我们收集了无数的
we've taken millions and millions
面部图像并训练了一种算法
of facial images and trained an algorithm.
这种算法♥会♥关注脸部的颜色
The algorithm is looking at colour,
纹理和各种脸型
texture, various shapes of the face.
当然 在这个例子里 还要看年龄
And of course, in this case, age as well.
它训练出了一种算法
And it's trained an algorithm to
可以区分不同的面孔
distinguish between various faces.
这种算法都会应用在哪些地方呢
And what are the applications of this?
我们如何利用它来造福社会
How can it be used to benefit society?
下面这个例子 我们把这个软件用于
So, in this case, we are using this kind of software
犯罪调查
for criminal investigation,
比方说 你的数据库里
for example, you have millions and
有成百上千万张图片
millions of images in a database.
你需要从中快速提取出相关的图像
You want to quickly pull out the relevant image from that.
对人类来说 这是一项非常耗费体力的任务
For humans, it will be a very laborious task and computers
而计算机可以非常快地完成
can do it very, very quickly.
这是其中一点 -最后一点就是隐私问题
So that's one of the things. - Finally, privacy.
在大数据和社交媒体的时代
Privacy is so important in the world
隐私非常重要
of big data and social media and so on.
此技术又是如何避免侵犯隐私的呢
How does privacy come into all of this?
所以我再说一遍 这一点非常非常重要
So, as I say, again, it's very, very important that we make sure
我们要确保正确的数据确实是被用于训练
that the right data is used actually to train,
还有我们测试的时候
and also when we test,
我们也要使用正确的数据进行测试
we use the right data to test it as well.
好的 非常感谢 哈桑
OK, thank you so much, Hassan,
谢谢你来为我们展示你的软件 谢谢你
for coming on and showing us your software. Thank you.
现在...
Now...
人工智能显然变得越来越智能
AI is clearly becoming more capable
而且越来越复杂
and more sophisticated by the day.
还有 它变化的速度 进步的速度
And the pace of change, the pace of improvement
真的非常了不起
is really quite remarkable.
但是有些人担心这可能会给我们带来灾难
But some people fear that this could be leading us to disaster.
在此之前 我问过德米斯·哈萨比斯
Earlier, I asked Demis Hassabis
他对人工智能的未来都有什么担忧
about his concerns for the future of AI.
是的 我所担心的风险有两类
Yeah, I think there are two categories of risks that I worry about.
一类是恶意行为体 无论是个人还是政体
One is bad actors, whether that's individuals or nation, rogue states
为了邪恶目的 改变通用型AI技术的用途
repurposing general purpose AI technology for bad ends,
使其背离人们开♥发♥这些技术时的初衷
you know, not for all the good they were intended to do,
通过改变AI目标等方法去做坏事
but repurposing their goals and other things to do harm.
这是我们需要担心的其中一点
So that's one thing I think we have to worry about.
说到近期的事
Probably the nearer term thing.
这种情况你可以想到的有虚假信息
You can think of things like misinformation and deepfakes
和深度伪造
as part of that bucket.
我担心的第二点是人工智能技术的风险
And then the second thing I worry about is technical AI risk,
实际上 随着这些系统
which is actually, as these systems
变得越来越复杂
become more and more sophisticated,
不是指现在的系统 甚至也不是未来几年的
so not today's systems, not even the next couple of years,
但是如果你想想未来的五到十年
but if you think about the next five to ten years, you know,
这些系统的能力变得越来越强
these systems will become ever more capable.
它们会变得更有能力 可以制定计划
They'll become more able to do things like planning
执行任务
and execute goals
当然更有用的是做一些世界规模的工作
and more useful, of course, do actions in the world.
但是 比如我们为游戏创建的
But then also the risks associated with those types
基于代理的那一类系统
of more agent-based systems
也伴随着一些风险
like we build for games,
它们在变得有用的同时
you know, they will carry more risks
也蕴含着更多的风险
as well as being more useful.
所以我们必须更好地了解这些系统
So we have to understand those systems better,
能够对它们进行分♥析♥以确保
be able to analyse them and make sure
我们可以控制它们
that we can keep control of them.
好的
Ok.
对于一些研究者来说 人工智能的终极目标
For some researchers, the end goal of artificial intelligence
就是所谓的AGI 即人工通用智能
is what's called AGI, artificial general intelligence.
现在对于什么是AGI并没有精确的定义
Now, there is no precise definition of what AGI is,
但从广义上讲 它意味着我们拥有一个可以
but broadly it means that we have an AI system
以人类水平或接近于人类水平
which can perform a wide range of tasks
执行各种任务的人工智能系统
at or very close to the level of human beings.
我们在好莱坞电影中
And we see AGI in Hollywood movies
在电子游戏和书里都能看到AGI
and video games and books.
如果人工通用智能真的成功了
And if AGI really, really succeeded,
那它就能做所有
then it would be able to do anything
人类可以做的事情了
that a human being could do.
现在即使人工通用智能还做不到这一点
Now, even AGI that was much less capable of that
它也将彻底改变社会
would be completely transformative for society.
但是如果人工智能可以和我们一样聪明
But if AI can be as smart as us,
那么它当然也可以更胜一筹
then of course it could be even smarter.
那时 人类又该如何自处呢
And then where would that leave humanity?
那种水平的机器智能是否会
Could that level of machine intelligence
威胁人类的未来
threaten the future of humanity?
有人坚信我们现在
And some people believe that's an issue
就应警惕这一问题 以免追悔莫及
we need to take seriously now before it's too late.
针对这一问题 欢迎斯图尔特·罗素教授
To discuss this issue, please welcome Professor Stuart Russell.
欢迎 斯图尔特
Welcome, Stuart.
我们刚刚在聊对人工智能未来发展的恐慌
So we're talking about fears about the future of AI,
就我所知的影视书籍
and in the movies and the books that I read,
科幻电影中
the science fiction movies that I read,
人类的问题始于人工智能的
the problems for humanity start when the AI becomes
自知 自觉 自醒
conscious or self-aware or sentient.
人工智能开始考虑自身
It starts to think for itself
然后决定不再围着人类打转
and then decides that it's not interested in us any more.
或者嫌我们碍事
Or maybe we're in the way.
这种担心有必要吗
Is that what we should be worried about?
简单来说 没有 -好
In a word, no. - Ok.
关键在于能力
What matters actually is competence.
对吧
Right?
在于机器做什么 尤其是对人类做什么
It's what the machine does, or what it does to us in particular.
如果从地球上其他生物的角度
So if you think about this from the point of view
来看待这一问题
of other species on Earth,
它们为什么害怕我们
why are they worried about us?
因为我们对它们的所作所为
It's because of what we do to them.
它们怕我们 不是因为我们有自我意识
They're not worried about whether we're conscious or not.
而是因为我们是智慧生物
They're worried about the fact that we are intelligent
而且能主宰地球
and we get to decide what happens on Earth.
这才是恐惧的根源
And so that's really the source of concern.
如果将来 我们面对机器
If we then put ourselves in that position
正如动物面对我们
with respect to the machines,
如果我们赋予机器高于我们的智慧
if we make the machines more intelligent than us,
那我们就得扪心自问
then we have to ask a question.
"人类要如何在比自己更强大的存在面前
"How do we humans retain power forever over entities
保留永久的权力"
that are more powerful than ourselves?"
如果未来真的有超高智慧人工智能
If we reach a future where there's sort of superintelligent AI,
我们就不能关掉它吗 实在不行
can't we just switch it off, or, in the worst case,
就用核弹炸了数据中枢
just nuke the data centres?
这样不能解决问题吗
Isn't that going to solve the problem?
既然是超高智慧 那它肯定早就料到了
Since it's superintelligent, it will already have thought
我们有可能会想要关掉它
about the possibility that we might try to switch it off
如果它想进行一些我们不喜欢的行动的话
if it wants to pursue some objectives that we don't like.
所以 它可能早就将自己
And so it will probably have copied
复♥制♥到全世界的数据中枢里
itself onto all the data centres in the world.
它还可能会入侵我们的武器系统
It'll probably have infiltrated our weapons systems
致使它们失效
so they won't work,
甚至利用它们来对付我们
or so that it could even use them against us.
我认为眼下唯一的出路
And I think the only solution here
就是从一开始就不要
is not to be in that conflict
陷入这种冲突
in the first place,
确保绝不会有
to actually make sure that we never have AI systems
执行目标不符合人类未来期望的
that are pursuing objectives that are not what we want
人工智能系统
the future to be like.
目前确实存在一些困难 对吧
Now, of course, there's a few difficulties with that, right?
你们马上会想