Nick Bostrom: What happens when our computers get smarter than we are?
尼克·博斯特罗姆: 当电脑比我们还聪明时会发生什么?
Nick Bostrom asks big questions: What should we do, as individuals and as a species, to optimize our long-term prospects? Will humanity’s technological advancements ultimately destroy us? Full bio
Double-click the English transcript below to play the video.
philosophers and computer scientists,
哲学家和电脑学家一起工作,
the future of machine intelligence,
things are sort of science fiction-y,
human condition.
guests on this planet,
was created one year ago,
would be 10 minutes old.
two seconds ago.
world GDP over the last 10,000 years,
想一下在过去一万年间的世界 GDP 状况。
to plot this for you in a graph.
for a normal condition.
of this current anomaly?
如此不寻常的现状?
through human history,
不断积累下来的果实。
advances extremely rapidly --
so very productive.
to the ultimate cause.
distinguished gentlemen:
tokens, an incredible feat.
一个难以置信的成就。
superstring revolution.
this is what we find:
这是我们的发现:
in the exact way it's wired.
be too complicated, however,
been 250,000 generations
take a long time to evolve.
需要很长的时间来进化得到。
to intercontinental ballistic missiles.
到发射洲际导弹。
that everything we've achieved,
changes that made the human mind.
is that any further changes
the substrate of thinking
enormous consequences.
think we're on the verge
a profound change in that substrate,
about putting commands in a box.
是把指令输入到一个箱子里。
handcraft knowledge items.
for some purposes,
you couldn't scale them.
what you put in.
in the field of artificial intelligence.
around machine learning.
representations and features,
often from raw perceptual data.
that the human infant does.
limited to one domain --
between any pairs of languages,
on the Atari console.
the same powerful, cross-domain
as a human being has.
algorithmic tricks
how to match in machines.
将这些技巧用于机器。
to match those tricks?
让机器复制这种能力?
leading A.I. experts,
做了一次问卷调查,
and one of the questions we asked was,
there is a 50 percent probability
human-level machine intelligence?"
as the ability to perform
as an adult human,
完成得至少和一名成年人一样好。
within some limited domain.
而不是仅限于一些领域。
group of experts we asked.
much later, or sooner,
也有可能提前实现。
limit to information processing
处理信息的能力的最终点,
the limits in biological tissue.
at 200 hertz, 200 times a second.
大约位于 200 赫兹,每秒 200 次。
operates at the Gigahertz.
都以千兆赫的频率运行。
100 meters per second, tops.
最多 100 米每秒。
at the speed of light.
to fit inside a cranium,
of a warehouse or larger.
lies dormant in matter,
lay dormant throughout human history,
the power of artificial intelligence.
an intelligence explosion.
about what is smart and what is dumb,
什么是聪明什么是笨的时候,
roughly like this.
or whoever your favorite guru is.
或者其他大师。
of artificial intelligence,
probably more like this:
at zero intelligence,
years of really hard work,
mouse-level artificial intelligence,
cluttered environments
of really hard work, lots of investment,
和投资之后,
chimpanzee-level artificial intelligence.
of really, really hard work,
artificial intelligence.
we are beyond Ed Witten.
我们能超越 Ed Witten。
at Humanville Station.
to questions of power.
twice as strong as a fit human male.
两个健康男性那么强壮。
and his pals depends a lot more
更多取决于
what the chimpanzees do themselves.
而不是猩猩能做到什么。
on what the superintelligence does.
那个超级智慧体要做什么。
that humanity will ever need to make.
at inventing than we are,
on digital timescales.
a telescoping of the future.
that you could have imagined
in the fullness of time:
of minds into computers,
和大脑到电脑的传输,
with the laws of physics.
develop, and possibly quite rapidly.
technological maturity
it would be able to get what it wants.
它能得到它想要的东西。
be shaped by the preferences of this A.I.
这个超级智慧体的喜好所主宰。
what are those preferences?
这些喜好是什么呢?
avoid anthropomorphizing.
every newspaper article
因为每一个关于人工智能的未来
has a picture of this:
to conceive of the issue more abstractly,
用更抽象的方法看待这个问题,
as an optimization process,
into a particular set of configurations.
一个特殊组合结构的过程。
a really strong optimization process.
非常强大的优化过程。
available means to achieve a state
conenction between
would find worthwhile or meaningful.
to make humans smile.
or amusing actions
它能做出有用或好笑的表演,
effective way to achieve this goal:
能达到这个效果:
muscles of humans
a difficult mathematical problem.
是解出很难的数学题,
to get the solution to this problem
into a giant computer,
an instrumental reason
给予人工智能一个模式型的理由
might not approve of.
problem from being solved.
让这个数学问题不能被解出。
go wrong in these particular ways;
这种事情不会错到这样的地步,
optimization process
that your definition of x
in many a myth.
he touches be turned into gold.
希望他碰到的所有东西都能变成金子。
she turns into gold.
optimization process
or poorly specified goals.
后果会是什么。
sticking electrodes into people's faces,
如果一个电脑开始在人类脸上插电极,
if we've grown dependent on the system --
如果我们变得非常依赖这个系统:
to the Internet?
flicked the off switch to humanity,
没有关掉人类的开关,
for example, right here.
an intelligent adversary;
and plan around them.
at that than we are.
that we have this under control here.
表示我们能控制所有事情。
a little bit easier by, say,
我们应该试着,比如,
from which it cannot escape.
the A.I. couldn't find a bug.
能让它逃出的漏洞吗?
find bugs all the time,
to create an air gap,
using social engineering.
out there somewhere
her account details
from the I.T. department.
around in your internal circuitry
环绕复杂缠绕的电极
can use to communicate.
you up to see what went wrong with you,
to a really nifty technology,
非常漂亮的科技蓝图,
that the A.I. had planned.
秘密的副作用。
not be confident in our ability
对我们能够永远控制
locked up in its bottle forever.
表示过度自信。
is to figure out
such that even if -- when -- it escapes,
fundamentally on our side
this difficult problem.
that this problem can be solved.
a long list of everything we care about,
所有我们在乎的事情,
in some computer language
that uses its intelligence
用它自己的智慧
in such a way that it is motivated
that it predicts we would approve of.
或者去做我们会赞成的事情。
its intelligence as much as possible
very good for humanity.
for the intelligence explosion
in just the right way
need to match ours,
要和我们的价值观相辅相成,
how the A.I. behaves,
that the A.I. might encounter
前所未有的情况下,
与我们的价值观相辅相成。
that would need to be solved, sorted out:
uncertainty and so forth.
to be solved to make this work
a superintelligent A.I.,
is a really hard challenge.
challenge on top of that.
how to crack the first challenge
the additional challenge
work out a solution
by the time it is needed.
the entire control problem in advance
预先解决全部的控制性问题,
can only be put in place
architecture where it will be implemented.
that we solve in advance,
to the machine intelligence era
that is well worth doing
things turn out okay,
look back at this century
the one thing we did that really mattered
我们所做的最最重要的事情,
ABOUT THE SPEAKER
Nick Bostrom - PhilosopherNick Bostrom asks big questions: What should we do, as individuals and as a species, to optimize our long-term prospects? Will humanity’s technological advancements ultimately destroy us?
Why you should listen
Philosopher Nick Bostrom envisioned a future full of human enhancement, nanotechnology and machine intelligence long before they became mainstream concerns. From his famous simulation argument -- which identified some striking implications of rejecting the Matrix-like idea that humans are living in a computer simulation -- to his work on existential risk, Bostrom approaches both the inevitable and the speculative using the tools of philosophy, probability theory, and scientific analysis.
Since 2005, Bostrom has led the Future of Humanity Institute, a research group of mathematicians, philosophers and scientists at Oxford University tasked with investigating the big picture for the human condition and its future. He has been referred to as one of the most important thinkers of our age.
Nick was honored as one of Foreign Policy's 2015 Global Thinkers .
His recent book Superintelligence advances the ominous idea that “the first ultraintelligent machine is the last invention that man need ever make.”
Nick Bostrom | Speaker | TED.com