人工智能的真相（2023）（2）

也让人工智能很难掌握它的玩法
and it makes it very, very hard for artificial intelligence.
所以它成了人工智能的重要学习目标
And that's what made it an important target for AI.
围棋的分支因子大约为二百五十
Go has a branching factor of around about 250.
记得吗 分支因子
Remember, that branching factor,
指的是玩家在任意时刻
that's the average number of moves
下一步的走法选项数的平均值
that a player can make at any given time.
我们来拿这些数字举个例子
Let's illustrate those numbers.
你们的座位下方有一张卡片
If you take a look under your seat, you should find a card.
看到了吗 看一下这张卡片
OK? So take a look at the card.
暂时不用做别的 只是看一下卡片内容
You don't need to do anything yet. Just take a look at the card.
好的 如果你的卡片上有井字棋方格
Now, if your card has a tic-tac-toe grid on it,
请起立
stand up now.
好的 看
OK, so here we are.
我们有四位观众
We've got four people in the audience
举着井字棋方格
holding a tic-tac-toe grid.
这就是井字棋的分支因子
That's the branching factor of tic-tac-toe.
如果你的卡片是国际象棋棋盘
If your card has a chess board on it,
像这样
which looks like this,
请起立
then stand up now.
现在有三十五位观众起立了
That's 35 of our audience members that are standing up now.
这就是国际象棋的分支因子
That's the branching factor of the game of chess.
最后 如果你座位下放着的是围棋
And, finally, if you have a Go board underneath your seat,
看起来像这样
something that looks like this,
请起立
stand up now.
基本上全体观众都起立了
That's pretty much the entire audience.
这就是围棋的分支因子
That's the branching factor of the game of Go,
你们可以把它和国际象棋的分支因子
and compare that to the branching factor of the game of chess
以及刚开始起立的那四个人比较一下
and those four people that were stood up there at the beginning.
好的各位 你们可以坐下了 谢谢
OK, everybody, you can sit down now. Thank you.
我们来思考一下围棋里的
Let's think about the number of game
游戏状态数量
states there are in the game of Go.
它会是这个数
And it's this number -
十的一百七十次方
ten to the power of 170.
我们都难以在屏幕上放下这个数
We can hardly fit that number on the screen.
日常生活中
We never encounter numbers that large
我们从来不会碰到这么大的数字
in our everyday life.
它确实是字面意思上的天文数字
They are literally astronomical.
我们今天学到的
And what we've learned today is a
是关于人工智能的很重要的一课
really important lesson about AI.
通过尝试所有可能性
Trying to solve a problem by simply
来解决问题的方法
looking at all the alternatives
被称为"暴♥力♥破解"
is called brute force.
而人工智能的这一课就是
And the AI lesson is this.
对于像围棋这样的问题来说 暴♥力♥破解没有用
For problems like playing Go, brute force doesn't work.
这么做行不通
It will never work.
因为可能性的数量是个天文数字
Because the numbers are just too astronomically large.
即使我们把整个宇宙变成一台计算机
Even if we turned the entire universe into a computer,
它也无法仅靠暴♥力♥破解来下围棋
it wouldn't be able to play just using brute force.
所以我们需要某个方法
So we need some way of reducing the
来减少需要检验的可能性
number of alternatives to check.
而这种方法常被称为"启发法"
And these are often called heuristics.
我们可以使用机器学习来帮助我们建构启发法
And we can use machine learning to help us build heuristics.
我问了德米斯·哈萨比斯
And I asked Demis Hassabis
DeepMind是如何训练人工智能下围棋的
how DeepMind trained AI to play Go.
对阿尔法GO来说
In the case of AlphaGo,
阿尔法GO学习如何下围棋
AlphaGo learnt for itself how to play Go,
以及该用什么策略等等的方式是
and the strategies it would use and so on,
数百万次地自己和自己下棋
by playing against itself many millions of times.
它使用了一些学习系统
And it used learning systems,
比如强化学习和树搜索技术
so reinforcement learning and tree search techniques,
来确定哪种围棋策略
in order to figure out for itself what the right strategies
可以获胜
were to be good at Go.
2016年 阿尔法GO对战并战胜李世乭
In 2016, AlphaGo played against and beat Lee Sedol,
世界上最好的围棋手之一
one of the world's greatest players.
出乎了很多人的意料
And a lot of people didn't expect that.
当然 这件事很棒的一点是
And so, of course, the cool thing about that is that, actually,
阿尔法GO不仅在2016年击败了世界冠军
AlphaGo, not only did it beat the world champion in 2016,
还创造了全新的围棋策略
it also came up with completely new ideas about Go
是其他人类玩家从未想到过的策略
new strategies that no human players had ever thought of,
即使围棋已存在上千年了
even though Go is several thousands of years old.
我以为阿尔法GO是基于概率计算运行的
它不过是个机器
但当我看到它下这步棋时 我改变了看法
阿尔法GO绝对是有创造力的
这步棋很有想法下得很妙
阿尔法GO的成就是人工智能世界
AlphaGo's achievement was a really big step forward
迈出的很大一步
in the world of AI.
因为围棋十分复杂
Because of its complexity,
我们一度认为人工智能需要十年甚至更久
we thought that Go was a barrier that AI wouldn't overcome
才能跨越这个障碍
for another decade, or maybe even more.
类似围棋的游戏
And games like Go have been a really
为人工智能提供了非常好的训练场地
great training ground for AI,
而我们通过训练人工智能玩游戏
and the knowledge that we've derived
获取的知识
from building AI to play games
让科学有了重大突破
has led to major breakthroughs in science,
我们之后会提到
as we're going to see later on.
人工智能也越来越多地进入我们的生活
And, increasingly, AI is embedded in our lives,
目之所及都有它们的身影
everywhere that we look.
让我们回到讲座开始时
Let's go back to that video that we showed you
我们放的那个影片
at the beginning of the lecture.
这次看的时候 我希望你们数数看
And this time, when you watch it, I want you to try to keep count
人工智能出现的次数
of how many times you think the AI appears.
我今天有哪些安排
What appointments do I have today?
今天有英国皇家科学院圣诞讲座
Today is the Royal Institution Christmas lecture.
请播放一首提神醒脑的歌♥
Play me a "wake me up" morning mix.
* 你的生活又过去了一天 *
* Another day's passing in your life... *
* 我听到你说 *
* I heard you saying *
* 现在是我们的时代 *
* Now is our time *
* 我正在高速公路上飞驰 *
* I'm on the highway I'm on my way *
麦克 今晚的讲座你准备好了吗
Mike, you all set for tonight?
我喜欢你的背景
I love your background!
是啊 我正在感受圣诞氛围
Yeah, I'm just getting in the Christmas mood.
电脑 我今晚需要穿外套吗
Computer, will I need a coat tonight?
是的 请携带外套
Yes, take a coat.
* 这一次我不会停下 *
* This time I ain't Ain't gonna stop *
* 我们正要登上顶峰 *
* We're on a ride to reach the top *
* 我口袋空空但风趣幽默 *
* Ain't got no money But I'm full of fun *
* 你无法忽视 不会选择跑开 *
* You can't ignore it No you won't run away *
我想向你们介绍这段影片的制♥作♥者
I'd like to introduce you to the person who made that video,
科学作者兼主持人 埃米莉·格罗斯曼博士
science author and broadcaster Dr Emily Grossman.
你好呀 -你好
Hiya. - Hello.
你好
Hi.
欢迎 埃米莉
And welcome, Emily!
谢谢
Thank you.
我们刚才看到的是你的日常吗
Was what we watched just a normal day for you?
是的 差不多
Yeah, pretty much.
我的一天基本就这样
I mean, it was a pretty typical day,
只是不会每天都来参加圣诞讲座
although I don't usually end up at the Christmas lectures.
好的 埃米莉 在你上台前 我让观众
OK. Now, Emily, before you came on, I asked the audience to watch
在看影片时记下人工智能
that video and to keep track of how many times they thought
出现的次数
that AI appeared in it.
我们要不要问问观众
Shall we ask the audience?
好啊 来吧 -好的 交给你
Yeah, let's do that. - OK, over to you.
你觉得呢 有多少次
OK, what do you think? How many times?
十四 -十四次
14. - 14 times?!
有没有 有没有其他答案
Any other...? Any other suggestions?
这里
Over here, yeah!
十 -十次
Er, ten. - Ten times.
埃米莉 你觉得那个视频里
Emily, how many times did you think
人工智能出现了几次
that AI appeared in that video?
实际上是介于两者之间
Well, actually, somewhere between the two.
视频中人工智能其实出现了十二次
It was actually 12 times that AI was in the video.
所以你们俩都非常接近了