This computer is learning to read your mind | DIY Neuroscience, a TED series

Translator: Joseph Geni Reviewer: Krystian Aparta

翻译人员: yiwen zhang 校对人员: Wanting Zhong

Greg Gage: Mind-reading. You've seen this in sci-fi movies:

格雷戈 · 盖奇（Greg Gage）：读心术。 你在科幻电影中曾经看到过：

machines that can read our thoughts.

那是可以读出我们想法的机器。

However, there are devices today

然而，如今有很多机器

that can read the electrical activity from our brains.

可以读出我们大脑中的电波。

We call this the EEG.

我们把它叫做 “EEG”。

Is there information contained in these brainwaves?

这些脑电波中含有信息吗？

And if so, could we train a computer to read our thoughts?

如果含有信息，我们可以训练 计算机读懂我们的思想吗？

My buddy Nathan has been working to hack the EEG

我的好友内森一直 致力于研究如何破译 EEG

to build a mind-reading machine.

以建造一台可以读心的机器。

[DIY Neuroscience]

【DIY 神经科学】

So this is how the EEG works.

这就是 EEG 的工作原理。

Inside your head is a brain,

在你的脑袋里有一个大脑，

and that brain is made out of billions of neurons.

大脑是由数十亿个神经元构成的，

Each of those neurons sends an electrical message to each other.

每个神经元都在 互相传送电子信息，

These small messages can combine to make an electrical wave

这些微小的信息可以结合在一起

that we can detect on a monitor.

形成我们能在显示器上 探测到的电波。

Now traditionally, the EEG can tell us large-scale things,

传统意义上而言， EEG 能告诉我们大维度的事情，

for example if you're asleep or if you're alert.

例如你是睡着还是清醒。

But can it tell us anything else?

但它可以告诉我们其它事情吗？

Can it actually read our thoughts?

它是否能够读出我们心中所想？

We're going to test this,

我们要去测试这一点，

and we're not going to start with some complex thoughts.

而我们不打算从一些 复杂的想法开始。

We're going to do something very simple.

我们打算做一件非常简单的事情。

Can we interpret what someone is seeing using only their brainwaves?

我们可以仅仅依据脑电波 判读出一个人看到了什么吗？

Nathan's going to begin by placing electrodes on Christy's head.

内森先要在克里斯蒂的头上安装电极。

Nathan: My life is tangled.

内：我的人生一团糟。

(Laughter)

（笑声）

GG: And then he's going to show her a bunch of pictures

格：之后他会给她看一些图片，

from four different categories.

这些图片出自四种不同类别：

Nathan: Face, house, scenery and weird pictures.

内森：面孔，房子， 风景和古怪的图片。

GG: As we show Christy hundreds of these images,

格：当我们向克里斯蒂 展示数百张这种图片时，

we are also capturing the electrical waves onto Nathan's computer.

我们也在内森的电脑上 捕捉她的脑电波。

We want to see if we can detect any visual information about the photos

我们想知道我们是否 能通过这些脑电波，

探测到任何与这些 图片相关的视觉信息。

contained in the brainwaves,

so when we're done, we're going to see if the EEG

在实验结束后， 我们将会看到 EEG 是否

can tell us what kind of picture Christy is looking at,

可以告诉我们克里斯蒂 在看哪种图片。

and if it does, each category should trigger a different brain signal.

如果可以，不同类别的图片 应该会触发不同的大脑信号。

OK, so we collected all the raw EEG data,

好的，我们收集完了 所有的原始 EEG 数据，

and this is what we got.

这就是我收集到的样子。

It all looks pretty messy, so let's arrange them by picture.

它看上去很混乱，于是我们 根据图片类别将它们排序。

Now, still a bit too noisy to see any differences,

现在，还是有点太嘈杂， 无法看出任何区别，

but if we average the EEG across all image types

但是如果我们根据图片 出现的时间将信号对齐，

by aligning them to when the image first appeared,

并对每种图片类别的 EEG 取平均值，

we can remove this noise,

我们就能移除其中的噪声。

and pretty soon, we can see some dominant patterns

很快，我们就可以从各个类别中

emerge for each category.

看到一些明显的规律。

Now the signals all still look pretty similar.

现在这些信号看起来还是很相似，

Let's take a closer look.

让我们再仔细看看。

About a hundred milliseconds after the image comes on,

大约在一张图片 出现后的一百毫秒后，

we see a positive bump in all four cases,

我们在四个类别中 都看到了正向波动，

and we call this the P100, and what we think that is

我们把它叫作 P100 我们认为这是

is what happens in your brain when you recognize an object.

当你识别物体时 大脑中发生的活动。

But damn, look at that signal for the face.

但是见鬼，看看“面孔“ 图片对应的信号，

It looks different than the others.

它看起来与众不同，

There's a negative dip about 170 milliseconds

在图片出现后的约 170 毫秒时，

after the image comes on.

出现了负向波动。

What could be going on here?

这里可能发生了什么？

Research shows that our brain has a lot of neurons that are dedicated

研究显示，我们大脑有大量神经元

专门负责识别人类的面孔，

to recognizing human faces,

so this N170 spike could be all those neurons

所以这个 N170 负波可能是

firing at once in the same location,

所有这些神经元 在同一地方同时激活，

and we can detect that in the EEG.

而我们可以在 EEG 中探测到。

So there are two takeaways here.

于是从中得出两个结论，

One, our eyes can't really detect the differences in patterns

第一，在没有经过平均化降噪时，

我们的眼睛并不能真的 识别脑波规律的不同；

without averaging out the noise,

and two, even after removing the noise,

第二，即使移除噪声后，

our eyes can only pick up the signals associated with faces.

我们的眼睛也只能 识别出和面孔有关的信号。

So this is where we turn to machine learning.

于是我们在此转而借助机器学习。

Now, our eyes are not very good at picking up patterns in noisy data,

我们的眼睛并不擅长 在嘈杂的数据中发现规律，

but machine learning algorithms are designed to do just that,

但是机器学习算法的设计 初衷就是解决这类问题。

so could we take a lot of pictures and a lot of data

所以我们可以将许多图片和数据

and feed it in and train a computer

输入到电脑中进行训练，

to be able to interpret what Christy is looking at in real time?

从而实时判断克里斯蒂正在看什么。

We're trying to code the information that's coming out of her EEG

我们尝试将她的 EEG 信息

进行实时编码，

in real time

and predict what it is that her eyes are looking at.

并预测她眼睛在看的东西。

And if it works, what we should see

如果这样有效，我们应该能看到

is every time that she gets a picture of scenery,

每次她看到风景的图片时，

it should say scenery, scenery, scenery, scenery.

机器应该显示风景， 风景，风景，风景。

A face -- face, face, face, face,

如果她看到面孔，机器则显示 面孔，面孔，面孔，面孔，

but it's not quite working that way, is what we're discovering.

但是我们发现， 实际上并非如此。

(Laughter)

（笑声）

OK.

好的。

Director: So what's going on here? GG: We need a new career, I think.

导演：所以发生了什么？ 格：我觉得我们应该转行。

(Laughter)

（笑声）

OK, so that was a massive failure.

好吧，所以刚刚那是个重大失败。

But we're still curious: How far could we push this technology?

但是我们依然好奇： 我们能这项技术发展到多深？

And we looked back at what we did.

于是我们回顾了我们的做法。

We noticed that the data was coming into our computer very quickly,

我们发现电脑在飞快地获取数据，

without any timing of when the images came on,

但没有对图片出现的时间进行计时，

and that's the equivalent of reading a very long sentence

这等同于读一个

在单词间没有空格的长句。

without spaces between the words.

It would be hard to read,

这样的句子很难读懂，

but once we add the spaces, individual words appear

不过一旦我们添加了空格， 我们就能看到独立的单词，

and it becomes a lot more understandable.

句子也就变得容易理解多了，

But what if we cheat a little bit?

但如果我们做一点弊呢？

By using a sensor, we can tell the computer when the image first appears.

通过使用传感器， 我们能告诉电脑每张图片出现的时机。

That way, the brainwave stops being a continuous stream of information,

这样，脑波就不再是 一个没有间断的信息流，

and instead becomes individual packets of meaning.

而是变成了一个个 有意义的信息小包裹。

Also, we're going to cheat a little bit more,

另外，我们还要再做一点弊，

by limiting the categories to two.

把图片限制到两个类别。

Let's see if we can do some real-time mind-reading.

让我们看看我们是否 能够进行实时读心。

In this new experiment,

在这个新实验中，

we're going to constrict it a little bit more

我们将限制实验条件：

so that we know the onset of the image

我们会知道图片出现的时间，

and we're going to limit the categories to "face" or "scenery."

并将类别限制为 "面孔” 或 “风景” 。

Nathan: Face. Correct.

内：面孔。正确。

Scenery. Correct.

风景。正确。

GG: So right now, every time the image comes on,

格：所以现在，每当图片出现时，

we're taking a picture of the onset of the image

我们对图片出现的时刻进行记录，

and decoding the EEG.

并对 EEG 解码。

It's getting correct.

它变得越来越正确。

Nathan: Yes. Face. Correct.

内：是的。面孔。正确。

GG: So there is information in the EEG signal, which is cool.

格：所以 EEG 的信号中 包含信息，这很棒。

We just had to align it to the onset of the image.

我们仅仅需要把它 和图片出现的时刻对齐。

Nathan: Scenery. Correct.

内：风景。正确。

Face. Yeah.

面孔。没错。

GG: This means there is some information there,

格：这意味着它包含了一些信息，

so if we know at what time the picture came on,

如果我们知道图片出现的时间，

we can tell what type of picture it was,

我们就有可能根据 这些由图片诱发的电位

possibly, at least on average, by looking at these evoked potentials.

判断它是哪个类别的图片， 至少一般可以做到。

Nathan: Exactly.

内：说得没错。

GG: If you had told me at the beginning of this project this was possible,

格：如果你一开始跟我说， 这个项目有可能实现，

I would have said no way.

我会说 “怎么可能” 。

I literally did not think we could do this.

我真的觉得我们不可能做到。

Did our mind-reading experiment really work?

我们的读心术实验 真的成功了吗？

Yes, but we had to do a lot of cheating.

成功了，但是我们必须做很多弊。

It turns out you can find some interesting things in the EEG,

结果就是，你能通过 EEG 发现一些有趣的事，

for example if you're looking at someone's face,

比如，你是否在看某人的脸，

but it does have a lot of limitations.

但它确实有很多限制。

Perhaps advances in machine learning will make huge strides,

也许机器学习领域的进步 会带来重大突破。

and one day we will be able to decode what's going on in our thoughts.

有朝一日，我们能够解码心中所想。

But for now, the next time a company says that they can harness your brainwaves

可是现在来说， 当一个公司说它能利用你的脑波

to be able to control devices,

来控制一些设备，

it is your right, it is your duty to be skeptical.

你有权利和义务对此保持怀疑。

【格雷戈 · 盖奇内森 · YH · 权】