翻译：约瑟夫•杰妮校验：克里斯汀•阿帕嗒
Translator: Joseph GeniReviewer: Krystian Aparta
格雷格•盖奇：读心术 你在科幻电影里看到过：
Greg Gage: Mind-reading.You’ve seen this in sci-fi movies:
机器可以读出你的想法
machines that can read our thoughts.
然而 当今的设备也能
However there are devices today that can
从我们的大脑读出脑电活动
read the electrical activity from our brains.
我们把这种设备称为脑电图扫描器
We call this the EEG.
这些脑电波包含些什么信息吗？
Is there informationcontained in these brainwaves?
如果有的话 我们能利用电脑读出我们的想法吗？
And if so, could we train a computer to read our thoughts?
我的兄弟 南森 已经在从事“入侵”脑电图扫描器
My buddy Nathanhas been working to hack the EEG
来建造一个读心的机器的工作
to build a mind-reading machine.
［DIY神经系统科学］
[DIY Neuroscience]
这就是脑电图扫描器工作的方式
So this is how the EEG works.
你的头里面是大脑
Inside your head is a brain,
大脑是由数以亿计的神经元组成的
and that brain is made out of billions of neurons.
神经元的每一个都会给彼此发送电讯息
Each of those neurons sendsan electrical message to each other.
这些小的电讯息能够结合起来
These small messages can combineto make an electrical wave
组成我们能从监视器里检测出来的电波
that we can detect on a monitor.
传统意义上来说 脑电图扫描器能告诉我们大致的事情
Now traditionally, the EEGcan tell us large-scale things,
比如 你是否睡着了 或是醒着
for example if you’re asleepor if you’re alert.
但是 它能告诉我们其他事情吗？
But can it tell us anything else?
它能真实地读出我们的想法吗？
Can it actually read our thoughts?
接下来我们测试一下
We’re going to test this,
我们不会从复杂的想法开始
and we’re not going to start with some complex thoughts.
我们会做一些简单的事情
We’re going to do something very simple.
只用某个人的脑波 我们能破译出他们正在看的东西吗？
Can we interpret what someone is seeingusing only their brainwaves?
南森打算在克里斯蒂的头上放些电极来开始实验
Nathan’s going to begin by placingelectrodes on Christy’s head. Nathan:
我的生活像这些线一样乱如麻
My life is tangled.
格雷格·盖奇： 然后从四个不同种类
And then he’s going to show her a bunch of pictures
展示给她一堆图片
from four different categories. Nathan:
南森：面孔 房子 风景 以及古怪的图片
Face, house, sceneryand weird pictures. GG:
格雷格•盖奇：给克里斯蒂展示上百个图像时
As we show Christyhundreds of these images,
我们也把电波捕捉到了南森的电脑上
we are also capturing the electrical wavesonto Nathan’s computer.
我们想要看看
We want to see
关于这些图片的包含在脑波里的任何视觉信息 我们是否能检测出来
if we can detect any visual information about the photos
当我们完成这一步时
contained in the brainwaves, so when we’re done,
我们会看到脑电图扫描器是否
we’re going to see if the EEG
可以告诉我们克里斯蒂在看哪种图片
can tell us what kind of pictureChristy is looking at,
如果可以的话 每一种应该能激发一个不同的脑信号
and if it does, each category should trigger a different brain signal. OK,
我们收集所有脑电图扫描器的原始数据
so we collected all the raw EEG data,
这就是我们得到的一些东西
and this is what we got.
它们看起来很乱 根据图片 我们来给它们分分类
It all looks pretty messy,so let’s arrange them by picture. Now,
现在 仍然很噪杂 看不出来不同之处
still a bit too noisyto see any differences,
但如果在图像第一次出现时 把它们对齐 通过所有的图片类型
but if we average the EEGacross all image types
算出脑电波扫描器结果的平均数
by aligning themto when the image first appeared,
我们就能去除噪音 而且用不了多久
we can remove this noise, and pretty soon,
我们就能看到每一种类的主要模式显现出来
we can see some dominant patterns emerge for each category.
现在 信号看起来都很相似
Now the signals allstill look pretty similar.
我们靠近点看
Let’s take a closer look.
大约在图像出现后一百毫秒
About a hundred millisecondsafter the image comes on,
在四个情况里 都能看到一个明显的突出
we see a positive bump in all four cases,
我们把它称为P100
and we call this the P100,
我们对它的认识就是
and what we think that is
当你在识别一个物体时 你的大脑正在发生的活动
is what happens in your brainwhen you recognize an object.
但是 瞧瞧针对面孔的信号
But damn, look atthat signal for the face.
它看起来有所不同
It looks different than the others.
在图像出现后的170毫秒
There’s a negative dipabout 170 milliseconds
有一个明显的下降
after the image comes on.
这里到底发生了什么？
What could be going on here?
研究表明 我们的大脑
Research shows that our brain has a lot
有专门识别人脸的
of neurons that are dedicated
大量神经元
to recognizing human faces,
所以这个N170长针可能都是
so this N170 spike could be all those neurons
立刻在同一地点触发的神经元
firing at once in the same location,
我们可以在脑电图扫描器里检测它
and we can detect that in the EEG.
所以这里有两个简单得到的结论
So there are two takeaways here. One,
第一 如果不把噪音平均化
our eyes can’t really detectthe differences in patterns
我们的眼睛不能真正地检测出模式的不同之处
without averaging out the noise, and two,
第二 即使是在去除噪音之后
even after removing the noise,
我们的眼睛只能收到与面孔相关联的信号
our eyes can only pick upthe signals associated with faces.
所以这就是我们致力于机器学习的地方
So this is where we turnto machine learning. Now,
现在 我们的眼睛并不擅长
our eyes are not very good
在乱糟糟的数据里提取模式
at picking up patterns in noisy data,
但是 机器学习算法就是要完成这个目的
but machine learning algorithmsare designed to do just that,
所以我们能够把大量的
so could we take a lot
图片和数据
of pictures and a lot of data
带入算法 然后利用电脑
and feed it in and train a computer
就能够翻译出克里斯蒂实时看到的东西吗？
to be able to interpret what Christy is looking at in real time?
我们尝试编码
We’re trying to code the
来自她脑电图扫描器结果的实时信息
information that’s coming out of her EEG
预测她的眼睛在看什么东西？
in real time and predict what it is that her eyes are looking at.
如果成功了
And if it works,
我们看到的应该就是每次她拿到的风景图像
what we should see is every time that she gets a picture of scenery,
机器可能会发出语音 风景 风景……
it should say scenery,scenery, scenery, scenery.
面孔 面孔……
A face — face, face, face, face,
但是机器并非那样子起作用的
but it’s not quite working that way,
而是找到了我们发现的东西
is what we’re discovering.
［笑声］
(Laughter)
好的
OK. Director:
导演：发生了什么？ 格雷格·盖奇：我认为我们需要一个新的职业
So what’s going on here?GG: We need a new career, I think.
［笑声］好的
(Laughter) OK,
那人类该多失败！
so that was a massive failure.
但是我们仍然好奇：我们能把这种科技推进多远呢？
But we’re still curious:How far could we push this technology?
我们回顾一下我们所做的
And we looked back at what we did.
我们意识到
We noticed that the data was coming
无论图像在什么时候出现
into our computer very quickly,
数据很快就会在电脑里生成
without any timingof when the images came on,
这就和读一个没有字间距的长句子
and that’s the equivalentof reading a very long sentence
的意义相同
without spaces between the words.
读起来很费劲
It would be hard to read,
但是一旦我们加了间距 单个词语就出现了
but once we add the spaces, individual words appear
这就更容易理解了
and it becomes a lot more understandable.
但如果我们稍稍作弊了会怎样？
But what if we cheat a little bit?
使用一个传感器
By using a sensor,
当图像第一次出现时 我们就能告诉电脑
we can tell the computer when the image first appears.
那样 脑波就会停止生成一连串的信息流
That way, the brainwave stops beinga continuous stream of information,
相反会成为独立的数据包
and instead becomesindividual packets of meaning. Also,
同时 我们打算再多作弊一点
we’re goingto cheat a little bit more,
把种类限定在两种
by limiting the categories to two.
我们来看看是否能做一些实时的读心术
Let’s see if we can dosome real-time mind-reading.
在这个新的实验里
In this new experiment,
我们将更多地设限
we’re going to constrict it a little bit more
所以 我们知道图像的开始
so that we know the onset of the image
以及 我们将把种类限制成面孔 或 风景
and we’re going to limitthe categories to”face” or”scenery.” Nathan:
南森：面孔 对的 风景
Face. Correct. Scenery.
对
Correct. GG:
格雷格·盖奇： 所以现在 每次只要图像出现
So right now,every time the image comes on,
我们就会拍一张图像开始时的照片
we’re taking a pictureof the onset of the image
破译脑电图扫描器结果
and decoding the EEG.
对
It’s getting correct. Nathan:
南森：是的 面孔 对
Yes. Face. Correct. GG:
格雷格·盖奇：所以脑电图扫描器信号里包含一些信息 真酷
So there is information in the EEG signal, which is cool.
我们只需把它和图像开始时对齐即可
We just had to align it to the onset of the image. Nathan:
南森：风景 对 面孔
Scenery. Correct. Face.
是的
Yeah. GG:
格雷格·盖奇：这意味着这里包含了一些信息
This means there is someinformation there,
所以如果我们知道图像出现的时间
so if we know at what timethe picture came on,
我们就能判别图像的种类
we can tell what type of picture it was,
看到这些诱发的电位
possibly, at least on average,
也许 至少有一半的成功率
by looking at these evoked potentials. Nathan:
南森·完全正确
Exactly. GG:
格雷格·盖奇：如果一开始你就告诉我
If you had told me at the beginning
这个项目是有理论可能性的话
of this project this was possible,
我可能会说 骗人
I would have said no way.
我坚定地以为 我们不会成功
I literally did not thinkwe could do this.
我们的读心术实验起作用了吗？是的
Did our mind-readingexperiment really work? Yes,
但是我们必须大量地作弊
but we had to do a lot of cheating.
你会发现原来
It turns out you can find
脑电图扫描器里包含一些有趣的东西
some interesting things in the EEG,
比如 你是否正在看某个人的脸
for example if you’relooking at someone’s face,
当然会存在很多局限性
but it does have a lot of limitations.
也许机器学习的提升会使人类进步
Perhaps advances in machine learningwill make huge strides,
总有一天 我们能够
and one day we will be able to
破译我们到底在想什么
decode what’s going on in our thoughts.
但是 现在 如果下一次 一个公司声明
But for now, the next time a company says
他们利用你的脑波
that they can harness your brainwaves
能够控制机器 你有权利
to be able to control devices, it is your right,
有义务去质疑
it is your duty to be skeptical.