ChatGPT背后的技术如何使读心术成为现实?

Neuroscientists at the University of Texas say they figured out a way to read minds by translating scans of brain activity into actual words using the very same artificial intelligence technology that powers ChatGPT.

德克萨斯大学神经学家称，他们找到了一种通过使用与支持ChatGPT的人工智能技术完全相同的技术，将大脑活动扫描结果翻译成实际文字的方法。

Now, does this mean that we could possibly communicate with someone who's unable or even unwilling?

那么，这是否意味着我们可以与那些不能说话甚至不愿说话的人交流呢?

Pop quiz, hotshot. How do brains say, hi, brainwaves.

突击测验，高手。大脑怎么说，嗨，脑电波。

Our Donie O'Sullivan has more.

本台记者多尼·奥沙利文将带来详细报道。

You're reading people's minds?

你在读心吗？

So we don't like to use the term mind reading.

我们不喜欢用读心这个词。

These neuroscientists at the University of Texas in Austin say they've made a major breakthrough.

德克萨斯大学奥斯汀分校的神经学家表示，他们已经取得了重大突破。

They've figured out how to translate brain activity into words using artificial intelligence.

他们已经找到了如何利用人工智能将大脑活动翻译成文字的方法。

These are different images.

这些是不同的图像。

Earlier this month, they published a paper explaining how they had research.

本月早些时候，他们发表了一篇解释他们是如何进行研究的论文。

Volunteers listened to audio clips while having their brains scanned by an fMRI machine.

志愿者听音频片段时，用功能磁共振成像仪对他们的大脑进行扫描。

Over time, AI algorithms, the very same tech that's behind ChatGPT, were able to figure out what the volunteers were listening to just by watching their brains.

随着时间的推移，人工智能算法，也就是ChatGPT背后的技术，能够通过观察志愿者的大脑来推断他们听到的内容。

It is just crazy. You can watch how blood flows through the brain. And using AI and GPT and everything else translated into words?

这太疯狂了。你可以看到血液是如何流经大脑的，以及如何利用人工智能、GPT和其他工具将其翻译成文字?

Yeah, it's wild that this works when you put it that way.

是啊，当你这么说时，这凑效了，太棒了。

Thumbs up, Donie.

太赞了，唐尼。

To test it all out, Professor Alexander Huth and I had our brain scanned while listening to parts of The Wizard of Oz audiobook.

为了验证，我和亚历山大·胡特教授在听《绿野仙踪》有声读物的部分内容时，对我们自己的大脑进行了扫描。

Hi, Donie, we have a picture of your brain.

唐尼，我们有一张你大脑的照片。

I have a brain?

我的大脑照片？

Yes. It looks good.

是的。看起来不错。

I was scanned first, followed by Professor Huth, capturing images of the changes in our brain's blood flow as we listened to the words from the audiobook and showing how our brains interpreted those words.

先扫描我的大脑，然后是胡特教授的，采集我们听有声书文字时大脑血流变化的图像，展示我们的大脑是如何理解这些文字的。

When she had finished her meal and was about to go back to the road of yellow brick, she was startled to hear a deep groan nearby.

当她吃完了东西，正要跑回到黄砖铺砌的路上去时，听得近旁有一声深长的呻吟声，给吓得跳了起来。

You can see that they're getting recordings every two seconds while he's listening to a story.

你可以看到，当他在听故事时，每两秒钟就会收到一次记录。

We will feed this data through our decoder and try to predict the story that he's currently listening to.

我们会先把这些数据输入解码器，再试着预测他正在听的故事。

The next morning, the results were in.

第二天早上，结果出来了。

OK, so it's been 24 hours since we got our brain scanned. You can confirm I have a brain?

好吧，自我们的大脑被扫描已经过去24小时了。你确定有我的大脑扫描结果吗?

Absolutely.

当然。

Brilliant.

太棒了。

So we were able to decode some stuff from my brain, not so much from yours.

我们能破译我大脑里的一些东西，你的能破译的就没那么多了。

So this is one from my brain. This is from The Wizard of Oz. So on the left side is the actual words that I heard.

这是我的大脑扫描结果。这是《绿野仙踪》的文字。左边是我听到的实际文字。

When she had finished her meal and was about to go back to the road of yellow brick, she was startled to hear a deep groan nearby.

当她吃完了东西，正要跑回到黄砖铺砌的路上去时，听得近旁有一声深长的呻吟声，给吓得跳了起来。

And the decoded version of this is on the right.

解码后的版本在右边。

I was about to head back to school and I hear this strange voice calling out to me.

我正要回学校，突然听到一个奇怪的声音朝我呼喊。

So it gets some things right. So this like, was about to go back, was about to head back.

因此，它解释对了一些文字。比如，正要回去、正要返回。

It completely misses some things like the road of yellow brick versus school, but then it gets this nice kind of example.

但它也完全理解错了一些文字，比如，黄砖路和学校，但它是个很好的例子。

So she hears something and then instead of a deep groan nearby, it said a strange voice calling out to me.

它说她听到了一个奇怪的声音朝我呼喊，而不是近旁深长的呻吟声。

That means something related, even if it's not exactly the right words.

这意味着这些文字相关，即使不是完全正确的文字。