Brain-IT: AI That Sees What You See
Welcome back to Learn English with Podcasts! Sarah, have you ever had a dream so strange that you wished you could show it to someone?欢迎回到 Learn English with Podcasts！Sarah，你有没有做过一个很奇怪的梦，希望能放给别人看？
All the time. Last week I dreamed my cat was giving a speech. I woke up and thought, if only I had a video of that.经常。上周我梦见我的猫在演讲。我醒来就在想，要是有个视频就好了。
Well, science just moved one step closer to that. A team in Israel built an AI model called Brain-IT. It can look at your brain activity and rebuild the picture you are seeing.科学刚向这个方向近了一步。以色列的一个团队做出一个叫 Brain-IT 的 AI 模型。它能看你的大脑活动，还原你正在看的画面。
Wait, really? It reads my mind through a brain scan? How does that even work?等等，真的吗？它通过脑扫描读我的心？这是怎么做到的？
The team is led by Michal Irani. She works at the Weizmann Institute of Science. Her team uses a machine called fMRI. It watches blood flow in your brain while you look at pictures.团队由 Michal Irani 带领。她在魏茨曼科学研究所工作。她的团队用一种叫 fMRI 的机器。你看图片时，它观察你大脑里的血流变化。
So different pictures make different parts of the brain light up? And the AI learns that pattern?所以不同的图片会让大脑不同部位亮起来？AI 就学习这个规律？
Exactly. To train the AI, they need pairs of data. One picture, plus the brain scan from the person looking at it. Thousands and thousands of such pairs.完全对。要训练 AI，他们需要成对的数据。一张图片，加上看图那个人的一份脑扫描。几千几万组这样的数据。
That sounds like a lot of work. How many people joined this study?这听起来工作量很大。有多少人参加了这项研究？
Only eight people. They used a public set called the Natural Scenes Dataset. It is the biggest set of its kind in the world. Each person went into the scanner 30 to 40 times. Each time they looked at thousands of pictures. In the end, the team got about 73,000 pairs.只有八个人。他们用了一个公开数据集，叫自然场景数据集。它是全世界同类中最大的。每个人进了扫描仪30到40次，每次看几千张图片。最后团队拿到约73,000组配对数据。
Eight people, 30 to 40 scans each? That must have taken forever. Lying still in a noisy machine and just looking at pictures?八个人，每人扫描30到40次？那得多花时间啊。躺在又吵的机器里一动不动，光看图片？
Yes, it is slow and tiring. That is the big problem. Good brain data is very hard to collect. So the team found a smart trick. They built a second model that works backwards.对，又慢又累。这是最大的难题。好的脑数据太难收集了。所以团队想了个聪明的办法。他们又建了一个反向工作的模型。
Backwards? You mean it starts from a picture and guesses the brain activity?反向？意思是它从图片出发，去猜大脑活动？
Right. Give it a normal photo, and it predicts what your brain would look like when you see it. In this way, they can make much more training data without putting more people into the scanner.对。给它一张普通照片，它就预测你看到它时大脑会是什么样。这样不用再让人进扫描仪，也能造出多得多的训练数据。
Oh, like making practice questions before a test. Smart. But does the new model really work better?哦，就像考试前自己出练习题。真聪明。但新模型真的更好用吗？
Irani says yes, in two ways. Older models could rebuild the main idea, but they often got the layout or the colors wrong. A red car could come back blue, or on the wrong side of the street. Brain-IT keeps the content and the small details much better.Irani 说是的，有两方面。老模型能还原大意，但经常搞错构图或颜色。红车可能变蓝车，或者跑到马路错的一边。Brain-IT 把内容和小细节都还原得好得多。
Colors and layout matter a lot. If my cat gives a speech, I want the cat, not a dog. And what is the second improvement?颜色和构图很重要。如果是我的猫演讲，我要看到猫，不是狗。第二个改进是什么？
Speed with new people. Older models needed tens of hours of scan data to learn how to read a new person. Every brain is a little different. Brain-IT needs only about one hour.对新人的适应速度。老模型要读懂一个新人，需要几十小时的扫描数据。每个人大脑都有点不一样。Brain-IT 只需要大约一小时。
Wow, from tens of hours to one hour. That is a huge drop. So what is next? Videos? Dreams?哇，从几十小时降到一小时，降太多了。那下一步呢？视频？梦境？
That is exactly their plan. Right now Brain-IT only works with still pictures. Irani wants to move to sounds first, and then to video. The dream part is the hardest and the most exciting.这正是他们的计划。现在 Brain-IT 只会处理静态图片。Irani 想先做声音，再做视频。梦境部分最难，也最让人激动。
I can see why video is hard. A video changes many times each second. But I read that one fMRI scan takes about two seconds. The machine is just too slow, right?我能想到视频为什么难。视频每秒变很多次。但我看到 fMRI 扫一次要大约两秒。机器太慢了，对吧？
You got it. The pictures in a video move fast, but the scanner moves slowly. It is like trying to film a race with a camera that takes one photo every two seconds. You miss most of the action. If they fix this, Irani says, maybe one day we can read dreams.你说中了。视频画面动得快，扫描仪动得慢。就像用一个两秒拍一张的相机去拍赛跑，大部分动作都错过了。如果能解决这个问题，Irani 说，也许有一天我们真能读梦。
Reading dreams sounds like science fiction. But is there a warm side to this? Can it help sick people?读梦听起来像科幻。但它有温暖的一面吗？能帮病人吗？
Yes. Kenneth Norman is a professor at Princeton University. He studies the brain too. Back in 2022, he said this kind of tool could help people with depression or worry. A therapist could show patients when their thoughts turn dark, and help them turn back.有。Kenneth Norman 是普林斯顿大学的教授，也研究大脑。早在2022年他就说，这类工具能帮抑郁或焦虑的人。治疗师可以让病人看到自己的想法何时变消极，再帮他们转回来。
That is beautiful. Like a mirror for your mind. You see the sad loop starting, and you learn to stop it. Some doctors also hope it can help people with brain illness, right?这很美好。就像心灵的镜子。你看到难过的循环开始了，就学会停下来。一些医生也希望它能帮脑部疾病患者，对吧？
Yes, some brain scientists praise Brain-IT for that reason. But others feel nervous. The same tool that helps a patient could also be used to take thoughts by force. That is the fear.对，一些脑科学家为此称赞 Brain-IT。但也有人紧张。能帮病人的同一个工具，也可能被用来强行拿走想法。这就是担心。
Force? You mean in a court? Like testing if someone is lying?强行？指在法庭上吗？比如测有没有人撒谎？
Exactly. Back in 2017, a paper in the Journal of Law and the Biosciences talked about mind reading with brain scans. It said courts could one day use it to check defendants, prisoners, or even jury members. It also warned it could be used in unfair and forced ways.正是。早在2017年，《法律与生物科学杂志》上的一篇论文就谈过用脑扫描读心。论文说，法庭未来可能用它来检查被告、囚犯，甚至陪审员。但也警告，它可能被用在不公平、强制的用途上。
That gives me chills. My brain talks all day. Most of it is shopping lists and silly songs. Nobody needs to see that without my permission.我听得发冷。我的大脑整天碎碎念，大多是购物单和傻歌。没我允许，谁也不该看。
And that is why scientists are split. One side sees healing. The other side sees spying. The model itself is amazing, but the rules about how we use it matter even more.所以科学家分成两派。一边看到治病，一边看到监视。模型本身很厉害，但怎么用它的规则更重要。
You know what I keep thinking about? My blank Monday-morning brain. If you scanned me at 7 a.m., the AI would rebuild a gray cloud and a cup of coffee. Mystery solved.你知道我一直在想什么吗？我周一早上放空的大脑。如果你早上7点扫我，AI 只能还原出一团灰云加一杯咖啡。谜底揭晓。
And mine would show pizza. Even during a science test. So friends, if this machine ever comes to your city, what is the first picture you would want it to see from your mind? Tell us, and see you next time!而我只会是披萨。就算在考科学时也是。所以朋友们，如果这台机器来到你的城市，你最想让它看到你脑中的哪张画面？告诉我们，下期见！