TED Talks · Neuroscience

The computer and your mind

You've seen this in sci-fi movies, Machines in the future that can read our thoughts.

Level: B2 · 145 sentences

Transcript

  1. You've seen this in sci-fi movies,
  2. Machines in the future that can read our thoughts.
  3. However, there are devices today
  4. That can read the electrical activity from our brains.
  5. We call this the EEG.
  6. Is there information contained in these brain waves?
  7. And if so, could we train a computer to read our thoughts?
  8. My buddy Nathan has been working to hack the EEG
  9. To build a mind-reading machine.
  10. So this is how the EEG works.
  11. Inside your head is a brain.
  12. And that brain is made out of billions of neurons.
  13. Each of those neurons send an electrical message to each other.
  14. Now, these small messages can combine to make an electrical wave
  15. That we can detect on a monitor.
  16. Now, traditionally, the EEG can tell us large-scale things, for example,
  17. If you're asleep or if you're alert.
  18. But can it tell us anything else?
  19. Can it actually read our thoughts?
  20. Well, we're going to test this.
  21. And we're not going to start with some complex thoughts.
  22. We're going to do something very simple.
  23. Can we interpret what someone is seeing using only their brain waves?
  24. Nathan's going to begin by placing electrodes on Christie's head.
  25. My life is tangled.
  26. And then he's going to show her
  27. A bunch of pictures from four different categories.
  28. Face, house, scenery, and weird pictures.
  29. As we show Christie hundreds of these images,
  30. We are also capturing the electrical waves onto Nathan's computer.
  31. We want to see if we can detect
  32. Any visual information about the photos contained in the brain waves.
  33. So when we're done, we're going to see
  34. If the EEG can tell us what kind of picture Christie is looking at.
  35. And if it does, each category should trigger a different brain signal.
  36. Okay, so we collected all the raw EEG data,
  37. And this is what we got.
  38. It all looks pretty messy, so let's arrange them by picture.
  39. Now it's still a bit too noisy to see any differences,
  40. But if we average the EEG across
  41. All image types by aligning them to when the image first appeared,
  42. We can remove this noise.
  43. And pretty soon we can see
  44. Some dominant patterns emerge for each category.
  45. Now the signals all still look pretty similar.
  46. Let's take a closer look.
  47. About 100 milliseconds after the image comes on,
  48. We see a positive bump in all four cases.
  49. And we call this the P100.
  50. And what we think that is is what happens in your brain
  51. When you recognize an object.
  52. But damn, look at that signal for the face.
  53. It looks different than the others.
  54. There's a negative dip about 170 milliseconds after the image comes on.
  55. What could be going on here?
  56. Research shows that our brain has a lot of neurons
  57. That are dedicated to recognizing human faces.
  58. So this N170 spike could be all those neurons
  59. Firing at once in the same location.
  60. We could detect that in the EEG.
  61. So there are two takeaways here.
  62. One, our eyes can't really detect the differences in patterns
  63. Without averaging out the noise.
  64. And two, even after removing the noise,
  65. Our eyes can only pick up the signals associated with faces.
  66. So this is where we turn to machine learning.
  67. Now our eyes are not very good
  68. At picking up patterns and noisy data.
  69. But machine learning algorithms are designed to do just that.
  70. So can we take a lot of pictures
  71. And a lot of data
  72. And feed it in and train a computer
  73. To be able to interpret what Christie's looking at in real time?
  74. We're trying to decode the information
  75. That's coming out of her EEG in real time
  76. And predict what it is that her eyes are looking at.
  77. And if it works, what we should see is every time
  78. That she gets a picture of scenery,
  79. It should say scenery, scenery, scenery, scenery.
  80. A face, face, face, face, face.
  81. It's not quite working that way is what we're discovering.
  82. OK.
  83. So what's going on here?
  84. We need a new career, I think.
  85. OK.
  86. So that was a massive failure.
  87. But we're still curious.
  88. How far can we push this technology?
  89. Now, we look back at what we did.
  90. We noticed that the data was coming into our computer very quickly
  91. Without any timing of when the images came on.
  92. And that's the equivalent of reading a very long sentence
  93. Without spaces between the words.
  94. Something like that would be hard to read,
  95. But once we add the spaces, individual words appear,
  96. And it becomes a lot more understandable.
  97. But what if we cheat a little bit?
  98. By using a sensor,
  99. We can tell the computer when the image first appears.
  100. That way, the brainwaves stop being a continuous stream of information
  101. And instead become individual packets of meaning.
  102. Also, we're going to cheat a little bit more
  103. By limiting the categories to two.
  104. Let's see if we can do some real-time mind reading.
  105. In this new experiment,
  106. We're going to sort of constrict it a little bit more
  107. So that we know the onset of the image,
  108. And we're going to limit the categories to face or scenery.
  109. Face.
  110. Correct.
  111. Scenery.
  112. Correct.
  113. So right now, every time the image comes on,
  114. We're taking a picture of the onset of the image
  115. And decoding the EEG.
  116. It's getting correct.
  117. Yes.
  118. Face.
  119. So there is information in the EEG signal, which is cool.
  120. We just had to align it to the onset of the image.
  121. Scenery, correct.
  122. Face, yeah.
  123. What this means is that there is some information there.
  124. So we can tell,
  125. If we know at what time the picture came on,
  126. We can tell what type of picture it was, possibly,
  127. At least on average,
  128. By looking at these evoked potentials.
  129. Exactly.
  130. Now, if you would have told me at the beginning
  131. Of this project this was possible, I would have said, no way.
  132. I literally did not think you could do this.
  133. Did our mind reading experiment really work?
  134. Yes, but we had to do a lot of cheating.
  135. Now it turns out you can find
  136. Some interesting things in the EEG.
  137. For example, if you're looking at someone's face.
  138. But it does have a lot of limitations.
  139. Perhaps advances in machine learning will make huge strides
  140. And one day we'll be able to decode
  141. What's going on in our thoughts.
  142. But for now, the next time a company says
  143. That they can harness your brainwaves to be able to control devices,
  144. It is your right,
  145. It is your duty to be skeptical.