This computer is learning to read your mind
-
0:00 - 0:03Greg Gage: Mind-reading.
You've seen this in sci-fi movies: -
0:03 - 0:05machines that can read our thoughts.
-
0:05 - 0:07However, there are devices today
-
0:07 - 0:09that can read the electrical
activity from our brains. -
0:09 - 0:11We call this the EEG.
-
0:12 - 0:15Is there information
contained in these brainwaves? -
0:15 - 0:17And if so, could we train a computer
to read our thoughts? -
0:17 - 0:20My buddy Nathan
has been working to hack the EEG -
0:20 - 0:22to build a mind-reading machine.
-
0:22 - 0:24[DIY Neuroscience]
-
0:25 - 0:26So this is how the EEG works.
-
0:27 - 0:28Inside your head is a brain,
-
0:28 - 0:31and that brain is made
out of billions of neurons. -
0:31 - 0:34Each of those neurons sends
an electrical message to each other. -
0:34 - 0:37These small messages can combine
to make an electrical wave -
0:37 - 0:38that we can detect on a monitor.
-
0:38 - 0:41Now traditionally, the EEG
can tell us large-scale things, -
0:41 - 0:44for example if you're asleep
or if you're alert. -
0:44 - 0:45But can it tell us anything else?
-
0:45 - 0:47Can it actually read our thoughts?
-
0:47 - 0:48We're going to test this,
-
0:48 - 0:51and we're not going to start
with some complex thoughts. -
0:51 - 0:53We're going to do something very simple.
-
0:53 - 0:56Can we interpret what someone is seeing
using only their brainwaves? -
0:56 - 0:59Nathan's going to begin by placing
electrodes on Christy's head. -
0:59 - 1:01Nathan: My life is tangled.
-
1:01 - 1:02(Laughter)
-
1:02 - 1:05GG: And then he's going to show her
a bunch of pictures -
1:05 - 1:06from four different categories.
-
1:06 - 1:09Nathan: Face, house, scenery
and weird pictures. -
1:09 - 1:11GG: As we show Christy
hundreds of these images, -
1:12 - 1:15we are also capturing the electrical waves
onto Nathan's computer. -
1:15 - 1:18We want to see if we can detect
any visual information about the photos -
1:18 - 1:20contained in the brainwaves,
-
1:20 - 1:22so when we're done,
we're going to see if the EEG -
1:22 - 1:25can tell us what kind of picture
Christy is looking at, -
1:25 - 1:28and if it does, each category
should trigger a different brain signal. -
1:28 - 1:31OK, so we collected all the raw EEG data,
-
1:31 - 1:32and this is what we got.
-
1:33 - 1:36It all looks pretty messy,
so let's arrange them by picture. -
1:37 - 1:39Now, still a bit too noisy
to see any differences, -
1:40 - 1:43but if we average the EEG
across all image types -
1:43 - 1:45by aligning them
to when the image first appeared, -
1:45 - 1:47we can remove this noise,
-
1:47 - 1:49and pretty soon, we can see
some dominant patterns -
1:49 - 1:51emerge for each category.
-
1:51 - 1:53Now the signals all
still look pretty similar. -
1:53 - 1:54Let's take a closer look.
-
1:54 - 1:57About a hundred milliseconds
after the image comes on, -
1:57 - 1:59we see a positive bump in all four cases,
-
1:59 - 2:02and we call this the P100,
and what we think that is -
2:02 - 2:05is what happens in your brain
when you recognize an object. -
2:05 - 2:07But damn, look at
that signal for the face. -
2:07 - 2:09It looks different than the others.
-
2:09 - 2:12There's a negative dip
about 170 milliseconds -
2:12 - 2:13after the image comes on.
-
2:13 - 2:15What could be going on here?
-
2:15 - 2:18Research shows that our brain
has a lot of neurons that are dedicated -
2:19 - 2:20to recognizing human faces,
-
2:20 - 2:23so this N170 spike could be
all those neurons -
2:23 - 2:25firing at once in the same location,
-
2:25 - 2:27and we can detect that in the EEG.
-
2:27 - 2:29So there are two takeaways here.
-
2:29 - 2:32One, our eyes can't really detect
the differences in patterns -
2:32 - 2:34without averaging out the noise,
-
2:34 - 2:36and two, even after removing the noise,
-
2:36 - 2:39our eyes can only pick up
the signals associated with faces. -
2:39 - 2:41So this is where we turn
to machine learning. -
2:41 - 2:45Now, our eyes are not very good
at picking up patterns in noisy data, -
2:45 - 2:48but machine learning algorithms
are designed to do just that, -
2:48 - 2:51so could we take a lot of pictures
and a lot of data -
2:51 - 2:53and feed it in and train a computer
-
2:53 - 2:57to be able to interpret
what Christy is looking at in real time? -
2:57 - 3:01We're trying to code the information
that's coming out of her EEG -
3:01 - 3:02in real time
-
3:02 - 3:05and predict what it is
that her eyes are looking at. -
3:05 - 3:07And if it works, what we should see
-
3:07 - 3:09is every time that she gets
a picture of scenery, -
3:09 - 3:11it should say scenery,
scenery, scenery, scenery. -
3:11 - 3:13A face -- face, face, face, face,
-
3:13 - 3:17but it's not quite working that way,
is what we're discovering. -
3:21 - 3:25(Laughter)
-
3:25 - 3:26OK.
-
3:26 - 3:30Director: So what's going on here?
GG: We need a new career, I think. -
3:30 - 3:31(Laughter)
-
3:31 - 3:33OK, so that was a massive failure.
-
3:33 - 3:36But we're still curious:
How far could we push this technology? -
3:36 - 3:38And we looked back at what we did.
-
3:38 - 3:41We noticed that the data was coming
into our computer very quickly, -
3:41 - 3:43without any timing
of when the images came on, -
3:43 - 3:46and that's the equivalent
of reading a very long sentence -
3:46 - 3:48without spaces between the words.
-
3:48 - 3:49It would be hard to read,
-
3:49 - 3:53but once we add the spaces,
individual words appear -
3:53 - 3:55and it becomes a lot more understandable.
-
3:55 - 3:57But what if we cheat a little bit?
-
3:57 - 4:01By using a sensor, we can tell
the computer when the image first appears. -
4:01 - 4:04That way, the brainwave stops being
a continuous stream of information, -
4:04 - 4:07and instead becomes
individual packets of meaning. -
4:07 - 4:09Also, we're going
to cheat a little bit more, -
4:09 - 4:11by limiting the categories to two.
-
4:11 - 4:14Let's see if we can do
some real-time mind-reading. -
4:14 - 4:15In this new experiment,
-
4:15 - 4:17we're going to constrict it
a little bit more -
4:17 - 4:19so that we know the onset of the image
-
4:19 - 4:23and we're going to limit
the categories to "face" or "scenery." -
4:23 - 4:25Nathan: Face. Correct.
-
4:26 - 4:27Scenery. Correct.
-
4:28 - 4:31GG: So right now,
every time the image comes on, -
4:31 - 4:33we're taking a picture
of the onset of the image -
4:33 - 4:35and decoding the EEG.
-
4:35 - 4:36It's getting correct.
-
4:36 - 4:38Nathan: Yes. Face. Correct.
-
4:38 - 4:40GG: So there is information
in the EEG signal, which is cool. -
4:40 - 4:43We just had to align it
to the onset of the image. -
4:43 - 4:45Nathan: Scenery. Correct.
-
4:47 - 4:48Face. Yeah.
-
4:49 - 4:51GG: This means there is some
information there, -
4:51 - 4:54so if we know at what time
the picture came on, -
4:54 - 4:56we can tell what type of picture it was,
-
4:56 - 5:01possibly, at least on average,
by looking at these evoked potentials. -
5:01 - 5:02Nathan: Exactly.
-
5:02 - 5:06GG: If you had told me at the beginning
of this project this was possible, -
5:06 - 5:07I would have said no way.
-
5:07 - 5:09I literally did not think
we could do this. -
5:09 - 5:11Did our mind-reading
experiment really work? -
5:11 - 5:13Yes, but we had to do a lot of cheating.
-
5:13 - 5:16It turns out you can find
some interesting things in the EEG, -
5:16 - 5:18for example if you're
looking at someone's face, -
5:18 - 5:21but it does have a lot of limitations.
-
5:21 - 5:24Perhaps advances in machine learning
will make huge strides, -
5:24 - 5:27and one day we will be able to decode
what's going on in our thoughts. -
5:27 - 5:31But for now, the next time a company says
that they can harness your brainwaves -
5:31 - 5:33to be able to control devices,
-
5:33 - 5:36it is your right, it is your duty
to be skeptical.
- Title:
- This computer is learning to read your mind
- Speaker:
- DIY Neuroscience
- Description:
-
Modern technology lets neuroscientists peer into the human brain, but can it also read minds? Armed with the device known as a electroencephalogram, or EEG, and some computing wizardry, our intrepid neuroscientists attempt to peer into a subject's thoughts.
- Video Language:
- English
- Team:
- closed TED
- Project:
- TED Series
- Duration:
- 05:51
Brian Greene edited English subtitles for This computer is learning to read your mind | ||
Brian Greene edited English subtitles for This computer is learning to read your mind | ||
Brian Greene approved English subtitles for This computer is learning to read your mind | ||
Brian Greene accepted English subtitles for This computer is learning to read your mind | ||
Brian Greene edited English subtitles for This computer is learning to read your mind | ||
Brian Greene edited English subtitles for This computer is learning to read your mind |