0:00:01.747,0:00:03.234
This is Henry,

0:00:03.258,0:00:05.064
a cute boy,

0:00:05.088,0:00:07.226
and when Henry was three,

0:00:07.250,0:00:12.185
his mom found him having[br]some febrile seizures.

0:00:13.080,0:00:18.040
Febrile seizures are seizures that occur[br]when you also have a fever,

0:00:18.064,0:00:19.962
and the doctor said,

0:00:19.986,0:00:22.728
"Don't worry too much.[br]Kids usually outgrow these."

0:00:23.506,0:00:27.487
When he was four,[br]he had a convulsive seizure,

0:00:27.511,0:00:30.773
the kind that you lose[br]consciousness and shake --

0:00:30.797,0:00:33.693
a generalized tonic-clonic seizure --

0:00:33.717,0:00:40.701
and while the diagnosis of epilepsy[br]was in the mail,

0:00:41.763,0:00:44.439
Henry's mom went to get him[br]out of bed one morning,

0:00:45.324,0:00:47.228
and as she went in his room,

0:00:48.753,0:00:53.443
she found his cold, lifeless body.

0:00:55.904,0:00:57.960
Henry died of SUDEP,

0:00:57.984,0:01:00.389
sudden unexpected death in epilepsy.

0:01:01.714,0:01:04.959
I'm curious how many of you[br]have heard of SUDEP.

0:01:05.603,0:01:09.969
This is a very well-educated audience,[br]and I see only a few hands.

0:01:09.993,0:01:13.959
SUDEP is when an otherwise[br]healthy person with epilepsy

0:01:13.983,0:01:17.873
dies and they can't attribute it[br]to anything they can find in an autopsy.

0:01:20.082,0:01:23.685
There is a SUDEP[br]every seven to nine minutes.

0:01:24.217,0:01:27.096
That's on average two per TED Talk.

0:01:31.411,0:01:35.070
Now, a normal brain[br]has electrical activity.

0:01:35.094,0:01:37.229
You can see some of the electrical waves

0:01:37.253,0:01:40.317
coming out of this picture[br]of a brain here.

0:01:40.341,0:01:44.087
And these should look[br]like typical electrical activity

0:01:44.111,0:01:46.246
that an EEG could read on the surface.

0:01:46.270,0:01:50.687
When you have a seizure,[br]it's a bit of unusual electrical activity,

0:01:50.711,0:01:52.096
and it can be focal.

0:01:52.120,0:01:54.620
It can take place[br]in just a small part of your brain.

0:01:54.644,0:01:58.452
When that happens,[br]you might have a strange sensation.

0:01:58.476,0:02:01.509
Several could be happening[br]here in the audience right now,

0:02:01.533,0:02:04.135
and the person next to you[br]might not even know.

0:02:04.159,0:02:08.079
However, if you have a seizure[br]where that little brush fire spreads

0:02:08.103,0:02:10.095
like a forest fire over the brain,

0:02:10.119,0:02:11.793
then it generalizes,

0:02:11.817,0:02:16.388
and that generalized seizure[br]takes your consciousness away

0:02:16.412,0:02:17.988
and causes you to convulse.

0:02:18.680,0:02:22.822
There are more SUDEPs[br]in the United States every year

0:02:22.846,0:02:25.505
than sudden infant death syndrome.

0:02:26.298,0:02:29.300
Now, how many of you have heard[br]of sudden infant death syndrome?

0:02:29.324,0:02:31.280
Right? Pretty much every hand goes up.

0:02:31.304,0:02:33.027
So what's going on here?

0:02:33.051,0:02:38.194
Why is this so much more common[br]and yet people haven't heard of it?

0:02:38.218,0:02:40.654
And what can you do to prevent it?

0:02:40.678,0:02:42.980
Well, there are two things,[br]scientifically shown,

0:02:43.004,0:02:46.377
that prevent or reduce the risk of SUDEP.

0:02:47.115,0:02:49.670
The first is: "Follow[br]your doctor's instructions,

0:02:49.694,0:02:51.130
take your medications."

0:02:51.154,0:02:53.355
Two-thirds of people who have epilepsy

0:02:53.379,0:02:55.498
get it under control[br]with their medications.

0:02:55.848,0:03:00.019
The second thing that reduces[br]the risk of SUDEP is companionship.

0:03:00.043,0:03:04.737
It's having somebody there[br]at the time that you have a seizure.

0:03:04.761,0:03:08.682
Now, SUDEP, even though[br]most of you have never heard of it,

0:03:08.706,0:03:13.895
is actually the number two cause[br]of years of potential life lost

0:03:13.919,0:03:16.895
of all neurological disorders.

0:03:17.197,0:03:21.483
The vertical axis is the number of deaths

0:03:21.507,0:03:24.673
times the remaining life span,

0:03:24.697,0:03:28.136
so higher is much worse impact.

0:03:28.522,0:03:31.496
SUDEP, however, unlike these others,

0:03:31.520,0:03:37.236
is something that people right here[br]could do something to push that down.

0:03:37.778,0:03:44.778
Now, what is Roz Picard, an AI researcher,[br]doing here telling you about SUDEP, right?

0:03:44.802,0:03:47.323
I'm not a neurologist.

0:03:47.347,0:03:51.562
When I was working at the Media Lab[br]on measurement of emotion,

0:03:51.586,0:03:54.808
trying to make our machines[br]more intelligent about our emotions,

0:03:54.832,0:03:57.363
we started doing a lot of work[br]measuring stress.

0:03:59.212,0:04:01.148
We built lots of sensors

0:04:01.172,0:04:03.553
that measured it[br]in lots of different ways.

0:04:03.577,0:04:05.897
But one of them in particular

0:04:05.921,0:04:10.033
grew out of some of this very old work[br]with measuring sweaty palms

0:04:10.057,0:04:11.801
with an electrical signal.

0:04:11.825,0:04:13.574
This is a signal of skin conductance

0:04:13.598,0:04:15.716
that's known to go up[br]when you get nervous,

0:04:15.740,0:04:19.312
but it turns out it also goes up with[br]a lot of other interesting conditions.

0:04:19.336,0:04:22.358
But measuring it with wires on your hand[br]is really inconvenient.

0:04:22.382,0:04:26.127
So we invented a bunch of other ways[br]of doing this at the MIT Media Lab.

0:04:26.151,0:04:28.324
And with these wearables,

0:04:28.348,0:04:33.153
we started to collect the first-ever[br]clinical quality data 24-7.

0:04:33.609,0:04:36.142
Here's a picture of what that looked like

0:04:36.166,0:04:42.761
the first time an MIT student collected[br]skin conductance on the wrist 24-7.

0:04:43.187,0:04:45.545
Let's zoom in a little bit here.

0:04:45.569,0:04:48.594
What you see is 24 hours[br]from left to right,

0:04:48.618,0:04:50.545
and here is two days of data.

0:04:50.569,0:04:53.251
And first, what surprised us

0:04:53.275,0:04:56.621
was sleep was the biggest[br]peak of the day.

0:04:56.645,0:04:58.395
Now, that sounds broken, right?

0:04:58.419,0:05:02.597
You're calm when you're asleep,[br]so what's going on here?

0:05:02.621,0:05:05.074
Well, it turns out[br]that our physiology during sleep

0:05:05.098,0:05:07.741
is very different[br]than our physiology during wake,

0:05:07.765,0:05:09.949
and while there's still a bit of a mystery

0:05:09.973,0:05:13.717
why these peaks are usually[br]the biggest of the day during sleep,

0:05:13.741,0:05:16.911
we now believe they're related[br]to memory consolidation

0:05:16.935,0:05:18.990
and memory formation during sleep.

0:05:19.895,0:05:23.466
We also saw things[br]that were exactly what we expected.

0:05:23.490,0:05:25.840
When an MIT student[br]is working hard in the lab

0:05:25.864,0:05:27.253
or on homeworks,

0:05:27.277,0:05:31.824
there is not only emotional stress,[br]but there's cognitive load,

0:05:31.848,0:05:36.709
and it turns out that cognitive load,[br]cognitive effort, mental engagement,

0:05:36.733,0:05:38.927
excitement about learning something --

0:05:38.951,0:05:41.288
those things also make the signal go up.

0:05:43.014,0:05:46.752
Unfortunately, to the embarrassment[br]of we MIT professors,

0:05:46.776,0:05:47.855
(Laughter)

0:05:47.879,0:05:52.347
the low point every day[br]is classroom activity.

0:05:52.810,0:05:55.326
Now, I am just showing you[br]one person's data here,

0:05:55.350,0:05:57.960
but this, unfortunately,[br]is true in general.

0:06:00.400,0:06:04.789
This sweatband has inside it[br]a homebuilt skin-conductance sensor,

0:06:04.813,0:06:09.899
and one day, one of our undergrads[br]knocked on my door

0:06:09.923,0:06:12.598
right at the end of the December semester,

0:06:12.622,0:06:14.812
and he said, "Professor Picard,

0:06:14.836,0:06:17.550
can I please borrow[br]one of your wristband sensors?

0:06:17.574,0:06:21.303
My little brother has autism,[br]he can't talk,

0:06:21.327,0:06:23.494
and I want to see[br]what's stressing him out."

0:06:24.434,0:06:27.292
And I said, "Sure, in fact,[br]don't just take one, take two,"

0:06:27.316,0:06:30.108
because they broke easily back then.

0:06:30.132,0:06:32.752
So he took them home,[br]he put them on his little brother.

0:06:32.776,0:06:35.945
Now, I was back in MIT,[br]looking at the data on my laptop,

0:06:35.969,0:06:38.830
and the first day, I thought,[br]"Hmm, that's odd,

0:06:38.854,0:06:41.845
he put them on both wrists[br]instead of waiting for one to break.

0:06:41.869,0:06:44.189
OK, fine, don't follow my instructions."

0:06:44.213,0:06:45.931
I'm glad he didn't.

0:06:46.377,0:06:49.583
Second day -- chill.[br]Looked like classroom activity.

0:06:49.607,0:06:50.869
(Laughter)

0:06:51.389,0:06:52.718
A few more days ahead.

0:06:53.404,0:06:57.918
The next day, one wrist signal was flat

0:06:57.942,0:07:01.900
and the other had[br]the biggest peak I've ever seen,

0:07:02.496,0:07:04.814
and I thought, "What's going on?

0:07:04.838,0:07:08.298
We've stressed people out at MIT[br]every way imaginable.

0:07:08.933,0:07:10.671
I've never seen a peak this big."

0:07:12.409,0:07:14.378
And it was only on one side.

0:07:14.402,0:07:17.546
How can you be stressed on one side[br]of your body and not the other?

0:07:17.570,0:07:20.440
So I thought one or both sensors[br]must be broken.

0:07:21.185,0:07:23.102
Now, I'm an electroengineer by training,

0:07:23.126,0:07:25.795
so I started a whole bunch of stuff[br]to try to debug this,

0:07:25.819,0:07:28.318
and long story short,[br]I could not reproduce this.

0:07:28.962,0:07:31.995
So I resorted to old-fashioned debugging.

0:07:32.019,0:07:35.301
I called the student at home on vacation.

0:07:35.325,0:07:39.668
"Hi, how's your little brother?[br]How's your Christmas?

0:07:39.692,0:07:42.973
Hey, do you have any idea[br]what happened to him?"

0:07:42.997,0:07:44.953
And I gave this particular date and time,

0:07:44.977,0:07:46.278
and the data.

0:07:46.302,0:07:49.717
And he said, "I don't know,[br]I'll check the diary."

0:07:50.850,0:07:53.758
Diary? An MIT student keeps a diary?

0:07:53.782,0:07:55.601
So I waited and he came back.

0:07:55.625,0:07:57.164
He had the exact date and time,

0:07:57.188,0:08:01.041
and he says, "That was right before[br]he had a grand mal seizure."

0:08:02.942,0:08:06.291
Now, at the time, I didn't know[br]anything about epilepsy,

0:08:06.315,0:08:08.474
and did a bunch of research,

0:08:08.498,0:08:11.624
realized that another student's dad[br]is chief of neurosurgery

0:08:11.648,0:08:13.648
at Children's Hospital Boston,

0:08:13.672,0:08:15.961
screwed up my courage[br]and called Dr. Joe Madsen.

0:08:15.985,0:08:18.454
"Hi, Dr. Madsen,[br]my name's Rosalind Picard.

0:08:18.478,0:08:22.374
Is it possible somebody could have

0:08:22.398,0:08:26.801
a huge sympathetic[br]nervous system surge" --

0:08:26.825,0:08:29.403
that's what drives the skin conductance --

0:08:29.427,0:08:31.099
"20 minutes before a seizure?"

0:08:32.172,0:08:34.291
And he says, "Probably not."

0:08:35.997,0:08:37.331
He says, "It's interesting.

0:08:37.355,0:08:40.123
We've had people whose hair[br]stands on end on one arm

0:08:40.147,0:08:42.227
20 minutes before a seizure."

0:08:43.060,0:08:44.375
And I'm like, "On one arm?"

0:08:44.399,0:08:46.529
I didn't want to tell him that, initially,

0:08:46.553,0:08:48.942
because I thought this was too ridiculous.

0:08:48.966,0:08:51.259
He explained how this could[br]happen in the brain,

0:08:51.283,0:08:53.456
and he got interested.[br]I showed him the data.

0:08:53.480,0:08:56.428
We made a whole bunch more devices,[br]got them safety certified.

0:08:56.452,0:08:58.526
90 families were being[br]enrolled in a study,

0:08:58.550,0:09:01.779
all with children who were going[br]to be monitored 24-7

0:09:01.803,0:09:05.174
with gold-standard EEG on their scalp

0:09:05.198,0:09:06.741
for reading the brain activity,

0:09:06.765,0:09:08.708
video to watch the behavior,

0:09:08.732,0:09:12.392
electrocardiogram -- ECG --[br]and now EDA, electrodermal activity,

0:09:12.416,0:09:15.001
to see if there was[br]something in this periphery

0:09:15.025,0:09:17.471
that we could easily pick up,[br]related to a seizure.

0:09:18.263,0:09:24.704
We found, in 100 percent[br]of the first batch of grand mal seizures,

0:09:24.728,0:09:28.361
this whopper of responses[br]in the skin conductance.

0:09:28.385,0:09:30.388
The blue in the middle, the boy's sleep,

0:09:30.412,0:09:32.384
is usually the biggest peak of the day.

0:09:32.408,0:09:36.040
These three seizures you see here[br]are popping out of the forest

0:09:36.064,0:09:37.540
like redwood trees.

0:09:39.096,0:09:42.676
Furthermore, when you couple[br]the skin conductance at the top

0:09:42.700,0:09:46.217
with the movement from the wrist

0:09:46.241,0:09:51.041
and you get lots of data[br]and train machine learning and AI on it,

0:09:51.065,0:09:55.858
you can build an automated AI[br]that detects these patterns

0:09:55.882,0:10:00.097
much better than just[br]a shake detector can do.

0:10:00.121,0:10:04.388
So we realized that we needed[br]to get this out,

0:10:04.412,0:10:06.770
and with the PhD work of Ming-Zher Poh

0:10:06.794,0:10:09.821
and later great improvements by Empatica,

0:10:09.845,0:10:13.689
this has made progress and the seizure[br]detection is much more accurate.

0:10:13.713,0:10:16.618
But we also learned some other things[br]about SUDEP during this.

0:10:16.642,0:10:20.054
One thing we learned is that SUDEP,

0:10:20.078,0:10:22.990
while it's rare after[br]a generalized tonic-clonic seizure,

0:10:23.014,0:10:26.251
that's when it's most likely[br]to happen -- after that type.

0:10:26.275,0:10:29.394
And when it happens,[br]it doesn't happen during the seizure,

0:10:29.418,0:10:32.284
and it doesn't usually happen[br]immediately afterwards,

0:10:32.308,0:10:34.023
but immediately afterwards,

0:10:34.047,0:10:37.109
when the person just seems[br]very still and quiet,

0:10:37.133,0:10:42.080
they may go into another phase,[br]where the breathing stops,

0:10:42.104,0:10:45.128
and then after the breathing stops,[br]later the heart stops.

0:10:45.152,0:10:47.195
So there's some time[br]to get somebody there.

0:10:48.349,0:10:53.253
We also learned that there is a region[br]deep in the brain called the amygdala,

0:10:53.277,0:10:56.213
which we had been studying[br]in our emotion research a lot.

0:10:56.237,0:10:57.518
We have two amygdalas,

0:10:57.542,0:10:59.215
and if you stimulate the right one,

0:10:59.239,0:11:01.978
you get a big right[br]skin conductance response.

0:11:02.002,0:11:06.247
Now, you have to sign up right now[br]for a craniotomy to get this done,

0:11:06.271,0:11:09.104
not exactly something[br]we're going to volunteer to do,

0:11:09.128,0:11:11.641
but it causes a big right skin[br]conductance response.

0:11:11.665,0:11:15.518
Stimulate the left one, big left[br]skin conductance response on the palm.

0:11:15.542,0:11:19.582
And furthermore, when somebody[br]stimulates your amygdala

0:11:19.606,0:11:23.025
while you're sitting there[br]and you might just be working,

0:11:23.049,0:11:24.937
you don't show any signs of distress,

0:11:25.620,0:11:26.906
but you stop breathing,

0:11:28.470,0:11:31.787
and you don't start again[br]until somebody stimulates you.

0:11:32.599,0:11:34.112
"Hey, Roz, are you there?"

0:11:34.136,0:11:36.067
And you open your mouth to talk.

0:11:36.868,0:11:39.062
As you take that breath to speak,

0:11:39.086,0:11:40.591
you start breathing again.

0:11:42.762,0:11:46.382
So we had started with work on stress,

0:11:46.406,0:11:48.659
which had enabled us[br]to build lots of sensors

0:11:48.683,0:11:50.815
that were gathering[br]high quality enough data

0:11:50.839,0:11:53.760
that we could leave the lab[br]and start to get this in the wild;

0:11:53.784,0:11:56.648
accidentally found a whopper[br]of a response with the seizure,

0:11:56.672,0:11:59.531
neurological activation that can cause[br]a much bigger response

0:11:59.555,0:12:00.865
than traditional stressors;

0:12:00.889,0:12:04.087
lots of partnership with hospitals[br]and an epilepsy monitoring unit,

0:12:04.111,0:12:05.939
especially Children's Hospital Boston

0:12:05.963,0:12:07.114
and the Brigham;

0:12:07.138,0:12:09.663
and machine learning and AI on top of this

0:12:09.687,0:12:12.708
to take and collect lots more data

0:12:12.732,0:12:15.361
in service of trying[br]to understand these events

0:12:15.385,0:12:16.979
and if we could prevent SUDEP.

0:12:18.023,0:12:21.586
This is now commercialized by Empatica,

0:12:21.610,0:12:24.499
a start-up that I had[br]the privilege to cofound,

0:12:24.523,0:12:28.729
and the team there has done an amazing job[br]improving the technology

0:12:28.753,0:12:30.512
to make a very beautiful sensor

0:12:30.536,0:12:34.395
that not only tells time and does steps[br]and sleep and all that good stuff,

0:12:34.419,0:12:37.515
but this is running real-time[br]AI and machine learning

0:12:37.539,0:12:40.157
to detect generalized[br]tonic-clonic seizures

0:12:40.181,0:12:42.332
and send an alert for help

0:12:42.356,0:12:45.548
if I were to have a seizure[br]and lose consciousness.

0:12:45.572,0:12:48.388
This just got FDA-approved

0:12:48.412,0:12:52.552
as the first smartwatch[br]to get approved in neurology.

0:12:53.923,0:13:00.923
(Applause)

0:13:02.674,0:13:06.205
Now, the next slide is what made[br]my skin conductance go up.

0:13:06.811,0:13:08.517
One morning, I'm checking my email

0:13:08.541,0:13:10.838
and I see a story from a mom

0:13:10.862,0:13:12.748
who said she was in the shower,

0:13:12.772,0:13:15.225
and her phone was[br]on the counter by the shower,

0:13:15.249,0:13:17.406
and it said her daughter[br]might need her help.

0:13:18.196,0:13:21.577
So she interrupts her shower and goes[br]running to her daughter's bedroom,

0:13:21.601,0:13:25.344
and she finds her daughter[br]facedown in bed, blue and not breathing.

0:13:25.368,0:13:28.649
She flips her over -- human stimulation --

0:13:28.673,0:13:32.356
and her daughter takes a breath,[br]and another breath,

0:13:32.380,0:13:36.334
and her daughter turns pink and is fine.

0:13:37.551,0:13:40.775
I think I turned white reading this email.

0:13:40.799,0:13:43.121
My first response is,[br]"Oh no, it's not perfect.

0:13:43.145,0:13:45.470
The Bluetooth could break,[br]the battery could die.

0:13:45.494,0:13:47.971
All these things could go wrong.[br]Don't rely on this."

0:13:47.995,0:13:51.403
And she said, "It's OK.[br]I know no technology is perfect.

0:13:51.427,0:13:53.590
None of us can always[br]be there all the time.

0:13:54.844,0:13:59.186
But this, this device plus AI

0:13:59.210,0:14:02.314
enabled me to get there in time[br]to save my daughter's life."

0:14:05.933,0:14:07.839
Now, I've been mentioning children,

0:14:07.863,0:14:13.570
but SUDEP peaks, actually,[br]among people in their 20s, 30s and 40s,

0:14:13.594,0:14:15.410
and the next line I'm going to put up

0:14:15.434,0:14:17.887
is probably going to make[br]some people uncomfortable,

0:14:17.911,0:14:20.285
but it's less uncomfortable[br]than we'll all be

0:14:20.309,0:14:23.379
if this list is extended[br]to somebody you know.

0:14:24.437,0:14:26.775
Could this happen to somebody you know?

0:14:27.197,0:14:29.939
And the reason I bring up[br]this uncomfortable question

0:14:29.963,0:14:34.875
is because one in 26 of you[br]will have epilepsy at some point,

0:14:34.899,0:14:37.134
and from what I've been learning,

0:14:37.158,0:14:40.492
people with epilepsy often don't tell[br]their friends and their neighbors

0:14:40.516,0:14:41.674
that they have it.

0:14:41.698,0:14:46.619
So if you're willing to let them[br]use an AI or whatever

0:14:46.643,0:14:51.057
to summon you in a moment[br]of possible need,

0:14:51.081,0:14:52.699
if you would let them know that,

0:14:52.723,0:14:55.077
you could make a difference in their life.

0:14:55.728,0:14:58.579
Why do all this hard work to build AIs?

0:15:00.101,0:15:01.302
A couple of reasons here:

0:15:01.326,0:15:03.489
one is Natasha, the girl who lived,

0:15:03.513,0:15:05.905
and her family wanted me[br]to tell you her name.

0:15:06.977,0:15:08.589
Another is her family

0:15:08.613,0:15:10.804
and the wonderful people out there

0:15:10.828,0:15:13.653
who want to be there to support people[br]who have conditions

0:15:13.677,0:15:16.970
that they've felt uncomfortable[br]in the past mentioning to others.

0:15:17.901,0:15:19.774
And the other reason is all of you,

0:15:19.798,0:15:25.130
because we have the opportunity[br]to shape the future of AI.

0:15:25.154,0:15:27.511
We can actually change it,

0:15:27.535,0:15:29.939
because we are the ones building it.

0:15:29.963,0:15:31.940
So let's build AI

0:15:31.964,0:15:35.256
that makes everybody's lives better.

0:15:35.982,0:15:37.133
Thank you.

0:15:37.157,0:15:41.571
(Applause)