Can AI Catch What Doctors Miss? | Eric Topol | TED

159,019 views ใƒป 2023-12-09

TED


ืื ื ืœื—ืฅ ืคืขืžื™ื™ื ืขืœ ื”ื›ืชื•ื‘ื™ื•ืช ื‘ืื ื’ืœื™ืช ืœืžื˜ื” ื›ื“ื™ ืœื”ืคืขื™ืœ ืืช ื”ืกืจื˜ื•ืŸ.

ืชืจื’ื•ื: zeeva livshitz ืขืจื™ื›ื”: aknv tso
00:05
I've had the real fortune of working at Scripps Research
0
5961
3336
ื ืคืœื” ื‘ื—ืœืงื™ ื”ื–ื›ื•ืช ื”ืืžื™ืชื™ืช ืœืขื‘ื•ื“ ื‘ืžื›ื•ืŸ ื”ืžื—ืงืจ "ืกืงืจื™ืคืก"
00:09
for the last 17 years.
1
9339
1835
ื‘ืžืฉืš 17 ื”ืฉื ื™ื ื”ืื—ืจื•ื ื•ืช.
00:11
It's the largest nonprofit biomedical institution in the country.
2
11174
5255
ื–ื”ื• ื”ืžืœื›"ืจ ื”ื‘ื™ื•-ืจืคื•ืื™ ื”ื’ื“ื•ืœ ื‘ื™ื•ืชืจ ื‘ืžื“ื™ื ื”.
00:16
And I've watched some of my colleagues,
3
16972
2419
ื•ืฆืคื™ืชื™ ื‘ื›ืžื” ืžืขืžื™ืชื™
00:19
who have spent two to three years
4
19391
2294
ื›ืฉื”ืงื“ื™ืฉื• ื‘ื™ืŸ ืฉื ืชื™ื™ื ืœืฉืœื•ืฉ ืฉื ื™ื
00:21
to define the crystal 3-D structure of a protein.
5
21685
3253
ื›ื“ื™ ืœื”ื’ื“ื™ืจ ืืช ื”ืžื‘ื ื” ื”ืชืœืช-ืžืžื“ื™ ื”ื’ื‘ื™ืฉื™ ืฉืœ ื—ืœื‘ื•ืŸ.
00:26
Well, now that can be done in two or three minutes.
6
26231
3420
ืขื›ืฉื™ื• ื–ื” ื™ื›ื•ืœ ืœื”ื™ืขืฉื•ืช ืชื•ืš ืฉืชื™ื™ื-ืฉืœื•ืฉ ื“ืงื•ืช.
00:29
And that's because of the work of AlphaFold,
7
29693
2419
ื•ื–ื”, ื”ื•ื“ื•ืช ืœืขื‘ื•ื“ืชื” ืฉืœ "ืืœืคื-ืคื•ืœื“",
00:32
which is a derivative of DeepMind, Demis Hassabis and John Jumper,
8
32112
6882
ืฉื”ื™ื ืชื•ืฆืจ ืžืฉื•ืœื‘ ืฉืœ "ื“ื™ืค-ืžื™ื™ื ื“", ื“ืžื™ืก ื”ืืกืื‘ื™ืก ื•ื’โ€˜ื•ืŸ ื’โ€™ืืžืคืจ,
00:38
recognized by the American Nobel Prize in September.
9
38994
3503
ื—ืชื ื™ ืคืจืก ื ื•ื‘ืœ ื”ืืžืจื™ืงืื™ ื‘ืกืคื˜ืžื‘ืจ.
00:42
What's interesting, this work,
10
42998
1918
ืžื” ืฉืžืขื ื™ื™ืŸ, ื”ืขื‘ื•ื“ื” ื”ื–ื•,
00:44
which is taking the amino acid sequence in one dimension
11
44958
4588
ืฉืœื•ืงื—ืช ืืช ืจืฆืฃ ื—ื•ืžืฆืช ื”ืืžื™ื ื• ื‘ืžืžื“ ืื—ื“
00:49
and predicting the three-dimensional protein at atomic level,
12
49588
5296
ื•ืžืชืืจืช ืืช ื”ื—ืœื‘ื•ืŸ ื”ืชืœืช-ืžื™ืžื“ื™ ื‘ืจืžื” ื”ืื˜ื•ืžื™ืช,
00:54
[has] now inspired many other of these protein structure prediction models,
13
54926
5881
ื”ื™ื ืžืงื•ืจ ื”ื”ืฉืจืื” ืœื“ื’ืžื™ ื—ื™ื–ื•ื™ ืจื‘ื™ื ืื—ืจื™ื ืฉืœ ืžื‘ื ื™ ื—ืœื‘ื•ืŸ,
01:00
as well as RNA and antibodies,
14
60849
2711
ื›ืžื• ื’ื ืจื "ื ื•ื ื•ื’ื“ื ื™ื,
01:03
and even being able to pick up all the missense mutations in the genome,
15
63602
4838
ื•ื”ื™ื ืืคื™ืœื• ืžืกื•ื’ืœืช ืœื–ื”ื•ืช ืืช ื›ืœ ืžื•ื˜ืฆื™ื•ืช ื”ืกึถืœึถืฃ ื‘ื’ื ื•ื,
01:08
and even being able to come up wit proteins
16
68481
4046
ื•ืืฃ ืœื”ืžืฆื™ื ื—ืœื‘ื•ื ื™ื
01:12
that have never been invented before, that don't exist in nature.
17
72569
3920
ืฉืœื ื”ื•ืžืฆืื• ืœืคื ื™ ื›ืŸ, ื•ืฉืื™ื ื ืงื™ื™ืžื™ื ื‘ื˜ื‘ืข.
01:16
Now, the only thing I think about this is it was a transformer model,
18
76990
3295
ืื ื™ ืจืง ื—ื•ืฉื‘ ืขืœ ื–ื” -- ื–ื” ื”ื™ื” ืžื•ื“ืœ ื˜ืจื ืกืคื•ืจืžืจ,
01:20
we'll talk about that in a moment,
19
80285
2169
ืžื™ื“ ื ื“ื‘ืจ ืขืœ ื–ื” --
01:22
in this award, since Demis and John
20
82454
5046
ื‘ืคืจืก ื”ื–ื”, ื”ื™ื•ืช ืฉื“ืžื™ืก ื•ื’โ€™ื•ืŸ ื•ืฆื•ื•ืช 30 ื”ืžื“ืขื ื™ื ืฉืœื”ื
01:27
and their team of 30 scientists
21
87542
1877
01:29
don't understand how the transformer model works,
22
89419
4004
ืœื ืžื‘ื™ื ื™ื ืื™ืš ืขื•ื‘ื“ ืžื•ื“ืœ ื”ื˜ืจื ืกืคื•ืจืžืจ,
01:33
shouldn't the AI get an asterisk as part of that award?
23
93465
4963
ื”ืื ื”ื‘ื™ื ื” ื”ืžืœืื›ื•ืชื™ืช ืœื ืจืื•ื™ื” ืœืื™ื–ื• ื›ื•ื›ื‘ื™ืช ื‘ืคืจืก ื”ื–ื”?
01:39
I'm going to switch from life science,
24
99262
2127
ืืขื‘ื•ืจ ื›ืขืช ืžืžื“ืขื™ ื”ื—ื™ื™ื,
01:41
which has been the singular biggest contribution just reviewed,
25
101431
4129
ื”ืชืจื•ืžื” ื”ื™ื—ื™ื“ื” ื”ื’ื“ื•ืœื” ื‘ื™ื•ืชืจ ืฉื ืกืงืจื” ื–ื” ืขืชื”,
01:45
to medicine.
26
105560
1335
ืœืจืคื•ืื”.
01:47
And in the medical community,
27
107604
1501
ื•ื‘ืงื”ื™ืœื” ื”ืจืคื•ืื™ืช,
01:49
the thing that we don't talk much about are diagnostic medical errors.
28
109105
6215
ืื ื—ื ื• ืœื ืžืจื‘ื™ื ืœื“ื‘ืจ ืขืœ ื˜ืขื•ื™ื•ืช ืจืคื•ืื™ื•ืช ื‘ืื‘ื—ื•ืŸ.
01:55
And according to the National Academy of Medicine,
29
115362
2669
ื•ืœืคื™ ื”ืืงื“ืžื™ื” ื”ืœืื•ืžื™ืช ืœืจืคื•ืื”,
01:58
all of us will experience at least one in our lifetime.
30
118031
3462
ื›ื•ืœื ื• ื ื—ื•ื•ื” ืœืคื—ื•ืช ืื‘ื—ื•ืŸ ืฉื’ื•ื™ ืื—ื“ ื‘ืžื”ืœืš ื—ื™ื™ื ื•.
02:01
And we know from a recent Johns Hopkins study
31
121993
2294
ื•ืื ื—ื ื• ื™ื•ื“ืขื™ื ืžืžื—ืงืจ ืฉืœ ืฉื ืขืจืš ืœืื—ืจื•ื ื” ื‘"ื’โ€™ื•ื ืก ื”ื•ืคืงื™ื ืก"
02:04
that these errors have led to 800,000 Americans dead
32
124329
5672
ืฉื˜ืขื•ื™ื•ืช ืืœื” ื’ืจืžื• ืœืžื•ืชื ืฉืœ 800,000 ืืžืจื™ืงื ื™ื
02:10
or seriously disabled each year.
33
130043
3629
ืื• ืœื ื›ื•ื™ื•ืช ืงืฉื•ืช ื‘ื›ืœ ืฉื ื” ื•ืฉื ื”.
02:13
So this is a big problem.
34
133713
1502
ื›ืš ืฉื–ื• ื‘ืขื™ื” ื—ืžื•ืจื”.
02:15
And the question is, can AI help us?
35
135215
3211
ื•ื”ืฉืืœื” ื”ื™ื, ื”ืื ื”ื‘ื™ื ื” ื”ืžืœืื›ื•ืชื™ืช ืชื•ื›ืœ ืœืขื–ื•ืจ ืœื ื•?
02:18
And you keep hearing about the term โ€œprecision medicine.โ€
36
138843
3295
ื•ื›ืœ ื”ื–ืžืŸ ืฉื•ืžืขื™ื ืขืœ โ€œืจืคื•ืื” ืžื“ื•ื™ืงืชโ€œ.
02:22
Well, if you keep making the same mistake over and over again, that's very precise.
37
142806
5589
ืื– ืื ืžืžืฉื™ื›ื™ื ืœืขืฉื•ืช ืฉื•ื‘ ื•ืฉื•ื‘ ืืช ืื•ืชื” ื”ื˜ืขื•ืช, ื–ื” ืžื“ื•ื™ืง ืžืื•ื“.
02:28
(Laughter)
38
148436
1168
(ืฆื—ื•ืง)
ื›ืŸ. ืื ื—ื ื• ืœื ืฆืจื™ื›ื™ื ืืช ื–ื”,
02:30
We don't need that,
39
150188
1168
02:31
we need accuracy and precision medicine.
40
151398
2669
ืื ื—ื ื• ืฆืจื™ื›ื™ื ื“ื™ื•ืง ื•ื’ื ืจืคื•ืื” ืžื“ื•ื™ืงืช.
02:34
So can we get there?
41
154442
1585
ื”ืื ื ื•ื›ืœ ืœื”ื’ื™ืข ืœื›ืš?
02:36
Well, this is a picture of the retina.
42
156486
2252
ื–ืืช ืชืžื•ื ื” ืฉืœ ื”ืจืฉืชื™ืช.
02:39
And this was the first major hint,
43
159072
2961
ื•ื–ื” ื”ื™ื” ื”ืจืžื– ื”ื’ื“ื•ืœ ื”ืจืืฉื•ืŸ,
02:42
training 100,000 images with supervised learning.
44
162075
4963
ืœืื—ืจ ืื™ืžื•ืŸ ื‘-100,000 ืชืžื•ื ื•ืช ื‘ืœืžื™ื“ื” ื‘ืคื™ืงื•ื—,
02:47
Could the machine see things that people couldn't see?
45
167080
4880
ื”ืื ื”ืžื›ื•ื ื” ืชืฆืœื™ื— ืœืจืื•ืช ื“ื‘ืจื™ื ืฉื‘ื ื™-ืื“ื ื”ื—ืžื™ืฆื•?
02:52
And so the question was, to the retinal experts,
46
172919
2961
ืื– ื”ืฉืืœื” ืฉื”ื•ืฆื’ื” ืœืžื•ืžื—ื™ ื”ืจืฉืชื™ืช,
02:55
is this from a man or a woman?
47
175880
2002
ื”ื™ืชื”, ื”ืื ื”ื™ื ืฉืœ ื’ื‘ืจ ืื• ืฉืœ ืื™ืฉื”?
02:58
And the chance of getting it accurate was 50 percent.
48
178717
3420
ื›ืฉื”ืกื™ื›ื•ื™ ืœืงืœื•ืข ื”ื™ื” 50 ืื—ื•ื–.
03:02
(Laughter)
49
182137
1167
(ืฆื—ื•ืง)
03:03
But the AI got it right, 97 percent.
50
183346
3754
ืื‘ืœ ื”ื‘ื™ื ื” ื”ืžืœืื›ื•ืชื™ืช ื“ื™ื™ืงื”. 97%.
03:07
So that training,
51
187142
2043
ื‘ืื™ืžื•ืŸ ื”ื”ื•ื, ื”ืชื›ื•ื ื•ืช ืฉื™ืืคืฉืจื• ืืช ื–ื”
03:09
the features are not even fully defined of how that was possible.
52
189227
4171
03:14
Well that gets then to all of medical images.
53
194274
3086
ื–ื” ืžืชื™ื™ื—ืก ืœื›ืœ ื”ืชืžื•ื ื•ืช ื”ืจืคื•ืื™ื•ืช.
03:17
This is just representative, the chest X-ray.
54
197652
2628
ื–ื” ืจืง ืžื™ื™ืฆื’. ืฆื™ืœื•ื ื”ืจื ื˜ื’ืŸ ืฉืœ ื”ื—ื–ื”.
03:20
And in fact with the chest X-ray,
55
200822
2169
ื•ืœืžืขืฉื” ืขื ืฆื™ืœื•ื ื”ืจื ื˜ื’ืŸ ืฉืœ ื”ื—ื–ื”,
03:22
the ability here for the AI to pick up,
56
202991
3086
ื”ื™ื›ื•ืœืช ื›ืืŸ ืฉืœ ื”ื‘ื™ื ื” ื”ืžืœืื›ื•ืชื™ืช ืœื–ื”ื•ืช...
03:26
the radiologists, expert radiologists missing the nodule,
57
206077
4880
ื”ืจื“ื™ื•ืœื•ื’ื™ื, ื”ืจื“ื™ื•ืœื•ื’ื™ื ื”ืžื•ืžื—ื™ื ื”ื—ืžื™ืฆื• ืืช ื”ื’ื•ืฉ,
03:30
which turned out to be picked up by the AI as cancerous,
58
210999
3128
ื•ื”ืชื‘ืจืจ ืฉื”ื‘ื™ื ื” ื”ืžืœืื›ื•ืชื™ืช ื–ื™ื”ืชื” ืื•ืชื• ื›ืกืจื˜ื ื™,
03:34
and this is, of course, representative of all of medical scans,
59
214127
4004
ื•ื–ื”, ื›ืžื•ื‘ืŸ, ืžื™ื™ืฆื’ ืืช ื›ืœ ื”ืกืจื™ืงื•ืช ื”ืจืคื•ืื™ื•ืช,
03:38
whether itโ€™s CT scans, MRI, ultrasound.
60
218173
3837
ื˜ื•ืžื•ื’ืจืคื™ื” ืžืžื•ื—ืฉื‘ืช, ืกืจื™ืงื•ืช ื“ื™ืžื•ืช ืžื’ื ื˜ื™, ืื•ืœื˜ืจืกืื•ื ื“.
03:42
That through supervised learning of large, labeled, annotated data sets,
61
222051
5881
ืฉื‘ืืžืฆืขื•ืช ืœืžื™ื“ื” ื‘ืคื™ืงื•ื— ืฉืœ ืžืขืจื›ื™ ื ืชื•ื ื™ื ื’ื“ื•ืœื™ื, ืžืชื•ื™ื’ื™ื ื•ืžื‘ื•ืืจื™ื,
03:47
we can see AI do at least as well, if not better,
62
227974
3921
ืื ื• ื™ื›ื•ืœื™ื ืœืจืื•ืช ืฉื”ื‘ื™ื ื” ื”ืžืœืื›ื•ืชื™ืช
ืžืฆืœื™ื—ื” ืœืคื—ื•ืช ื›ืžื• ืจื•ืคืื™ื ืžื•ืžื—ื™ื, ืื ืœื ื˜ื•ื‘ ื™ื•ืชืจ,
03:51
than expert physicians.
63
231895
1960
03:55
And 21 randomized trials of picking up polyps --
64
235023
4755
ื•ื‘-21 ื ื™ืกื•ื™ื™ื ืืงืจืื™ื™ื ืฉืœ ื–ื™ื”ื•ื™ ืคื•ืœื™ืคื™ื -
03:59
machine vision during colonoscopy -- have all shown
65
239819
4171
ืจืื™ื™ืช ืžื›ื•ื ื” ื‘ืžื”ืœืš ืงื•ืœื•ื ื•ืกืงื•ืคื™ื”,
ื”ืจืื• ื›ื•ืœื ืฉื–ื™ื”ื•ื™ ื”ืคื•ืœื™ืคื™ื ืžื‘ื•ืฆืข ื˜ื•ื‘ ื™ื•ืชืจ
04:03
that polyps are picked up better
66
243990
3003
04:06
with the aid of machine vision than by the gastroenterologist alone,
67
246993
3796
ื‘ืขื–ืจืช ืจืื™ื™ืช ื”ืžื›ื•ื ื” ืžืืฉืจ ืข"ื™ ื”ื’ืกื˜ืจื•ืื ื˜ืจื•ืœื•ื’ ื‘ืœื‘ื“,
04:10
especially as the day goes on, later in the day, interestingly.
68
250789
4337
ื‘ืžื™ื•ื—ื“ ื›ื›ืœ ืฉื”ื™ื•ื ื ืžืฉืš, ื•ื‘ืฉืขื•ืช ืžืื•ื—ืจื•ืช ื™ื•ืชืจ, ืฉื–ื” ืžืขื ื™ื™ืŸ.
04:15
We don't know whether picking up all these additional polyps
69
255168
3253
ืื™ื ื ื• ื™ื•ื“ืขื™ื ืื ื–ื™ื”ื•ื™ ื›ืœ ื”ืคื•ืœื™ืคื™ื ื”ื ื•ืกืคื™ื ื”ืืœื”
04:18
changes the natural history of cancers,
70
258463
2085
ืžืฉื ื” ืืช ื”ื”ื™ืกื˜ื•ืจื™ื” ื”ื˜ื‘ืขื™ืช ืฉืœ ื”ืกืจื˜ืŸ,
04:20
but it tells you about machine eyes,
71
260590
3086
ืื‘ืœ ื–ื” ืื•ืžืจ ืœื›ื ืžืฉื”ื• ืขืœ ืขื™ื ื™ ืžื›ื•ื ื”,
04:23
the power of machine eyes.
72
263718
1376
ืขืœ ื”ืขื•ืฆืžื” ืฉืœ ืขื™ื ื™ ื”ืžื›ื•ื ื”.
04:25
Now that was interesting.
73
265470
2377
ื–ื” ื”ื™ื” ืžืขื ื™ื™ืŸ.
04:27
But now still with deep learning models, not transformer models,
74
267889
5714
ืื‘ืœ... ืขื“ื™ื™ืŸ ืœื’ื‘ื™ ืžื•ื“ืœื™ื ืฉืœ ืœืžื™ื“ื” ืขืžื•ืงื”,
ืœื ืžื•ื“ืœื™ื ืฉืœ ื˜ืจื ืกืคื•ืจืžืจ,
04:33
we've seen and learned that the ability
75
273645
3253
ืจืื™ื ื• ื•ืœืžื“ื ื• ืฉื”ื™ื›ื•ืœืช ืฉืœ ืจืื™ื™ืช ืžื—ืฉื‘
04:36
for computer vision to pick up things that human eyes can't see
76
276898
5589
ืœื–ื”ื•ืช ื“ื‘ืจื™ื ืฉืขื™ื ื™ ืื ื•ืฉ ืœื ื™ื›ื•ืœื•ืช ืœืจืื•ืช
04:42
is quite remarkable.
77
282487
1460
ื”ื™ื ืžื“ื”ื™ืžื” ืœืžื“ื™. ื”ื ื” ื”ืจืฉืชื™ืช.
04:43
Here's the retina.
78
283988
1418
04:46
Picking up the control of diabetes and blood pressure.
79
286074
3378
ื–ื™ื”ื•ื™ ืชืกืžื™ื ื™ ืกื•ื›ืจืช ื•ืœื—ืฅ-ื“ื.
04:50
Kidney disease.
80
290495
1710
ืžื—ืœืช ื›ืœื™ื•ืช.
04:52
Liver and gallbladder disease.
81
292872
2586
ืžื—ืœืช ื›ื‘ื“ ื•ื›ื™ืก ื”ืžืจื”.
04:56
The heart calcium score,
82
296084
2043
ืจืžืช ื”ืกื™ื“ืŸ ื‘ืœื‘,
04:58
which you would normally get through a scan of the heart.
83
298127
4004
ืฉื‘ื“ืจืš ื›ืœืœ ื”ื™ื™ืชื ืžืงื‘ืœื™ื ื‘ืืžืฆืขื•ืช ืกืจื™ืงื” ืฉืœ ื”ืœื‘.
05:03
Alzheimer's disease before any clinical symptoms have been manifest.
84
303174
4129
ืžื—ืœืช ืืœืฆื”ื™ื™ืžืจ ืœืคื ื™ ื”ื•ืคืขืช ืชืกืžื™ื ื™ื ืงืœื™ื ื™ื™ื ื›ืœืฉื”ื.
05:08
Predicting heart attacks and strokes.
85
308012
2586
ื—ื™ื–ื•ื™ ื”ืชืงืคื™ ืœื‘ ื•ืฉื‘ืฅ ืžื•ื—ื™.
05:11
Hyperlipidemia.
86
311599
1543
ื”ื™ืคืจืœื™ืคื™ื“ืžื™ื”.
05:13
And seven years before any symptoms of Parkinson's disease,
87
313518
4546
ื•ืฉื‘ืข ืฉื ื™ื ืœืคื ื™ ื”ื•ืคืขืช ืชืกืžื™ื ื™ื ื›ืœืฉื”ื ืฉืœ ืžื—ืœืช ืคืจืงื™ื ืกื•ืŸ,
05:18
to pick that up.
88
318064
1251
ื•ืจืง ืื– ื ืื‘ื—ืŸ ื–ืืช.
05:19
Now this is interesting because in the future,
89
319941
3587
ื•ื–ื” ืžืขื ื™ื™ืŸ ื›ื™ ื‘ืขืชื™ื“,
05:23
we'll be taking pictures of our retina at checkups.
90
323570
3753
ื ืฆืœื ืชืžื•ื ื•ืช ืฉืœ ื”ืจืฉืชื™ืช ื‘ื‘ื“ื™ืงื•ืช ืชืงื•ืคืชื™ื•ืช.
05:27
This is the gateway to almost every system in the body.
91
327365
3462
ื–ื”ื• ื”ืฉืขืจ ืœื›ืœ ืžืขืจื›ืช ื‘ื’ื•ืฃ ื›ืžืขื˜.
05:31
It's really striking.
92
331369
1168
ื–ื” ืžืžืฉ ืžื“ื”ื™ื.
05:32
And we'll come back to this because each one of these studies
93
332579
4087
ื•ืขื•ื“ ื ื—ื–ื•ืจ ืœื–ื” ื›ื™ ื›ืœ ืื—ื“ ืžื”ืžื—ืงืจื™ื ื”ืืœื”
05:36
was done with tens or hundreds [of] thousands of images
94
336666
4213
ื ืขืฉื” ืขื ืขืฉืจื•ืช ืื• ืžืื•ืช ืืœืคื™ ืชืžื•ื ื•ืช ื‘ืœืžื™ื“ื” ืžืคื•ืงื—ืช,
05:40
with supervised learning,
95
340879
1251
05:42
and theyโ€™re all separate studies by different investigators.
96
342171
3921
ื•ื›ื•ืœื ืžื—ืงืจื™ื ื ืคืจื“ื™ื ืฉืœ ื—ื•ืงืจื™ื ืฉื•ื ื™ื.
05:46
Now, as a cardiologist, I love to read cardiograms.
97
346426
4045
ื›ืงืจื“ื™ื•ืœื•ื’, ืื ื™ ืื•ื”ื‘ ืœืงืจื•ื ืงืจื“ื™ื•ื’ืจืžื•ืช.
05:50
I've been doing it for over 30 years.
98
350513
2169
ืื ื™ ืขื•ืฉื” ืืช ื–ื” ื›ื‘ืจ ืœืžืขืœื” ืž-30 ืฉื ื”.
05:53
But I couldn't see these things.
99
353808
2086
ืื‘ืœ ืœื ื™ื›ื•ืœืชื™ ืœืจืื•ืช ืืช ื”ื“ื‘ืจื™ื ื”ืืœื”.
05:56
Like, the age and the sex of the patient,
100
356519
2920
ื›ืžื• ื”ื’ื™ืœ ื•ื”ืžื™ืŸ ืฉืœ ื”ืžื˜ื•ืคืœ,
05:59
or the ejection fraction of the heart,
101
359439
3086
ืื• ืžืงื˜ืข ื”ืคืœื™ื˜ื” ืฉืœ ื”ืœื‘,
06:02
making difficult diagnoses that are frequently missed.
102
362567
3503
ื‘ื™ืฆื•ืข ืื‘ื—ื ื•ืช ืงืฉื•ืช ืฉืœืขืชื™ื ืงืจื•ื‘ื•ืช ืžื•ื—ืžืฆื•ืช.
06:06
The anemia of the patient, that is, the hemoglobin to the decimal point.
103
366571
4212
ื”ืื ืžื™ื” ืฉืœ ื”ืžื˜ื•ืคืœ, ื›ืœื•ืžืจ ื”ื”ืžื•ื’ืœื•ื‘ื™ืŸ ืขื“ ืœื ืงื•ื“ื” ื”ืขืฉืจื•ื ื™ืช.
06:11
Predicting whether a person,
104
371951
1460
ืœื—ื–ื•ืช ืื ืื“ื,
06:13
who's never had atrial fibrillation or stroke
105
373453
2502
ืฉืžืขื•ืœื ืœื ื—ื•ื•ื” ืคืจืคื•ืจ ืคืจื•ื–ื“ื•ืจื™ื ืื• ืฉื‘ืฅ ืžื•ื—ื™
06:15
from the ECG,
106
375955
1418
ืขืœ ืคื™ ื”ืืงโ€œื’,
06:17
whether that's going to likely occur.
107
377415
2169
ื”ืื ืกื‘ื™ืจ ืœื”ื ื™ื— ืฉื–ื” ื™ืงืจื”.
06:20
Diabetes, a diagnosis of diabetes and prediabetes, from the cardiogram.
108
380418
4796
ืกื•ื›ืจืช, ืื‘ื—ื ื” ืฉืœ ืกื•ื›ืจืช ื•ื˜ืจื•ื-ืกื•ื›ืจืช, ืžื”ืงืจื“ื™ื•ื’ืจืžื”.
06:25
The filling pressure of the heart.
109
385965
2044
ืœื—ืฅ ื”ืžื™ืœื•ื™ ืฉืœ ื”ืœื‘.
06:28
Hypothyroidism
110
388509
2086
ืชืช-ืคืขื™ืœื•ืช ืฉืœ ื‘ืœื•ื˜ืช ื”ืชืจื™ืก
06:30
and kidney disease.
111
390637
1626
ื•ืžื—ืœืช ื›ืœื™ื•ืช.
06:32
Imagine getting an electrocardiogram to tell you about all these other things,
112
392305
3920
ืชืืจื• ืœืขืฆืžื›ื, ืœืงื‘ืœ ืืœืงื˜ืจื•ืงืจื“ื™ื•ื’ืจืžื” ืฉืชืกืคืจ ืœื›ื ืขืœ ื›ืœ ื”ื“ื‘ืจื™ื ื”ืืœื”,
06:36
not really so much about the heart.
113
396267
2711
ืฉืคื—ื•ืช ืžืชื™ื™ื—ืกื™ื ืœืœื‘.
06:39
Then there's the chest X-ray.
114
399729
1543
ื•ื™ืฉื ื• ื’ื ืฆื™ืœื•ื ื”ืจื ื˜ื’ืŸ ืฉืœ ื”ื—ื–ื”.
06:41
Who would have guessed that we could accurately determine
115
401314
3920
ืžื™ ื”ื™ื” ื—ื•ืฉื‘ ืฉื ื•ื›ืœ ืœืงื‘ื•ืข ื‘ืžื“ื•ื™ืง
06:45
the race of the patient,
116
405234
1377
ืืช ื’ื–ืข ื”ืžื˜ื•ืคืœ,
06:46
no less the ethical implications of that,
117
406611
2794
ืœื ืคื—ื•ืช ืžื”ื”ืฉืœื›ื•ืช ื”ืืชื™ื•ืช ืฉืœ ื–ื”,
06:49
from a chest X-ray through machine eyes?
118
409405
3379
ืขืœ ืคื™ ืฆื™ืœื•ื ืจื ื˜ื’ืŸ ื‘ื—ื–ื” ื“ืจืš ืขื™ื ื™ ื”ืžื›ื•ื ื”?
06:53
And interestingly, picking up the diagnosis of diabetes,
119
413201
4171
ื•ืžืขื ื™ื™ืŸ, ื–ื™ื”ื•ื™ ื”ืื‘ื—ื ื” ืฉืœ ืกื•ื›ืจืช,
06:57
as well as how well the diabetes is controlled,
120
417372
4212
ื›ืžื• ื’ื ืขื“ ื›ืžื” ื”ืกื•ื›ืจืช ืžื‘ื•ืงืจืช,
07:01
through the chest X-ray.
121
421584
1668
ื‘ืืžืฆืขื•ืช ืฆื™ืœื•ื ืจื ื˜ื’ืŸ ืฉืœ ื”ื—ื–ื”.
07:04
And of course, so many different parameters about the heart,
122
424629
3795
ื•ื›ืžื•ื‘ืŸ, ื›ืœ-ื›ืš ื”ืจื‘ื” ืคืจืžื˜ืจื™ื ืฉื•ื ื™ื ืœื’ื‘ื™ ื”ืœื‘,
07:08
which we could never,
123
428424
2169
ืฉืœืขื•ืœื ืœื ื ื•ื›ืœ -
07:10
radiologists or cardiologists, could never be able to come up
124
430593
3837
ืจื“ื™ื•ืœื•ื’ื™ื ืื• ืงืจื“ื™ื•ืœื•ื’ื™ื - ืœืขื•ืœื ืœื ื ื•ื›ืœ ืœื”ื’ื™ืข
07:14
with what machine vision can do.
125
434430
2878
ืœืžื” ืฉืจืื™ื™ืช ื”ืžื›ื•ื ื” ื™ื›ื•ืœื” ืœืขืฉื•ืช.
07:17
Pathologists often argue about a slide,
126
437976
3169
ื”ืคืชื•ืœื•ื’ื™ื ืžืจื‘ื™ื ืœื”ืชื•ื•ื›ื— ืขืœ ืฉืงื•ืคื™ืช ื–ื• ืื• ืื—ืจืช,
07:21
about what does it really show?
127
441187
1794
ืžื” ื‘ืืžืช ืจื•ืื™ื ื‘ื”?
07:23
But with this ability of machine eyes,
128
443314
4338
ืื‘ืœ ืขื ื™ื›ื•ืœืช ื–ื• ืฉืœ ืขื™ื ื™ ืžื›ื•ื ื”,
07:27
the driver genomic mutations of the cancer can be defined,
129
447694
3878
ื ื™ืชืŸ ืœื”ื’ื“ื™ืจ ืืช ื”ืžื•ื˜ืฆื™ื•ืช ื”ื’ื ื˜ื™ื•ืช ื”ื™ืกื•ื“ื™ื•ืช ืฉืœ ื”ืกืจื˜ืŸ,
07:31
no less the structural copy number variants
130
451614
2878
ื•ืœื ืคื—ื•ืช ืžื›ืš - ื”ื’ืจืกืื•ืช ืฉืœ ืžืกืคืจ ื”ืขื•ืชืง ื”ืžื‘ื ื™
07:34
that are accounting or present in that tumor.
131
454534
2878
ืฉืžืกื‘ื™ืจื™ื ืืช ื”ื’ื™ื“ื•ืœ ืื• ืงื™ื™ืžื™ื ื‘ื•.
07:37
Also, where is that tumor coming from?
132
457787
2336
ื›ืžื• ื›ืŸ, ืžื ื™ื™ืŸ ื‘ื ื”ื’ื™ื“ื•ืœ ื”ื–ื”?
07:40
For many patients, we donโ€™t know.
133
460164
2253
ืืฆืœ ืžื˜ื•ืคืœื™ื ืจื‘ื™ื, ืื ื—ื ื• ืœื ื™ื•ื“ืขื™ื.
07:42
But it can be determined through AI.
134
462458
4255
ืื‘ืœ ื–ื” ื™ื›ื•ืœ ืœื”ื™ืงื‘ืข ื‘ืขื–ืจืช ื‘ื™ื ื” ืžืœืื›ื•ืชื™ืช.
07:46
And also the prognosis of the patient,
135
466754
2836
ื•ื’ื ื”ืคืจื•ื’ื ื•ื–ื” ืฉืœ ื”ืžื˜ื•ืคืœ,
07:49
just from the slide,
136
469590
2169
ืจืง ืžื”ืฉืงื•ืคื™ืช,
07:51
by all of the training.
137
471801
1627
ืชื•ื“ื•ืช ืœื›ืœ ื”ืื™ืžื•ื ื™ื.
07:53
Again, this is all just convolutional neural networks,
138
473469
4797
ืฉื•ื‘, ื›ืœ ืืœื” ื”ื ืจืง ืจืฉืชื•ืช ืงื•ื ื‘ื•ืœืฆื™ื”,
07:58
not transformer models.
139
478307
1669
ืœื ืžื•ื“ืœื™ื ืฉืœ ื˜ืจื ืกืคื•ืจืžืจ.
08:00
So when we go from the deep neural networks to transformer models,
140
480852
5630
ืื– ื›ืฉืื ื• ืขื•ื‘ืจื™ื ืžื”ืจืฉืชื•ืช ื”ืขืฆื‘ื™ื•ืช ื”ืขืžื•ืงื•ืช ืœืžื•ื“ืœื™ื ืฉืœ ื˜ืจื ืกืคื•ืจืžืจ,
08:06
this classic pre-print,
141
486524
2085
ื”ื˜ื™ื•ื˜ื” ื”ืงืœืืกื™ืช ื”ื–ืืช,
08:08
one of the most cited pre-prints ever,
142
488651
2586
ืื—ืช ืžื”ื˜ื™ื•ื˜ื•ืช ื”ืžื•ืงื“ืžื•ืช ื‘ื™ื•ืชืจ ืฉืฆื•ื˜ื˜ื• ืื™-ืคืขื,
08:11
"Attention is All You Need,"
143
491237
1418
"ืฆืจื™ืš ืจืง ืชืฉื•ืžืช-ืœื‘",
08:12
the ability to now be able to look at many more items,
144
492697
4296
ื”ื™ื›ื•ืœืช ืœื”ื™ื•ืช ืžืกื•ื’ืœื™ื ืœื”ืกืชื›ืœ ื›ื™ื•ื ืขืœ ืคืจื™ื˜ื™ื ืจื‘ื™ื ื ื•ืกืคื™ื,
ื‘ื™ืŸ ืื ื–ื• ืฉืคื” ืื• ืชืžื•ื ื•ืช,
08:17
whether it be language or images,
145
497035
3837
08:20
and be able to put this in context,
146
500913
2962
ืœื”ื™ื•ืช ืžืกื•ื’ืœื™ื ืœื”ื›ื ื™ืก ืืช ื–ื” ืœื”ืงืฉืจ,
08:23
setting up a transformational progress in many fields.
147
503916
4588
ื•ืœื”ื’ื“ื™ืจ ื”ืชืงื“ืžื•ืช ื‘ืžื•ื“ืœ ื˜ืจื ืกืคื•ืจืžืจ ื‘ืชื—ื•ืžื™ื ืจื‘ื™ื.
08:29
The prototype is, the outgrowth of this is GPT-4.
148
509172
4504
ื”ืื‘ื˜ื™ืคื•ืก ื”ื–ื” ืฆืžื— ืžื”ื’'ื™-ืคื™-ื˜ื™ 4,
08:34
With over a trillion connections.
149
514510
2628
ืขื ื™ื•ืชืจ ืžื˜ืจื™ืœื™ื•ืŸ ืงืฉืจื™ื.
08:37
Our human brain has 100 trillion connections or parameters.
150
517138
4713
ืœืžื•ื— ื”ืื ื•ืฉื™ ืฉืœื ื• ื™ืฉ ืžืื” ื˜ืจื™ืœื™ื•ืŸ ืงืฉืจื™ื ืื• ืคืจืžื˜ืจื™ื.
08:42
But one trillion,
151
522185
1167
ืื‘ืœ ื˜ืจื™ืœื™ื•ืŸ ืื—ื“... ื—ื™ืฉื‘ื• ืจืง ืขืœ ื›ืœ ื”ืžื™ื“ืข, ื”ื™ื“ืข,
08:43
just think of all the information, knowledge,
152
523352
2128
08:45
that's packed into those one trillion.
153
525480
1876
ืฉื’ืœื•ืžื™ื ื‘ื˜ืจื™ืœื™ื•ืŸ ื”ืื—ื“ ื”ื–ื”.
08:47
And interestingly, this is now multimodal with language, with images,
154
527398
4880
ื•ืžืขื ื™ื™ืŸ ืฉืขื›ืฉื™ื• ื–ื” ืจื‘-ืžื•ื“ืืœื™ ืขื ืฉืคื”, ืขื ืชืžื•ื ื•ืช,
08:52
with speech.
155
532320
1376
ืขื ื“ื™ื‘ื•ืจ.
08:53
And it involves a massive amount of graphic processing units.
156
533696
3921
ื•ืžืขื•ืจื‘ืช ื‘ื–ื” ื›ืžื•ืช ืื“ื™ืจื” ืฉืœ ื™ื—ื™ื“ื•ืช ืขื™ื‘ื•ื“ ื’ืจืคื™.
08:58
And it's with self-supervised learning,
157
538076
2293
ื•ื–ื” ื ืขืฉื” ื‘ืœืžื™ื“ื” ื‘ืคื™ืงื•ื— ืขืฆืžื™,
09:00
which is a big bottleneck in medicine
158
540369
2128
ืฉื”ื™ื ืฆื•ื•ืืจ ื‘ืงื‘ื•ืง ื’ื“ื•ืœ ื‘ืจืคื•ืื”
09:02
because we can't get experts to label images.
159
542497
3169
ื›ื™ ืื ื—ื ื• ืœื ื™ื›ื•ืœื™ื ืœื”ื‘ื™ื ืžื•ืžื—ื™ื ืœืชื™ื™ื’ ืชืžื•ื ื•ืช.
09:05
This can be done with self-supervised learning.
160
545708
2795
ื–ื” ื‘ืจ-ื‘ื™ืฆื•ืข ื‘ืœืžื™ื“ื” ื‘ืคื™ืงื•ื— ืขืฆืžื™.
09:08
So what does this set up in medicine?
161
548961
2837
ืื– ืžื” ื–ื” ืžื ื‘ื ื‘ืจืคื•ืื”?
09:11
It sets up, for example, keyboard liberation.
162
551839
4421
ื–ื” ืžื ื‘ื ืœืžืฉืœ ืฉื—ืจื•ืจ ืžืžืงืœื“ืช.
09:16
The one thing that both doctors, clinicians
163
556803
3920
ื”ื“ื‘ืจ ื”ื™ื—ื™ื“ ืฉื’ื ืจื•ืคืื™ื, ืงืœื™ื ืื™ื,
09:20
and patients would like to see.
164
560765
2377
ื•ื’ื ืžื˜ื•ืคืœื™ื ื”ื™ื• ืจื•ืฆื™ื ืœืจืื•ืช.
09:23
Everyone hates being data clerks as clinicians,
165
563851
3921
ื›ืงืœื™ื ืื™ื, ื›ื•ืœื ืฉื•ื ืื™ื ืœื”ื™ื•ืช ืคืงื™ื“ื™ ื ืชื•ื ื™ื,
09:27
and patients would like to see their doctor
166
567814
2836
ื•ืžื˜ื•ืคืœื™ื ื”ื™ื• ืจื•ืฆื™ื ืœืคื’ื•ืฉ ืืช ื”ืจื•ืคื ืฉืœื”ื
09:30
when they finally have the visit they've waited for a long time.
167
570650
3753
ื›ืฉืžื’ื™ืข ืกื•ืฃ-ืกื•ืฃ ื”ื‘ื™ืงื•ืจ ืฉืœื• ื”ื ื—ื™ื›ื• ื–ืžืŸ ืจื‘.
09:34
So the ability to change the face-to-face contact
168
574445
4713
ื”ื™ื›ื•ืœืช ืœืฉื ื•ืช ืืช ื”ืงืฉืจ ื”ื‘ืœืชื™-ืืžืฆืขื™
09:39
is just one step along the way.
169
579200
2502
ื”ื™ื ืจืง ืฉืœื‘ ืื—ื“ ื‘ื“ืจืš.
09:41
By having the liberation from keyboards with synthetic notes
170
581744
5005
ืขืœ ื™ื“ื™ ืฉื—ืจื•ืจ ืžืžืงืœื“ื•ืช ืขื ืชื•ื•ื™ื ืกื™ื ืชื˜ื™ื™ื
09:46
that are driven, derived from the conversation,
171
586791
2753
ื”ืžื•ื ืขื™ื, ื”ื ื’ื–ืจื™ื ืžื”ืฉื™ื—ื”,
09:49
and then all the downstream normal data clerk functions that are done,
172
589585
4880
ื•ื›ืœ ืคื•ื ืงืฆื™ื•ืช ืคืงื™ื“ื•ืช ื”ื ืชื•ื ื™ื ื”ืจื’ื™ืœื•ืช ืฉื ืขืฉื•ืช ื‘ื”ืžืฉืš,
09:54
often off-hours.
173
594507
1668
ื‘ืขื‘ื•ื“ื” ืื• ืžื—ื•ืฅ ืœืฉืขื•ืช ื”ืขื‘ื•ื“ื”.
09:56
Now we're seeing in health systems across the United States
174
596217
3587
ื›ืขืช ืื ื• ืจื•ืื™ื ื‘ืžืขืจื›ื•ืช ื‘ืจื™ืื•ืช ื‘ืจื—ื‘ื™ ืืจืฆื•ืช ื”ื‘ืจื™ืช
09:59
where people, physicians are saving many hours of time
175
599846
3920
ืฉืื ืฉื™ื, ืจื•ืคืื™ื, ื—ื•ืกื›ื™ื ืฉืขื•ืช ืจื‘ื•ืช
10:03
and heading towards ultimately keyboard liberation.
176
603808
3587
ื•ืžืชืงื“ืžื™ื ื‘ืขืชื™ื“ ืœื—ื•ืคืฉ ืžืŸ ื”ืžืงืœื“ืช.
10:08
We recently published, with the group at Moorfields Eye Institute,
177
608396
3587
ืœืื—ืจื•ื ื” ืคืจืกืžื ื•, ืขื ื”ืงื‘ื•ืฆื” ื‘"ืžื›ื•ืŸ ื”ืขื™ืŸ ืฉืœ ืžื•ืจืคื™ืœื“ืก",
ื‘ืจืืฉื•ืช ืคื™ืจืก ืงื™ืŸ,
10:12
led by Pearse Keane,
178
612024
1335
10:13
the first foundation model in medicine from the retina.
179
613401
3295
ืืช ืžื•ื“ืœ ื”ื™ืกื•ื“ ื”ืจืคื•ืื™ ื”ืจืืฉื•ืŸ ืขืœ ืคื™ ื”ืจืฉืชื™ืช.
10:16
And remember those eight different things that were all done by separate studies?
180
616737
4380
ื–ื•ื›ืจื™ื ืืช ืฉืžื•ื ืช ื”ื“ื‘ืจื™ื ืฉื›ื•ืœื ื ืขืฉื• ื‘ืžื—ืงืจื™ื ื ืคืจื“ื™ื?
10:21
This was all done with one model.
181
621159
2335
ื›ืœ ื–ื” ื ืขืฉื” ืขื ืžื•ื“ืœ ืื—ื“,
10:23
This is with 1.6 million retinal images
182
623494
3879
ื•ืขื 1.6 ืžื™ืœื™ื•ืŸ ืชืžื•ื ื•ืช ืจืฉืชื™ืช
10:27
predicting all these different outcome likelihoods.
183
627415
4546
ื”ืžื ื‘ืื•ืช ืืช ื›ืœ ื”ืกื‘ื™ืจื•ื™ื•ืช ื”ืœืœื•.
10:32
And this is all open-source,
184
632003
1710
ื•ื›ืœ ื–ื” ืงื•ื“ ืคืชื•ื—,
10:33
which is of course really important that others can build on these models.
185
633754
4380
ื•ื–ื” ื›ืžื•ื‘ืŸ ืžืื•ื“ ื—ืฉื•ื‘, ืฉืื—ืจื™ื ื™ื•ื›ืœื• ืœื‘ื ื•ืช ืขืœ ื”ืžื•ื“ืœื™ื ื”ืืœื”.
10:38
Now I just want to review a couple of really interesting patients.
186
638134
5547
ืขื›ืฉื™ื• ืื ื™ ืจืง ืจื•ืฆื” ืœืขื‘ื•ืจ ืขืœ ื›ืžื” ืžื˜ื•ืคืœื™ื ืžืžืฉ ืžืขื ื™ื™ื ื™ื.
10:44
Andrew, who is now six years old.
187
644098
3003
ืื ื“ืจื•, ืฉื”ื•ื ืขื›ืฉื™ื• ื‘ืŸ ืฉืฉ.
10:47
He had three years of relentlessly increasing pain, arrested growth.
188
647810
7007
ื”ื™ื• ืœื• ืฉืœื•ืฉ ืฉื ื™ื ืฉืœ ื›ืื‘ ื”ื•ืœืš ื•ื’ื•ื‘ืจ, ืฆืžื™ื—ื” ืžืขื•ื›ื‘ืช.
10:55
His gait suffered with a dragging of his left foot,
189
655318
2544
ื”ื”ืœื™ื›ื” ืฉืœื• ืกื‘ืœื” ืžื’ืจื™ืจืช ื›ืฃ ื”ืจื’ืœ ื”ืฉืžืืœื™ืช,
10:57
he had severe headaches.
190
657862
1918
ื”ื™ื• ืœื• ื›ืื‘ื™ ืจืืฉ ืงืฉื™ื.
10:59
He went to 17 doctors over three years.
191
659780
3337
ื”ื•ื ื”ืœืš ืœ-17 ืจื•ืคืื™ื ื‘ืžืฉืš ืฉืœื•ืฉ ืฉื ื™ื.
11:03
His mother then entered all his symptoms into ChatGPT.
192
663743
4254
ื•ืื– ืืžื• ื”ื›ื ื™ืกื” ืืช ื›ืœ ื”ืชืกืžื™ื ื™ื ืฉืœื• ืœืฆ'ืื˜-ื’'ื™-ืคื™-ื˜ื™,
11:08
It made the diagnosis of occulta spina bifida,
193
668706
4254
ืฉืื‘ื—ืŸ ืื•ืชื• ื‘ืฉื“ืจื” ืฉืกื•ืขื” ื ืกืชืจืช,
11:12
which meant he had a tethered spinal cord that was missed by all 17 doctors
194
672960
5297
ื›ืœื•ืžืจ, ื—ื•ื˜ ืฉื“ืจื” ืงืจื•ืข, ืฉื›ืœ 17 ื”ืจื•ืคืื™ื ื”ื—ืžื™ืฆื•
11:18
over three years.
195
678257
1168
ื‘ืžืฉืš ืฉืœื•ืฉ ืฉื ื™ื.
11:19
He had surgery to release the cord.
196
679467
2002
ื”ื•ื ืขื‘ืจ ื ื™ืชื•ื— ืœืฉื—ืจื•ืจ ื—ื•ื˜ ื”ืฉื“ืจื”.
11:21
He's now perfectly healthy.
197
681469
1793
ื•ืขื›ืฉื™ื• ื”ื•ื ื‘ืจื™ื ืœื—ืœื•ื˜ื™ืŸ.
11:24
(Applause)
198
684889
5630
(ืžื—ื™ืื•ืช ื›ืคื™ื™ื)
11:30
This is a patient that was sent to me,
199
690561
2920
ื–ื• ืžื˜ื•ืคืœืช ืฉื ืฉืœื—ื” ืืœื™,
11:33
who was suffering with, she was told, long COVID.
200
693481
4671
ืฉืกื‘ืœื”, ื›ืš ื ืืžืจ ืœื”, ืžืชืกืžื•ื ืช ืคื•ืกื˜-ืงื•ืจื•ื ื”.
11:38
She saw many different physicians, neurologists,
201
698694
3379
ื”ื™ื ืจืืชื” ื›ืœ ืžื™ื ื™ ืจื•ืคืื™ื, ื ื•ื™ืจื•ืœื•ื’ื™ื,
11:42
and her sister entered all her symptoms after getting nowhere,
202
702073
4546
ื•ืื—ื•ืชื” ื”ื›ื ื™ืกื” ืืช ื›ืœ ื”ืชืกืžื™ื ื™ื ืฉืœื” ืœืื—ืจ ืฉื”ื’ื™ืขื” ืœืžื‘ื•ื™ ืกืชื•ื --
11:46
no treatment for long COVID,
203
706619
1418
ืื™ืŸ ื˜ื™ืคื•ืœ ื‘ืคื•ืกื˜-ืงื•ืจื•ื ื”, ืื™ืŸ ืฉื•ื ื˜ื™ืคื•ืœ ืžืื•ืžืช --
11:48
there is no treatment validated,
204
708079
1710
11:49
and her sister put all her symptoms into ChatGPT.
205
709789
4421
ืื—ื•ืชื” ื”ื›ื ื™ืกื” ืืช ื›ืœ ื”ืกื™ืžืคื˜ื•ืžื™ื ืฉืœื” ืœืฆ'ืื˜-ื’'ื™-ืคื™-ื˜ื™,
11:54
It found out it actually was not long COVID,
206
714252
2293
ื•ื”ืชื‘ืจืจ ืฉื–ื” ืœืžืขืฉื” ืœื ืคื•ืกื˜-ืงื•ืจื•ื ื”,
11:56
she had limbic encephalitis, which is treatable.
207
716587
3462
ืืœื ื“ืœืงืช ืžื•ื— ืœื™ืžื‘ื™ืช, ืฉื‘ื” ื ื™ืชืŸ ืœื˜ืคืœ.
12:00
She was treated, and now she's doing extremely well.
208
720091
3128
ื”ื™ื ื˜ื•ืคืœื”, ื•ืขื›ืฉื™ื• ื”ื™ื ื‘ืจื™ืื” ืœื’ืžืจื™.
12:03
But these are not just anecdotes anymore.
209
723594
2753
ืื‘ืœ ื›ื‘ืจ ืœื ืžื“ื•ื‘ืจ ื‘ืงื•ืจื™ื•ื–ื™ื ื‘ืœื‘ื“.
12:06
70 very difficult cases
210
726389
3461
70 ืžืงืจื™ื ืงืฉื™ื ืžืื•ื“,
12:09
that are the clinical pathologic conferences
211
729850
2461
ืฉื ื“ื•ื ื• ื‘ื•ื•ืขื™ื“ื•ืช ื”ืคืชื•ืœื•ื’ื™ื•ืช ืงืœื™ื ื™ื•ืช,
ืขืœ ืคื™ "ื ื™ื•-ืื™ื ื’ืœื ื“ ื’'ื•ืจื ืœ ืื•ืฃ ืžื“ื™ืกื™ืŸ",
12:12
at the New England Journal of Medicine
212
732353
1877
12:14
were compared to GPT-4,
213
734272
2836
ื”ื•ืฉื•ื• ืœื’'ื™-ืคื™-ื˜ื™ 4,
12:17
and the chatbot did as well
214
737149
3295
ื•ื”ืฆ'ืื˜ื‘ื•ื˜ ื”ืฆืœื™ื— ื‘ืื•ืชื” ืžื™ื“ื”
12:20
or better than the expert master clinicians
215
740486
3295
ืื• ื˜ื•ื‘ ื™ื•ืชืจ ืžื”ืจื•ืคืื™ื ื”ืžื•ืžื—ื™ื
12:23
in making the diagnosis.
216
743781
1960
ื‘ืื‘ื—ื•ืŸ.
12:26
So I just want to close with a recent conversation with my fellow.
217
746492
4713
ืื– ืื ื™ ืจืง ืจื•ืฆื” ืœืกื™ื™ื ื‘ืฉื™ื—ื” ืฉื”ื™ืชื” ืœื™ ืœืื—ืจื•ื ื” ืขื ื”ื—ื‘ืจ ืฉืœื™.
12:31
Medicine is still an apprenticeship,
218
751706
2085
ื”ืจืคื•ืื” ื”ื™ื ืขื“ื™ื™ืŸ ื”ืชืžื—ื•ืช,
12:33
and Andrew Cho is 30 years old,
219
753833
3837
ื•ืื ื“ืจื• ืฆ'ื• ื”ื•ื ื‘ืŸ 30,
12:37
in his second year of cardiology fellowship.
220
757670
2085
ื‘ืฉื ื” ื”ืฉื ื™ื™ื” ืฉืœื• ื‘ืงืจื“ื™ื•ืœื•ื’ื™ื”.
12:39
We see all patients together in the clinic.
221
759797
2669
ืฉื ื™ื ื• ืจื•ืื™ื ื™ื—ื“ ืืช ื›ืœ ื”ื—ื•ืœื™ื ื‘ืžืจืคืื”.
12:42
And at the end of clinic the other day,
222
762967
2252
ื•ื‘ืกื•ืฃ ื”ื™ื•ื ื‘ืžืจืคืื”, ืœืคื ื™ ื›ืžื” ื™ืžื™ื,
12:45
I sat down and said to him,
223
765261
1918
ื”ืชื™ื™ืฉื‘ืชื™ ื•ืืžืจืชื™ ืœื•, "ืื ื“ืจื•, ืืชื” ื›ืœ ื›ืš ื‘ืจ ืžื–ืœ.
12:47
"Andrew, you are so lucky.
224
767179
2795
12:50
You're going to be practicing medicine in an era of keyboard liberation.
225
770516
4838
"ืืชื” ืขืชื™ื“ ืœืขืกื•ืง ื‘ืจืคื•ืื” ื‘ืขื™ื“ืŸ ืฉืœ ืฉื—ืจื•ืจ ืžืงืœื“ืช.
12:55
You're going to be connecting with patients
226
775813
2044
"ืืชื” ืชื™ืฆื•ืจ ืงืฉืจ ื™ืฉื™ืจ ืขื ืžื˜ื•ืคืœื™ื
12:57
the way we haven't done for decades."
227
777857
2502
"ืžื” ืฉืœื ืขืฉื™ื ื• ื‘ืžืฉืš ืขืฉืจื•ืช ืฉื ื™ื.
13:00
That is the ability to have the note
228
780735
3086
"ื›ืœื•ืžืจ, ืชื•ื›ืœ ืœืงื‘ืœ ื”ืขืจื•ืช ื•ืœื“ืขืช ืžื” ืœืขืฉื•ืช
13:03
and the work from the conversation
229
783863
2502
"ื™ืฉื™ืจื•ืช ืžื”ืฉื™ื—ื” ืขื ื”ืžื˜ื•ืคืœ
13:06
to derive things like pre-authorization,
230
786407
3795
"ื›ื“ื™ ืœื”ื–ื™ื– ื“ื‘ืจื™ื ื›ืžื• ืื™ืฉื•ืจ ืžืจืืฉ,
13:10
billing, prescriptions, future appointments --
231
790202
4755
"ื—ื™ื•ื‘, ืžืจืฉืžื™ื, ืคื’ื™ืฉื•ืช ืขืชื™ื“ื™ื•ืช -
13:14
all the things that we do,
232
794999
1293
"ื›ืœ ื”ื“ื‘ืจื™ื ืฉืื ื—ื ื• ืขื•ืฉื™ื, ื›ื•ืœืœ ืขื™ื“ื•ื“ ื”ืžื˜ื•ืคืœ,
13:16
including nudges to the patient.
233
796334
1584
13:17
For example, did you get your blood pressure checks
234
797918
2461
"ื›ืžื•, ื”ืื ืงื™ื‘ืœืช ืืช ื‘ื“ื™ืงื•ืช ืœื—ืฅ ื”ื“ื ืฉืœืš,
13:20
and what did they show
235
800421
1168
"ื•ืžื” ื”ืŸ ื”ืจืื•?
13:21
and all that coming back to you.
236
801630
1544
"ืืชื” ืชืงื‘ืœ ืืช ื›ืœ ื–ื”,
13:23
But much more than that,
237
803215
1710
"ืื‘ืœ ื”ืจื‘ื” ื™ื•ืชืจ ืžื–ื”,
13:24
to help with making diagnoses.
238
804925
2086
"ื”ืขื–ืจื” ื‘ืื‘ื—ื•ืŸ ื•ืžืชื ืช ื”ื–ืžืŸ
13:27
And the gift of time
239
807928
2002
13:29
that having all the data of a patient
240
809972
2169
"ื›ืฉื›ืœ ื”ื ืชื•ื ื™ื ืฉืœ ื”ืžื˜ื•ืคืœ ื–ืžื™ื ื™ื,
13:32
that's all teed up before even seeing the patient.
241
812183
2961
"ืฉื›ื•ืœื ื ืืกืคื• ืขื•ื“ ืœืคื ื™ ืฉืืชื” ื‘ื›ืœืœ ืคื•ื’ืฉ ื‘ืžื˜ื•ืคืœ.
13:35
And all this support changes the future of the patient-doctor relationship,
242
815144
6632
"ื•ื›ืœ ื”ืชืžื™ื›ื” ื”ื–ื• ืžืฉื ื” ืืช ืขืชื™ื“ ืžืขืจื›ืช ื”ื™ื—ืกื™ื ื‘ื™ืŸ ืžื˜ื•ืคืœ ืœืจื•ืคื,
13:41
bringing in the gift of time.
243
821776
2460
"ื•ืžืขื ื™ืงื” ืืช ืžืชื ืช ื”ื–ืžืŸ.
13:44
So this is really exciting.
244
824612
1710
"ืื– ื–ื” ื‘ืืžืช ืžืจื’ืฉ",
13:46
I said to Andrew, everything has to be validated, of course,
245
826364
4295
ืืžืจืชื™ ืœืื ื“ืจื• -- ื•ื”ื›ืœ ืฆืจื™ืš ืœื”ื™ื•ืช ืžืื•ืžืช, ื›ืžื•ื‘ืŸ --
13:50
that the benefit greatly outweighs any risk.
246
830701
3796
ืฉื”ื™ืชืจื•ืŸ ืขื•ืœื” ื‘ื”ืจื‘ื” ืขืœ ื›ืœ ืกื™ื›ื•ืŸ.
13:54
But it is really a remarkable time for the future of health care,
247
834538
4505
ืื‘ืœ ื–ื” ื‘ืืžืช ื–ืžืŸ ืžื“ื”ื™ื,
ื•ืžื‘ื—ื™ื ืช ืขืชื™ื“ ื”ื‘ืจื™ืื•ืช ื–ื” ื›ืœ ื›ืš ืžืจื’ืฉ, ืœืขื–ืื–ืœ.
13:59
it's so damn exciting.
248
839085
2544
14:01
Thank you.
249
841962
1168
ืชื•ื“ื” ืœื›ื.
14:03
(Applause)
250
843172
2753
(ืžื—ื™ืื•ืช ื›ืคื™ื™ื)
ืขืœ ืืชืจ ื–ื”

ืืชืจ ื–ื” ื™ืฆื™ื’ ื‘ืคื ื™ื›ื ืกืจื˜ื•ื ื™ YouTube ื”ืžื•ืขื™ืœื™ื ืœืœื™ืžื•ื“ ืื ื’ืœื™ืช. ืชื•ื›ืœื• ืœืจืื•ืช ืฉื™ืขื•ืจื™ ืื ื’ืœื™ืช ื”ืžื•ืขื‘ืจื™ื ืขืœ ื™ื“ื™ ืžื•ืจื™ื ืžื”ืฉื•ืจื” ื”ืจืืฉื•ื ื” ืžืจื—ื‘ื™ ื”ืขื•ืœื. ืœื—ืฅ ืคืขืžื™ื™ื ืขืœ ื”ื›ืชื•ื‘ื™ื•ืช ื‘ืื ื’ืœื™ืช ื”ืžื•ืฆื’ื•ืช ื‘ื›ืœ ื“ืฃ ื•ื™ื“ืื• ื›ื“ื™ ืœื”ืคืขื™ืœ ืืช ื”ืกืจื˜ื•ืŸ ืžืฉื. ื”ื›ืชื•ื‘ื™ื•ืช ื’ื•ืœืœื•ืช ื‘ืกื ื›ืจื•ืŸ ืขื ื”ืคืขืœืช ื”ื•ื•ื™ื“ืื•. ืื ื™ืฉ ืœืš ื”ืขืจื•ืช ืื• ื‘ืงืฉื•ืช, ืื ื ืฆื•ืจ ืื™ืชื ื• ืงืฉืจ ื‘ืืžืฆืขื•ืช ื˜ื•ืคืก ื™ืฆื™ืจืช ืงืฉืจ ื–ื”.

https://forms.gle/WvT1wiN1qDtmnspy7