WEBVTT
1
00:00:00.260 --> 00:00:02.677
(soft music)
2
00:00:04.830 --> 00:00:05.850
2024 is going to be
3
00:00:05.850 --> 00:00:07.860
an exciting one for generative AI.
4
00:00:07.860 --> 00:00:09.270
My name is Ashkan Afkhami.
5
00:00:09.270 --> 00:00:10.500
I'm a managing director and partner,
6
00:00:10.500 --> 00:00:11.970
and I look after our healthcare practice
7
00:00:11.970 --> 00:00:13.500
from a digital analytics perspective.
8
00:00:13.500 --> 00:00:15.240
And I'll be moderating today's panel.
9
00:00:15.240 --> 00:00:18.150
People are getting ready for an exciting 2024.
10
00:00:18.150 --> 00:00:20.790
What are you most excited about in 2024 on this topic?
11
00:00:20.790 --> 00:00:23.280
Last year was all about understanding generative AI.
12
00:00:23.280 --> 00:00:26.310
I think we've done a great job collectively as an industry
13
00:00:26.310 --> 00:00:29.400
understanding the potential opportunities for generative AI.
14
00:00:29.400 --> 00:00:31.890
The area that I'm particularly excited
15
00:00:31.890 --> 00:00:34.290
is large vision models and multimodal model,
16
00:00:34.290 --> 00:00:37.470
because healthcare is predominantly multimodal images.
17
00:00:37.470 --> 00:00:41.010
How do we combine the EMR data
18
00:00:41.010 --> 00:00:42.570
with the medical imaging data,
19
00:00:42.570 --> 00:00:45.450
with the data that's coming from monitoring
20
00:00:45.450 --> 00:00:46.283
in meaningful way?
21
00:00:46.283 --> 00:00:48.930
That's where GenAI can really help streamline the data
22
00:00:48.930 --> 00:00:52.350
and then different persona users can query the same data
23
00:00:52.350 --> 00:00:53.520
to get meaningful insights.
24
00:00:53.520 --> 00:00:55.350
And there is massive opportunity
25
00:00:55.350 --> 00:00:57.390
when you see all the data in biology
26
00:00:57.390 --> 00:00:59.250
to use the technology of GenAI.
27
00:00:59.250 --> 00:01:00.240
Once you have enough data,
28
00:01:00.240 --> 00:01:02.910
you can, from the data, train models
29
00:01:02.910 --> 00:01:06.000
that capture information in the data.
30
00:01:06.000 --> 00:01:08.580
As companies become more familiar with GenAI,
31
00:01:08.580 --> 00:01:11.340
they're learning how to prioritize use cases.
32
00:01:11.340 --> 00:01:14.460
The first use cases are focusing on automation.
33
00:01:14.460 --> 00:01:16.080
We'll start with operational efficiency.
34
00:01:16.080 --> 00:01:18.840
How do we improve better bedside prediction,
35
00:01:18.840 --> 00:01:21.180
streamlining the scheduling, the EMR,
36
00:01:21.180 --> 00:01:22.830
reducing the product development cycle?
37
00:01:22.830 --> 00:01:24.990
There's a real ROI where a lot of things
38
00:01:24.990 --> 00:01:26.670
which could have taken five to 10 years
39
00:01:26.670 --> 00:01:28.290
can actually be done in one year.
40
00:01:28.290 --> 00:01:31.830
First, a particular single use case, start with security.
41
00:01:31.830 --> 00:01:33.960
Because ultimately, if you can work
42
00:01:33.960 --> 00:01:35.880
with your security teams to improve security,
43
00:01:35.880 --> 00:01:37.860
then you're actually going to build a solid foundation
44
00:01:37.860 --> 00:01:39.630
for how you build services on top of that-
45
00:01:39.630 --> 00:01:41.670
your data platform, your AI platform,
46
00:01:41.670 --> 00:01:43.200
your experience platforms.
47
00:01:43.200 --> 00:01:45.270
GenAI can also play an important role
48
00:01:45.270 --> 00:01:47.610
in improving the customer experience.
49
00:01:47.610 --> 00:01:49.350
We do truly believe that generative AI
50
00:01:49.350 --> 00:01:52.110
can actually reinvent customer engagement.
51
00:01:52.110 --> 00:01:54.810
Can we now use technology to augment that experience
52
00:01:54.810 --> 00:01:55.860
to make it better, richer?
53
00:01:55.860 --> 00:01:58.650
Can we move down the intelligent self-service route
54
00:01:58.650 --> 00:02:00.510
so patients or members can actually
55
00:02:00.510 --> 00:02:02.130
find out more information?
56
00:02:02.130 --> 00:02:04.110
And then on the other side, inside of the organization,
57
00:02:04.110 --> 00:02:06.690
can we equip the people in the organization
58
00:02:06.690 --> 00:02:08.100
to operate at the top of their license
59
00:02:08.100 --> 00:02:10.440
so they can spend more time with the patient
60
00:02:10.440 --> 00:02:12.720
rather than actually looking at the screen?
61
00:02:12.720 --> 00:02:13.980
Healthcare companies are facing
62
00:02:13.980 --> 00:02:15.660
a second major challenge.
63
00:02:15.660 --> 00:02:18.480
How do you scale GenAI use cases successfully?
64
00:02:18.480 --> 00:02:20.850
There are probably four areas that I'd kind of highlight.
65
00:02:20.850 --> 00:02:22.770
The first one is digital skills,
66
00:02:22.770 --> 00:02:24.690
making sure we're providing appropriate readiness
67
00:02:24.690 --> 00:02:25.980
at all layers of the organization
68
00:02:25.980 --> 00:02:27.720
to understand the opportunity,
69
00:02:27.720 --> 00:02:29.310
but also the risks and limitations.
70
00:02:29.310 --> 00:02:32.340
The second then is really having a well-defined approach
71
00:02:32.340 --> 00:02:33.600
for use case ideation.
72
00:02:33.600 --> 00:02:36.300
And so every use case that you will ideate on
73
00:02:36.300 --> 00:02:38.070
is validated in the same way.
74
00:02:38.070 --> 00:02:40.170
The third area is really taking a platform approach
75
00:02:40.170 --> 00:02:41.730
to the technology adoption.
76
00:02:41.730 --> 00:02:43.800
Making sure you're building an appropriate platform
77
00:02:43.800 --> 00:02:45.210
with the appropriate guardrails
78
00:02:45.210 --> 00:02:46.320
and the appropriate governance
79
00:02:46.320 --> 00:02:48.660
and organizational frameworks in place
80
00:02:48.660 --> 00:02:51.720
so you can actually scale fast once you're ready to go
81
00:02:51.720 --> 00:02:53.460
and you've approved those use cases.
82
00:02:53.460 --> 00:02:55.680
And then the fourth is the importance of partnerships.
83
00:02:55.680 --> 00:02:57.570
The way we develop AI solutions
84
00:02:57.570 --> 00:03:00.300
for drug discovery, et cetera,
85
00:03:00.300 --> 00:03:02.940
is that we need expertise and we need data.
86
00:03:02.940 --> 00:03:04.830
For our strategy to access the data,
87
00:03:04.830 --> 00:03:07.740
it's great to partner with hospitals, medical centers,
88
00:03:07.740 --> 00:03:09.530
and this is crucial because you need good data
89
00:03:09.530 --> 00:03:11.250
in order to train the model.
90
00:03:11.250 --> 00:03:13.020
Partnering with hospitals requires
91
00:03:13.020 --> 00:03:16.410
a long-term relationship, trust building, and technology
92
00:03:16.410 --> 00:03:18.660
to make sure that the data is not leaked,
93
00:03:18.660 --> 00:03:22.080
that there is no personal information leaking.
94
00:03:22.080 --> 00:03:24.125
It's the importance of building these connections
95
00:03:24.125 --> 00:03:25.745
because ultimately, there's
96
00:03:25.745 --> 00:03:27.360
not going to be a single frontier model
97
00:03:27.360 --> 00:03:29.293
or a open source model that's going to be able
98
00:03:29.293 --> 00:03:31.107
to solve all of the challenges in healthcare.
99
00:03:31.107 --> 00:03:34.380
You do need partners across the data, the domain,
100
00:03:34.380 --> 00:03:37.380
as well as partners, both in the ISV world,
101
00:03:37.380 --> 00:03:39.990
partners that actually sit in front of those data estates,
102
00:03:39.990 --> 00:03:42.270
as well as partners that actually do the implementation.
103
00:03:42.270 --> 00:03:45.180
We have been heavily investing around building
104
00:03:45.180 --> 00:03:47.760
what's called a medical imaging foundation models.
105
00:03:47.760 --> 00:03:50.190
So rather than having a smaller model,
106
00:03:50.190 --> 00:03:52.500
if you scale it and then bring it down,
107
00:03:52.500 --> 00:03:54.210
it actually gives you more accuracy.
108
00:03:54.210 --> 00:03:56.220
That smaller model we can even embed
109
00:03:56.220 --> 00:03:57.690
into our ultrasound devices.
110
00:03:57.690 --> 00:03:59.913
As scale is important,
111
00:04:01.020 --> 00:04:03.030
as with traditional AI and machine learning,
112
00:04:03.030 --> 00:04:04.410
data is equally important.
113
00:04:04.410 --> 00:04:07.110
It's possible that your data size might decrease,
114
00:04:07.110 --> 00:04:09.240
but you need to ensure the quality of the data,
115
00:04:09.240 --> 00:04:11.760
the governance around that is there.
116
00:04:11.760 --> 00:04:13.320
It also helps you reduce the cost.
117
00:04:13.320 --> 00:04:16.290
There's a huge opportunity collectively as an industry
118
00:04:16.290 --> 00:04:17.760
for us to start talking about
119
00:04:17.760 --> 00:04:19.650
where are we seeing value return
120
00:04:19.650 --> 00:04:21.990
and can we replicate that for the benefit of society
121
00:04:21.990 --> 00:04:24.928
rather than the benefit of our own individual organizations?
122
00:04:24.928 --> 00:04:27.345
(soft music)