WEBVTT

00:00:00.166 --> 00:00:01.126
Thomas, welcome.

00:00:01.501 --> 00:00:07.799
Can you give me some examples of
digital, AI, GenAI, that you

00:00:07.799 --> 00:00:10.427
think have the most use in
health care?

00:00:10.427 --> 00:00:11.845
Broadly in health care,

00:00:11.845 --> 00:00:15.932
I'm convinced that AI it
will help making better decisions.

00:00:15.932 --> 00:00:18.935
So in the not-too-distant
future, we will see the life of

00:00:18.935 --> 00:00:22.188
physicians changing dramatically
because there will be GenAI.

00:00:22.522 --> 00:00:24.566
So we'll see validated GenAI

00:00:24.566 --> 00:00:28.611
being assisting physicians, and
I suspect in a not-too-distant

00:00:28.611 --> 00:00:32.282
future working more directly
involved in decision-making

00:00:32.282 --> 00:00:36.119
diagnostics, leaving the human
interactions to the doctors.

00:00:36.578 --> 00:00:39.706
So that's one way. In the
pharmaceutical industry, we will see

00:00:39.706 --> 00:00:41.916
we are already seeing AI drive

00:00:41.916 --> 00:00:45.920
enormous acceleration
in thenprocesses.

00:00:45.920 --> 00:00:49.466
So from productivity point of
view, but also from a scientific

00:00:49.466 --> 00:00:52.635
insight point of view, quality
by design point of view.

00:00:52.969 --> 00:00:56.056
So we see AI already being
deployed in our company.

00:00:56.056 --> 00:00:58.683
We already deployed broadly, but
I expect it will be accelerated

00:00:58.683 --> 00:01:04.481
significantly in the next 6, 8,
12, 18 months.

00:01:05.023 --> 00:01:07.150
And that's just an R&D, but
we'll see the same in

00:01:07.150 --> 00:01:09.611
production, in other parts of
the industry.

00:01:09.611 --> 00:01:13.782
Where you're seeing the most success...

00:01:13.782 --> 00:01:15.158
Are there themes?

00:01:15.909 --> 00:01:20.038
It really depends on what
success is defined as because,

00:01:20.038 --> 00:01:24.292
because you can say we
already see it in report writing.

00:01:24.292 --> 00:01:28.004
So we do no longer have human
beings writing the study reports

00:01:28.004 --> 00:01:29.380
of our clinical trials.

00:01:29.380 --> 00:01:30.715
It's all done by AI.

00:01:31.007 --> 00:01:34.052
So that's an example of very
successful implementation and

00:01:34.052 --> 00:01:37.055
scaling, and we're expanding
that into other parts of the

00:01:37.055 --> 00:01:38.932
what we call the
regulatory dossier.

00:01:39.849 --> 00:01:42.894
Another example is in deployment
and code generation.

00:01:42.894 --> 00:01:46.481
So we basically, will gradually
see no need for starting to code

00:01:46.481 --> 00:01:47.482
manually yourself.

00:01:48.066 --> 00:01:49.943
Image analysis--very successful.

00:01:49.943 --> 00:01:53.530
You know, we, we do know that AI
is better at looking at images

00:01:53.530 --> 00:01:56.282
than radiologists, quite
frankly, in many cases.

00:01:56.658 --> 00:02:00.328
And again, it goes much faster
adjudication processes, which is

00:02:00.328 --> 00:02:03.581
during very large clinical
trials, if you have endpoints

00:02:03.581 --> 00:02:07.210
that need to be adjudicated by
physicians where we are building

00:02:07.210 --> 00:02:10.672
AI algorithms that can do that.
And the list goes on . . .

00:02:10.672 --> 00:02:12.799
So that's just in development.
In research you see it in

00:02:12.799 --> 00:02:16.219
molecule design, target
discovery, protein-folding

00:02:16.219 --> 00:02:20.098
predictions, which are very
complex, and then simulations of

00:02:20.098 --> 00:02:22.642
clinical trials and simulations
of processes.

00:02:22.934 --> 00:02:25.812
And then there are lots of other
process automation errors.

00:02:25.812 --> 00:02:30.191
So the list is really long and
success is defined in many ways:

00:02:30.191 --> 00:02:34.112
is scientific insight is
success, increased productivity

00:02:34.112 --> 00:02:37.282
is success, you know,
reduction of time spent is success.

00:02:37.282 --> 00:02:39.033
So I wouldn't pick one area.

00:02:40.410 --> 00:02:44.497
I was speaking about digital and
AI, but what else, what other

00:02:44.497 --> 00:02:48.585
trends, innovations do you see
really shaping this year ahead?

00:02:48.585 --> 00:02:52.088
From a digital point of view or
an AI point of view,

00:02:52.088 --> 00:02:57.886
it's the enormous development in
generative AI that we keep seeing.

00:02:58.094 --> 00:03:00.722
Actually, half a year ago, I
would have thought we were

00:03:00.722 --> 00:03:03.558
plateauing a bit, but I can say
that's not going to happen.

00:03:03.892 --> 00:03:04.976
We now see the GenAI

00:03:04.976 --> 00:03:09.564
models developing very rapidly
into, into areas where you start

00:03:09.564 --> 00:03:13.943
seeing something which would be
closer to real intelligence.

00:03:14.194 --> 00:03:18.740
So reasoning models that really
now start behaving very differently.

00:03:19.115 --> 00:03:22.410
And . . .  and it's . . . it's
just amazing the speed of that.

00:03:22.493 --> 00:03:24.871
It's also a challenge for us in
the regulated area.

00:03:25.163 --> 00:03:28.708
How do we keep that balance
between the pace of model

00:03:28.708 --> 00:03:31.836
development and being able to
validate and say this is good

00:03:31.836 --> 00:03:35.632
enough also versus regulators
and legislation?

00:03:35.632 --> 00:03:37.967
How do you strike that balance?

00:03:37.967 --> 00:03:42.680
We strike that balance by having
built a very good AI governance

00:03:42.680 --> 00:03:44.557
set-up first and foremost.

00:03:44.557 --> 00:03:48.019
We do not just deploy everything
into production so that we have

00:03:48.019 --> 00:03:49.979
experimentation--that's
good enough.

00:03:50.230 --> 00:03:52.815
But if it goes into production,
it has to go through our

00:03:52.815 --> 00:03:54.984
governance, which is . . .
includes a lot of technological

00:03:54.984 --> 00:03:58.655
checks and balances, but also
ethical and legal checks and balances.

00:03:58.655 --> 00:03:59.781
And that's one way of doing it.

00:04:00.073 --> 00:04:03.493
But I don't have the full answer
yet because it's not a something

00:04:03.493 --> 00:04:05.036
that there isn't a manual for

00:04:05.036 --> 00:04:05.954
how do you balance that.

00:04:05.954 --> 00:04:07.163
Thomas, thank you so much.

00:04:07.163 --> 00:04:08.122
An absolute pleasure.

00:04:08.331 --> 00:04:09.165
Likewise, Thank you.