WEBVTT

00:00:04.277 --> 00:00:09.057
Machine learning algorithms using deep
learning are as good as radiologists

00:00:09.207 --> 00:00:15.655
in analysing images, for example,
mammograms or to detect breast cancer

00:00:15.925 --> 00:00:21.975
or other forms of CT scans or MRI
scans for early cancer detection.

00:00:22.435 --> 00:00:26.055
Performance of machine learning
algorithms are equivalent to

00:00:26.065 --> 00:00:28.475
those of the best human experts.

00:00:28.985 --> 00:00:31.844
The interesting thing is
with these new techniques.

00:00:32.284 --> 00:00:33.514
We can go further.

00:00:35.013 --> 00:00:39.413
Welcome to Oxford+, the podcast series
that takes you deep into the myths and

00:00:39.413 --> 00:00:41.623
truths of the Oxford investing landscape.

00:00:42.233 --> 00:00:45.343
I'm your host, Susannah de
Jager, and I've spent over 16

00:00:45.343 --> 00:00:46.843
years in UK asset management.

00:00:47.643 --> 00:00:51.199
My guest today is Lord
Lionel Tarassenko, CBE.

00:00:51.490 --> 00:00:54.650
Lionel is a world leading expert
in the application of signalling

00:00:54.650 --> 00:00:58.350
processing, artificial intelligence
and machine learning to healthcare,

00:00:58.450 --> 00:01:01.590
with a strong track record in
translation to clinical medicine.

00:01:02.020 --> 00:01:05.630
Professor Tarassenko's work has had
a major impact on the identification

00:01:05.640 --> 00:01:09.774
of deterioration in acute care and
on the management of chronic disease.

00:01:09.935 --> 00:01:12.985
Lionel was born in Paris in 1957.

00:01:13.115 --> 00:01:17.475
He received a BA in Engineering
Science in 1978 and the DPhil in

00:01:17.485 --> 00:01:22.315
Medical Electronics in 1985, both
from the University of Oxford.

00:01:22.705 --> 00:01:26.715
After a period in industry working for
Rakel Electronics, he was appointed

00:01:26.875 --> 00:01:32.205
University Lecturer and Tutorial Fellow
in Oxford, St. Hugh's College in 1988.

00:01:32.625 --> 00:01:36.610
He was elected to the Chair of Electrical
Engineering, and to a professional

00:01:36.610 --> 00:01:39.310
fellowship at St. John's College in 1997.

00:01:39.910 --> 00:01:43.350
He was the driving force behind the
creation of the Institute of Biomedical

00:01:43.350 --> 00:01:48.610
Engineering, which he directed from
its opening in 2008 to October 2012.

00:01:48.780 --> 00:01:52.130
Lionel was elected to the Fellowship
of the Royal Academy of Engineering

00:01:52.130 --> 00:01:57.090
in 2000 and to the Fellowship of the
Academy of Medical Sciences in 2013.

00:01:57.310 --> 00:02:00.430
He is a Director of the University's
wholly owned Technology Transfer Centre.

00:02:01.450 --> 00:02:04.260
Oxford University Innovation
and has founded four

00:02:04.260 --> 00:02:05.780
university spin out companies.

00:02:06.300 --> 00:02:10.420
He was the head of the Department
of Engineering Science from 2014

00:02:10.490 --> 00:02:14.560
to 2019 and is now the founding
president of Reuben College, the

00:02:14.560 --> 00:02:16.600
University of Oxford's newest college.

00:02:17.020 --> 00:02:20.426
He was editor in chief for the Topol
Review, preparing the healthcare

00:02:20.426 --> 00:02:22.596
workforce to deliver the digital future.

00:02:22.926 --> 00:02:26.486
Lionel Tarassenko is at the core of
translational science in the University

00:02:26.486 --> 00:02:30.196
of Oxford, and with his new leadership
of Rubin College, where we are

00:02:30.196 --> 00:02:33.956
recording today, I'm very privileged
to be able to discuss his views on

00:02:33.956 --> 00:02:36.106
how he sees Oxford evolving further.

00:02:36.412 --> 00:02:38.792
Lionel, thank you for joining me today.

00:02:38.897 --> 00:02:41.038
Glad to be able to  chat to you today.

00:02:41.088 --> 00:02:47.154
So, you have an extraordinary history
with Oxford, I'd love to hear some of

00:02:47.154 --> 00:02:53.014
your experience where you've translated
research out of the lab and into a

00:02:53.014 --> 00:02:54.814
practical application in the real world.

00:02:54.909 --> 00:02:59.849
Yes, well, I've been very fortunate to
be in the right place at the right time.

00:03:00.439 --> 00:03:05.826
I was appointed as a professor back
in 1988, then elected to the chair of

00:03:05.826 --> 00:03:11.706
electrical engineering in 1997 and that
election was partly due to the fact that

00:03:11.746 --> 00:03:16.296
I'd done some translational work with
established companies, with Rolls Royce,

00:03:16.676 --> 00:03:21.102
with Sharp, who were on the founding
company of the Oxford Science Park in

00:03:21.102 --> 00:03:27.982
the 1990s, but after the appointment
to the chair in 1997, instead of

00:03:27.982 --> 00:03:32.772
working with established companies, I
got involved in setting up spin outs.

00:03:32.782 --> 00:03:38.062
In fact, my first spin out was in 1999
with two ex students of mine who'd

00:03:38.062 --> 00:03:41.812
ended up in Cambridge, and they came
back to Oxford so that we could do a

00:03:41.822 --> 00:03:46.307
spin out together and at that time when
you were doing a spin out in Oxford,

00:03:46.757 --> 00:03:48.617
it was still slightly frowned upon.

00:03:48.967 --> 00:03:53.917
Then within a decade it was tolerated
and now it's positively encouraged.

00:03:54.177 --> 00:03:59.107
So I've been through all the stages
from colleagues sort of frowning

00:03:59.107 --> 00:04:03.813
upon me doing all this kind of
commercial activity to now saying,

00:04:04.183 --> 00:04:05.843
how can you help us do the same?

00:04:06.628 --> 00:04:10.708
So that's been a very interesting kind
of progression in the last 25 years.

00:04:10.998 --> 00:04:16.228
But the reason why I did it in 1999
and did the latest one a decade ago

00:04:16.618 --> 00:04:23.338
was really as an engineer, passionate
about translating from the lab into a

00:04:23.338 --> 00:04:27.213
domain where it can be used by as many
people as possible and the two areas

00:04:27.213 --> 00:04:30.953
I've worked in was further work with
Rolls Royce on engine health monitoring.

00:04:31.503 --> 00:04:36.643
So, if you've flown on an Airbus plane
recently, you're very likely to have been

00:04:36.673 --> 00:04:41.463
monitored by software that was developed
by one of the spin-out companies until

00:04:41.463 --> 00:04:42.963
it was brought back in by Rolls Royce.

00:04:42.983 --> 00:04:46.903
They got slightly concerned that they
were relying on this software, but it

00:04:46.903 --> 00:04:52.003
was an SME and SMEs do go down as well
as up, so they thought it'd be safer

00:04:52.003 --> 00:04:54.333
to bring it back into Rolls Royce.

00:04:54.433 --> 00:04:59.543
But mostly, since those activities
finished, it's been mostly healthcare

00:04:59.903 --> 00:05:05.283
and I'm passionate because what it means,
its not just a research paper, it's not

00:05:05.283 --> 00:05:08.523
just the results of a clinical trial,
it's something that makes a difference

00:05:08.603 --> 00:05:13.063
on the hospital ward or in somebody's
home as they try to manage their diabetes

00:05:13.063 --> 00:05:16.633
and so on and that's what's really
driven my passion for doing spin outs,

00:05:17.588 --> 00:05:22.498
is the fact that you can reach so many
more people and a spin-out is a good

00:05:22.498 --> 00:05:25.778
vehicle for doing it, because when I
worked with Sharp or when I worked with

00:05:25.788 --> 00:05:30.288
Rolls Royce and Oxford Instruments,
we're one out of maybe 50 projects.

00:05:30.398 --> 00:05:34.218
When you do a spin-out, you're one out
of one projects, maybe one of two in the

00:05:34.218 --> 00:05:38.068
second year of the company, but there's
a real focus on taking it from the

00:05:38.068 --> 00:05:44.798
lab into the commercial arena in order
to diffuse it as widely as possible.

00:05:45.228 --> 00:05:49.188
I used to live in Headington quite close
to the John Radcliffe Hospital and one

00:05:49.198 --> 00:05:52.628
thing that got me out of bed in the
morning and on my bike to go off to the

00:05:52.628 --> 00:05:56.838
Institute of Biobank Engineering was the
fact that as I rode past the hospital,

00:05:57.298 --> 00:06:03.908
everybody in the hospital that morning
was having their risk of their disease

00:06:04.018 --> 00:06:08.298
getting worse assessed by an algorithm
that we developed in my research group

00:06:08.638 --> 00:06:11.958
and that's the kind of thing that got
me up and that's the kind of thing that

00:06:11.968 --> 00:06:13.828
drew my passion for doing spinouts.

00:06:14.083 --> 00:06:19.015
And so there you're talking about OBS
medical and patient monitoring, and it's

00:06:19.015 --> 00:06:23.545
one of the first FDA approved technologies
that actually use machine learning.

00:06:23.685 --> 00:06:25.983
How did you come to pioneer in that area?

00:06:26.088 --> 00:06:30.508
What I usually say is that I was
doing machine learning before it

00:06:30.508 --> 00:06:31.728
was even called machine learning.

00:06:32.328 --> 00:06:38.911
I had come back in 1981 to do a
doctorate and I'm unusual because my

00:06:38.911 --> 00:06:40.771
doctorate was actually in pediatrics.

00:06:41.321 --> 00:06:45.721
I was in a small unit at
the John Radcliffe Hospital,

00:06:46.011 --> 00:06:47.161
an accident of history.

00:06:47.211 --> 00:06:52.621
It was because the first professor
of pediatrics in Oxford was Sir Peter

00:06:52.621 --> 00:06:57.051
Tizard,  his father, Sir Henry Tizard
had been one of the inventors of radar.

00:06:57.731 --> 00:07:01.361
So as a professor of paediatrics,
he had huge interest in

00:07:01.361 --> 00:07:02.651
electronic instrumentation.

00:07:02.661 --> 00:07:06.861
So there's a small team of us and my
project for my doctorate at the John

00:07:06.861 --> 00:07:11.031
Radcliffe Hospital was to develop
methods to monitor the blood flow

00:07:11.031 --> 00:07:15.321
to the brain of premature babies at
the risk of brain hemorrhages known

00:07:15.321 --> 00:07:16.831
as intraventricular hemorrhage.

00:07:16.931 --> 00:07:22.366
That got me interested in neuroscience
and so when I went back to mainstream

00:07:22.366 --> 00:07:26.986
electrical engineering after finishing
my doctorate in biomedical engineering,

00:07:27.596 --> 00:07:35.393
neural networks, so computing using some
models that are loosely based on the way

00:07:35.473 --> 00:07:39.743
neurons in our brains compute and pass
information from one neuron to the other

00:07:39.743 --> 00:07:43.788
via  billions, trillions of synapses.

00:07:44.458 --> 00:07:46.048
That was called neural networks.

00:07:46.298 --> 00:07:50.961
I went straight from finishing my
doctorate, my DPhil in Department of

00:07:50.981 --> 00:07:54.611
Pediatrics, to starting to work with
artificial versions of those neural

00:07:54.611 --> 00:08:00.231
networks at a time when somebody
else down the road at the Gatsby

00:08:00.231 --> 00:08:04.501
in London, somebody called Geoff
Hinton was doing exactly the same.

00:08:04.881 --> 00:08:11.911
Geoff Hinton published a key paper on
the Error Back Propagation Algorithm

00:08:11.911 --> 00:08:15.601
in 1986, which underpins the whole of
machine learning and I was doing that in

00:08:15.601 --> 00:08:21.286
the late 1980s and was then appointed as
a professor in 1988 and what's wonderful

00:08:21.286 --> 00:08:25.076
about being a professor in Oxford, you
can explore really what you want to

00:08:25.076 --> 00:08:28.976
explore because I was an electrical
engineer, what I originally explored

00:08:29.206 --> 00:08:33.696
was building special purpose hardware
to implement these neural networks.

00:08:33.936 --> 00:08:38.516
I then gradually moved into developing
new algorithms and then having developed

00:08:38.516 --> 00:08:42.606
those new algorithms, moved into
applications, monitoring jet engines,

00:08:42.916 --> 00:08:46.336
and then back to my first love,
because the university had asked me

00:08:46.336 --> 00:08:49.806
to set up the Institute of Biomedical
Engineering, back into applying those

00:08:49.806 --> 00:08:51.906
algorithms to healthcare applications.

00:08:52.002 --> 00:08:55.853
So you have an amazing journey there
with machine learning as a kind of

00:08:56.083 --> 00:08:58.623
underpinning technology and algorithm.

00:08:59.043 --> 00:08:59.843
That's amazing.

00:08:59.853 --> 00:09:04.279
So you're one of the absolute first people
that was doing this interdisciplinary

00:09:04.279 --> 00:09:08.559
work and you're completely across
both, which is not unique anymore, but

00:09:08.559 --> 00:09:10.879
at the time I imagine very much so.

00:09:11.214 --> 00:09:13.804
It was at the start, as far
as I know, I was probably the

00:09:13.804 --> 00:09:15.724
first person to be doing this.

00:09:16.191 --> 00:09:20.961
I shared an office in the first six months
with one of the world's best machine

00:09:20.961 --> 00:09:25.051
learning and computer vision experts,
Professor Andrew Zisserman and as I

00:09:25.051 --> 00:09:28.701
was coding this up, he's a far better
computer software programmer than I am.

00:09:28.961 --> 00:09:32.211
There are bugs in my software and
Andrew used to sort them out for me.

00:09:32.651 --> 00:09:35.371
But he would say, look, you're wasting
your timeline with these algorithms,

00:09:35.371 --> 00:09:36.621
they're a complete waste of time.

00:09:36.631 --> 00:09:41.919
Data driven learning is not the way
of the future and I often remind him

00:09:41.919 --> 00:09:45.206
when he gives, some of these invited
lectures at the Royal Society, he's

00:09:45.206 --> 00:09:49.316
a fellow of the Royal Society, based
on the amazing machine learning and

00:09:49.826 --> 00:09:51.306
computer vision work that he's done.

00:09:51.606 --> 00:09:52.746
So I was a pioneer.

00:09:53.121 --> 00:09:57.351
Probably, again, in the right place at the
right time and I've been very privileged

00:09:57.361 --> 00:10:01.071
to work with some fantastic people in
this field and to apply it to healthcare

00:10:01.541 --> 00:10:05.201
and you asked me in the previous
question about taking it through the FDA.

00:10:05.741 --> 00:10:09.821
Through the company OBS Medical, we
developed a software algorithm to

00:10:09.821 --> 00:10:15.709
identify early physiological deterioration
in patients in step down care or in

00:10:15.709 --> 00:10:22.549
critical care and doctors are very
good at doing pairs of vital signs.

00:10:22.549 --> 00:10:25.529
For example, if your blood pressure goes
one way and the heart rate the other way,

00:10:25.809 --> 00:10:27.339
they might be very worried about that.

00:10:27.689 --> 00:10:31.259
There are actually five or six vital
signs and to think in five or six

00:10:31.259 --> 00:10:35.319
dimensions, it's quite hard for a
doctor, but an algorithm can do that.

00:10:35.549 --> 00:10:39.569
So we designed a machine learning
algorithm that use all six vital

00:10:39.569 --> 00:10:44.421
signs, integrated the information and
then as a result of that, was able to

00:10:44.421 --> 00:10:50.181
identify deterioration earlier than
algorithms or doctors that would only

00:10:50.181 --> 00:10:52.271
look at one or two of these parameters.

00:10:52.641 --> 00:10:54.301
We took that through the FDA.

00:10:54.721 --> 00:10:59.421
We had to have three different existing
devices to get approval as predicate

00:10:59.431 --> 00:11:04.636
devices, but we did and one of the
interesting phenomena, I think, is that

00:11:04.906 --> 00:11:11.326
the FDA has now approved somewhere between
600 and 700 medical devices and software

00:11:11.326 --> 00:11:13.496
algorithms that rely on machine learning.

00:11:13.686 --> 00:11:17.286
So it's already happening in
healthcare and probably healthcare

00:11:17.286 --> 00:11:20.456
is one of the leading fields for the
applications of machine learning.

00:11:20.586 --> 00:11:23.132
And we all benefit in ways that
we're probably not aware of.

00:11:23.406 --> 00:11:28.066
So you've spoken there about some very
specific granular applications of machine

00:11:28.066 --> 00:11:32.482
learning and AI in healthcare that
you pioneered and now are commonplace.

00:11:32.590 --> 00:11:37.210
But one of the areas that still feels
like it's very much in its nascent

00:11:37.230 --> 00:11:42.340
phase is the application to the large
sets of data and you spoke recently

00:11:42.390 --> 00:11:46.890
in the House of Lords about how the
UK shouldn't be looking to try and get

00:11:46.890 --> 00:11:50.680
ahead in large language models and that
the horse is effectively bolted there.

00:11:51.300 --> 00:11:55.760
But that we should be looking at
large multi modality models and

00:11:55.770 --> 00:11:58.930
using our national asset of NHS data.

00:11:59.330 --> 00:12:03.730
Please can you tell us a bit more about
that and how you see that evolving?

00:12:04.020 --> 00:12:06.290
Well, that's already a brilliant
summary that you've given me here.

00:12:06.790 --> 00:12:12.750
The House of Lords Digital Committee
did an inquiry on large language

00:12:12.750 --> 00:12:17.370
models throughout 2023 and published
its report in February 2024.

00:12:17.630 --> 00:12:20.800
It's actually a very good report called
Large Language Models and Generative

00:12:20.800 --> 00:12:24.750
AI, and I thoroughly recommend it
to anybody who wants to read it.

00:12:24.760 --> 00:12:29.924
It's a very good snapshot and the
debate in Parliament was on the 21st of

00:12:29.924 --> 00:12:36.879
November on that particular topic, to
lead the House of Lords and really, what

00:12:36.879 --> 00:12:41.149
I was trying to say, because one of the
recommendations that they were not quite

00:12:41.149 --> 00:12:46.609
sure about was whether the UK should
develop a sovereign large language model.

00:12:46.629 --> 00:12:54.539
So, effectively, its own capability to
compete or to stand alongside GPT 4 or

00:12:54.539 --> 00:13:01.959
Gemini or Anthropos Cloud Sonnet, but it
seems to me that we know that GPT 4 took

00:13:02.349 --> 00:13:07.449
approximately a hundred million dollars
to train and huge amounts of resources,

00:13:07.449 --> 00:13:11.429
both in terms of the people working on
it and the amount of electricity that

00:13:11.429 --> 00:13:16.719
it required in order to be able to train
this hyperscale large language model.

00:13:17.259 --> 00:13:22.164
GPT 5 is the next one, will
probably come out sometime in 2025

00:13:22.274 --> 00:13:26.404
and that's probably closer to a
billion dollar in terms of the cost.

00:13:26.814 --> 00:13:31.954
Now, I don't think the UK can compete
with that and I think that the time

00:13:31.954 --> 00:13:35.354
has probably gone whereby a country
could think that they might want

00:13:35.364 --> 00:13:40.394
to set up their own sovereign AI
capability to compete alongside those

00:13:40.844 --> 00:13:44.924
hyperscale large language models from
the big tech companies in California.

00:13:45.499 --> 00:13:50.259
So, you have two ways of thinking about
how nationally we should deal with that.

00:13:50.429 --> 00:13:53.729
We could work with those big tech
companies and fine tune them for

00:13:53.729 --> 00:13:57.719
some applications and I think there's
probably some room to do that.

00:13:58.199 --> 00:14:02.339
But more interestingly, you could
reflect upon the data sets that you

00:14:02.339 --> 00:14:06.509
have that are unique to your country,
that are an advantage compared to any

00:14:06.509 --> 00:14:11.971
other country and we have several of
these data assets in this country.

00:14:12.021 --> 00:14:14.381
I think we should call
them Sovereign Data Assets.

00:14:14.381 --> 00:14:17.971
That's what I was trying to argue
in my speech in the House of Lords.

00:14:18.551 --> 00:14:23.101
For example, the main example, the
one that I know best, of course, is

00:14:23.101 --> 00:14:29.641
NHS data, but it's also all the BBC
databases, some geospatial databases,

00:14:29.901 --> 00:14:31.371
the Office of National Statistics.

00:14:31.611 --> 00:14:33.541
We've got some amazing resources.

00:14:33.541 --> 00:14:39.807
In terms of NHS data,  67 million
population, 98 percent of people in this

00:14:39.807 --> 00:14:44.967
country use the NHS or at least have
used the NHS at some point in their life.

00:14:45.507 --> 00:14:51.381
It is literally a sovereign data asset
and as well as making that speech in

00:14:51.381 --> 00:14:55.711
the House of Lords, I have put down
my first amendment to the data bill

00:14:55.881 --> 00:15:01.051
trying to define with two or three other
peers, what a sovereign data asset is

00:15:01.301 --> 00:15:05.321
so that we can actually control access
to those sovereign data assets and my

00:15:05.321 --> 00:15:10.941
view, for example, is if we develop
then, on top of these sovereign data

00:15:10.941 --> 00:15:16.201
assets, what you quite right, are not
large language models, they're not

00:15:16.301 --> 00:15:23.421
LLMs, they're LMMs, Large Multimodal
Models, simply because the NHS data is

00:15:23.421 --> 00:15:29.351
not just text, it's not just discharge
summaries, it's much more than that.

00:15:29.371 --> 00:15:33.111
It's the vital signs like the blood
pressure and heart rate, it's the

00:15:33.391 --> 00:15:37.541
lab data, your blood, your blood
glucose, your creatinine and so on.

00:15:37.971 --> 00:15:43.796
It's also the pathology lab data, which
these days comes in form of images, it's

00:15:43.876 --> 00:15:50.476
MR scans, CT scans, etc. So, multiple
modalities and if you integrate all of

00:15:50.496 --> 00:15:54.586
that together, you will be able to do
things that have not been done before

00:15:54.916 --> 00:15:58.531
and let me give you a very simple
example to try and illustrate that.

00:15:59.191 --> 00:16:04.601
We now know that machine learning
algorithms using deep learning are as

00:16:04.601 --> 00:16:11.609
good as radiologists in analysing images,
for example, mammograms or to detect

00:16:11.629 --> 00:16:18.819
breast cancer or other forms of CT scans
or MRI scans for early cancer detection.

00:16:19.279 --> 00:16:22.899
Performance of machine learning
algorithms are equivalent to

00:16:22.909 --> 00:16:25.319
those of the best human experts.

00:16:25.829 --> 00:16:28.688
The interesting thing is
with these new techniques.

00:16:29.128 --> 00:16:34.588
We can go further because we can use
not just the scans, we can use the whole

00:16:34.588 --> 00:16:39.498
profile of the patient at the time the
scan was done and probably in the run

00:16:39.508 --> 00:16:44.418
up to the scan, there probably is either
some hospital data or some primary care

00:16:44.418 --> 00:16:49.378
data on that patient before they have the
scan and also because we'll be doing this

00:16:49.708 --> 00:16:55.328
training our models on retrospective data,
we'll have the outcome two years from the

00:16:55.328 --> 00:17:00.168
scan, five years from the scan, maybe even
ten years from the scan eventually and

00:17:00.168 --> 00:17:06.618
because you know the outcomes, the Large
Multimodal Model will be able to identify

00:17:07.168 --> 00:17:12.258
some hidden features in the scan that even
the radiologists themselves do not know

00:17:12.528 --> 00:17:15.238
are the features that affect the outcome.

00:17:15.698 --> 00:17:20.478
So we'll have better analytical
prediction than human experts can do

00:17:20.758 --> 00:17:25.328
and that's a fantastic goal to set for
ourselves for those Large Multimodal

00:17:25.708 --> 00:17:30.348
Models and the NHS dataset is the
best in class, best in the world.

00:17:30.538 --> 00:17:33.738
There are countries like Estonia
and so on that have got all this

00:17:33.758 --> 00:17:37.218
data available, but Estonia's
population is one and a half million.

00:17:37.438 --> 00:17:40.328
It's actually smaller than the number
of people that work for the NHS.

00:17:41.488 --> 00:17:45.308
Just taking that, it's amazing
and very exciting to hear.

00:17:45.658 --> 00:17:49.498
You can also see a way that not only
would those that are being scanned be

00:17:49.498 --> 00:17:53.168
captured earlier because the model would
pick up on things that the human eye

00:17:53.188 --> 00:17:57.778
perhaps doesn't even know to look for,
but you will be able to potentially

00:17:57.868 --> 00:18:01.538
create a profile of those that did
develop and scan those earlier and

00:18:01.538 --> 00:18:06.392
therefore potentially put assets and
focus on screening for particular cohorts?

00:18:06.912 --> 00:18:07.522
Absolutely.

00:18:07.632 --> 00:18:10.802
It's interesting, healthcare,
you sort of work backwards.

00:18:11.032 --> 00:18:16.532
The first problem you solve is to
try and improve the treatment of

00:18:16.532 --> 00:18:19.682
those who have already been diagnosed
based on what you've learned with

00:18:19.682 --> 00:18:25.032
your Large Multimodal Models, for
example, in terms of their outcomes.

00:18:25.192 --> 00:18:30.282
Then, you might improve diagnosis and
reduce the false positives or the false

00:18:30.322 --> 00:18:34.172
negatives and then, of course, the
Holy Grail, you're absolutely right, is

00:18:34.192 --> 00:18:39.552
prevention and so, once we build these
huge data sets and of course, they will

00:18:39.552 --> 00:18:46.752
be augmented two fold by genomics data
and also by these extraordinary data sets.

00:18:47.162 --> 00:18:50.402
The one that already exists and has
existed for more than a decade, UK

00:18:50.402 --> 00:18:55.752
Biobank and we've just had under the new
government, the first permission to link

00:18:55.752 --> 00:19:02.052
UK Biobank data to those individuals
primary care data and the follow on

00:19:02.052 --> 00:19:06.582
project from UK Biobank is Our Future
Health, which has just recruited its

00:19:06.582 --> 00:19:12.182
1,000,000th participants and if you
integrate all of that information together

00:19:12.572 --> 00:19:18.247
and build the Time Series, as we call
it, the historical longitudinal data

00:19:18.247 --> 00:19:20.547
set, you'll be able to go backwards.

00:19:20.547 --> 00:19:25.777
For example, already in UK Biobank,
and just with UK Biobank data, we can

00:19:25.777 --> 00:19:32.847
identify from their records 10 years
ago, using simple, wearable data, those

00:19:32.877 --> 00:19:36.777
people are far more likely to have
Parkinson's disease ten years down the

00:19:36.777 --> 00:19:41.657
road and we can start therefore early
intervention to try and minimise the

00:19:41.657 --> 00:19:46.307
risk of Parkinson's or minimise the
severity, of the Parkinson's disease.

00:19:46.517 --> 00:19:52.107
So all that becomes available, provided
we have the whole history and we can

00:19:52.107 --> 00:19:53.657
start doing some of that already.

00:19:53.707 --> 00:19:58.347
We can start screening people in hospital,
for example, we could do it today for

00:19:58.347 --> 00:20:03.267
colorectal cancer, because we know that
there are early warning signs, maybe

00:20:03.267 --> 00:20:07.724
in your blood profile, two years ahead
of you developing colorectal cancer.

00:20:08.014 --> 00:20:11.734
So we could do this automatically
for everybody who's in the hospital.

00:20:11.994 --> 00:20:16.494
It might only be a fraction of a percent
that we detect early, but when you think

00:20:16.494 --> 00:20:20.704
that there are tens of millions of visits,
because some patients go several times a

00:20:20.704 --> 00:20:27.234
year to the hospital, but if you analysed
all the data when there are lab results

00:20:27.259 --> 00:20:31.979
available, you could screen people for
colorectal cancer two or three years ahead

00:20:31.989 --> 00:20:33.556
of when the first symptoms would appear.

00:20:34.026 --> 00:20:36.186
That's something that we
know we could do today.

00:20:36.336 --> 00:20:41.776
We could do so much more if we build
those data sets and integrate all

00:20:41.776 --> 00:20:43.346
the information that's available.

00:20:43.786 --> 00:20:50.056
One other thing I should stress is Large
multimodal Models are incredibly good at

00:20:50.056 --> 00:20:55.556
dealing  with the variability from one
patient to another, because not every

00:20:55.556 --> 00:20:57.086
patient will have the whole record.

00:20:57.356 --> 00:20:59.016
Some will only have a
fraction of the record.

00:20:59.206 --> 00:21:03.016
We can still do something in the
way that we train Large Multimodal

00:21:03.486 --> 00:21:07.426
Models, even if we only have 10
percent of the complete data set.

00:21:07.756 --> 00:21:10.336
So, I think this could
be transformational.

00:21:10.376 --> 00:21:15.676
It's totally dependent on us having
access to the data and therefore I

00:21:15.676 --> 00:21:19.916
see part of my work in the House of
Lords in trying to set up the rules for

00:21:20.006 --> 00:21:24.916
access to this data in order to benefit
patients in this country and in order

00:21:24.916 --> 00:21:30.406
eventually financially to help the NHS
so that the NHS budget will continue

00:21:30.406 --> 00:21:35.591
to grow as we know, but we can't just
keep putting money from taxation.

00:21:35.591 --> 00:21:39.481
There needs to be other ways and this
is one way of doing it in a way which

00:21:39.481 --> 00:21:43.411
I think we can convince the general
population is a worthwhile way of doing.

00:21:43.526 --> 00:21:44.146
I hope so.

00:21:44.656 --> 00:21:47.136
I actually am a participant
in MyFutureHealth.

00:21:47.401 --> 00:21:47.911
Well done.

00:21:48.161 --> 00:21:49.911
I'm a UK Biobank participant.

00:21:50.212 --> 00:21:56.362
Taking a slight step away from the UK,
which is so exciting, back into Oxford

00:21:56.362 --> 00:22:01.462
more specifically, one of the big changes
that's coming into Oxford is the Ellison

00:22:01.502 --> 00:22:06.482
Institute of Technology and a lot of the
themes that we're talking about here are

00:22:06.522 --> 00:22:11.272
part of their stated goals of pioneering
healthcare and clearly the relationship

00:22:11.282 --> 00:22:15.732
with Larry Ellison and the Tony Blair
Institute speaks to internationally

00:22:15.762 --> 00:22:18.452
looking to digitise healthcare.

00:22:18.682 --> 00:22:24.462
I'd love to know how you see the Ellison
Institute impacting Oxford and really

00:22:24.462 --> 00:22:28.337
just to get your views on what it's
going to change that's currently here.

00:22:28.517 --> 00:22:31.107
So I think it's a fantastic
development for Oxford.

00:22:31.317 --> 00:22:34.542
Some time ago, AstraZeneca
tried to come to Oxford.

00:22:34.762 --> 00:22:37.482
We could not accommodate
them, they went to Cambridge.

00:22:37.852 --> 00:22:40.322
So the fact that the Ellison
Institute of Technology has come

00:22:40.322 --> 00:22:42.952
to Oxford is wonderful for Oxford.

00:22:43.512 --> 00:22:44.812
That's the first thing to say.

00:22:44.942 --> 00:22:50.782
Secondly, the agreement which the
university has just signed with the

00:22:50.782 --> 00:22:54.152
Ellison Institute of Technology, I
think it's the right kind of agreement.

00:22:54.652 --> 00:22:58.022
I think we've been very fortunate
in that Sir John Bell, who is Regius

00:22:58.022 --> 00:23:02.982
Professor of Medicine since 2002,
has moved away from the university to

00:23:02.982 --> 00:23:08.342
become the first president of this EIT,
Ellison Institute of Technology, in

00:23:08.342 --> 00:23:14.102
Oxford and because John's come from the
university, he has seen what has happened,

00:23:14.412 --> 00:23:16.382
lower scale, but in similar ways.

00:23:16.392 --> 00:23:18.352
So let me try and illustrate that.

00:23:18.482 --> 00:23:25.888
When I became head of engineering science
in 2014, Within a couple of weeks of

00:23:25.888 --> 00:23:30.108
becoming head of department and the
same thing happened in computer science,

00:23:30.108 --> 00:23:34.428
the next door department, two or three
of the professors in my department

00:23:34.818 --> 00:23:39.328
were being offered the opportunity to
leave the university and go and work

00:23:39.328 --> 00:23:45.228
for, it was then, Google DeepMind, or
Amazon, or Facebook as it was then,

00:23:45.238 --> 00:23:47.948
Meta now of course, NVIDIA and others.

00:23:48.198 --> 00:23:52.378
So across engineering and computer
science, there was this, what appeared

00:23:52.378 --> 00:23:57.048
to be a land grab, and it was really an
intellectual land grab is about coming

00:23:57.048 --> 00:24:02.438
in and grabbing the professors, but over
a weekend, because it had to be done

00:24:02.438 --> 00:24:07.998
that quickly, we actually negotiated a
deal whereby it was 50 50 and 10 years

00:24:07.998 --> 00:24:11.898
later, I can say, and if I could mention
again, Professor Andrew Zisserman,

00:24:11.998 --> 00:24:18.418
FRS, ever since 2014, he has had a 50
percent appointment at Google DeepMind

00:24:18.418 --> 00:24:22.138
and 50 percent in the University in
the Department of Engineering Science

00:24:22.608 --> 00:24:24.768
and that has worked pretty well.

00:24:25.248 --> 00:24:32.058
Partly because the university has a
lot of very good PhD or DPhils, as we

00:24:32.058 --> 00:24:35.718
call them, students, so postgraduate
research students who want to come

00:24:35.718 --> 00:24:38.998
and work for professor Zisserman,
because he's so well known, so he

00:24:38.998 --> 00:24:41.748
has a good supply of PhD students.

00:24:41.968 --> 00:24:43.428
That's his university lab.

00:24:43.953 --> 00:24:49.213
But all the postdocs are now employed by
these big tech companies, because unless

00:24:49.213 --> 00:24:55.063
you want to stay in academia, if you've
got a PhD, a DPhil, which you did in

00:24:55.063 --> 00:24:58.893
Professor Zisserman's group, Professor
Paul Newman's group, other groups are

00:24:58.893 --> 00:25:03.223
available in engineering and computer
science and stats and mathematics and

00:25:03.633 --> 00:25:10.548
big Data Institute,  then you can command
a very worthwhile salary as a postdoc

00:25:10.598 --> 00:25:14.668
in any of these big tech companies
and the resources are so much better.

00:25:14.748 --> 00:25:18.948
So I worked with Andrew Zisserman
and I was head department, trying to

00:25:18.948 --> 00:25:25.168
get him state of the art GPU cluster
for his computer vision algorithms.

00:25:25.458 --> 00:25:28.698
That took several months, and
we went as fast as we could.

00:25:29.078 --> 00:25:31.568
By the time the cluster arrives,
it's almost out of date.

00:25:31.908 --> 00:25:36.128
Whereas if you work in one of the big
tech companies, that compute resource

00:25:36.138 --> 00:25:40.108
is available on your desk tomorrow
if you need it and you have these

00:25:40.108 --> 00:25:44.628
wonderful postdocs and that's the
reason why these professors want to

00:25:44.628 --> 00:25:46.098
work with those big tech companies.

00:25:46.308 --> 00:25:50.828
It's both a human resource at
postdoc level and the computational

00:25:50.828 --> 00:25:55.338
resource and Ellison Institute of
Technology will do the same, I think.

00:25:55.548 --> 00:26:00.338
So, the idea is that the professors
will not all move across to Littlemore

00:26:00.618 --> 00:26:03.188
and become employees of EIT.

00:26:03.678 --> 00:26:09.798
The plan is for them to have these 50/50
jobs, again, working similarly across both

00:26:09.798 --> 00:26:13.828
the university and the Ellison Institute
of Technology and the way we should view

00:26:13.828 --> 00:26:17.943
it is that we should be able to do so
much more then we would have been able

00:26:17.943 --> 00:26:20.263
to do if EIT had not come to Oxford.

00:26:20.448 --> 00:26:23.408
And I think there's a theme coming
through the conversation here when

00:26:23.408 --> 00:26:27.578
you've spoken about large language
models, the expense of training them,

00:26:27.628 --> 00:26:31.818
DeepMind, the expense of putting the
data into that is well documented.

00:26:32.008 --> 00:26:35.698
The research that's available
here being funded or co funded

00:26:35.698 --> 00:26:37.488
by EIT will be a benefit.

00:26:37.998 --> 00:26:41.808
I suppose that is a very pragmatic
view and by the way, for what

00:26:41.808 --> 00:26:43.058
it's worth, I agree with it.

00:26:43.618 --> 00:26:47.578
The flip side of that is we
often are accused in the UK of

00:26:47.578 --> 00:26:49.258
having sold companies too soon.

00:26:49.993 --> 00:26:56.353
I'd love to know whether you think that's
inevitable or whether more capital supply

00:26:56.363 --> 00:27:01.683
with changes such as the Mansion House
Compact should or could change that.

00:27:02.433 --> 00:27:08.213
Yes, well, before the large language model
debate in the House of Lords, we also had

00:27:08.303 --> 00:27:15.063
a debate on science and technology and
its contribution to the UK economy and we

00:27:15.063 --> 00:27:19.203
only had four minutes each because there
were so many peers wanting to speak on

00:27:19.203 --> 00:27:21.043
the topic, which in many ways is great.

00:27:21.563 --> 00:27:24.553
That this will be a topic that
will be of interest to many

00:27:24.553 --> 00:27:26.043
members of the House of Lords.

00:27:26.793 --> 00:27:31.893
I started by mentioning the fact that
just two or three weeks before, we'd

00:27:31.903 --> 00:27:37.978
had the wonderful news of Geoff Hinton,
having won the Nobel Prize for Physics and

00:27:37.978 --> 00:27:42.368
then very soon afterwards Demis Hassabis
having won the Nobel Prize for Chemistry.

00:27:42.598 --> 00:27:45.018
Geoff Hinton had done
his PhD at Edinburgh.

00:27:45.458 --> 00:27:49.928
He did his Neural Network Algorithm
Development when he was at UCL.

00:27:50.508 --> 00:27:54.918
Demis Hassabis was a Cambridge
undergrad and his PhD at UCL.

00:27:54.918 --> 00:28:00.793
But they were awarded and Nobel Prizes
at a time when the latest developments

00:28:00.793 --> 00:28:06.153
they'd made of the last decade, they had
done for a two trillion dollar U. S. big

00:28:06.153 --> 00:28:11.553
tech company, Google and so I thought I
would do a gedankenexperiment in my four

00:28:11.563 --> 00:28:18.863
minutes available, which is, what would it
take to have a one trillion dollar company

00:28:19.163 --> 00:28:22.623
come out of the UK in, say, the AI field?

00:28:23.118 --> 00:28:30.398
And the first one was to have a good
supply of UK PhD students in Russell

00:28:30.398 --> 00:28:33.908
Group universities doing computer
science, information engineering,

00:28:33.908 --> 00:28:38.218
statistics, and so on and there
are some dangerous signs here.

00:28:38.268 --> 00:28:43.868
There's possibly an unintended consequence
of the £9,250 tuition fee for undergrad

00:28:43.868 --> 00:28:48.308
degrees, putting people off from staying
on and doing PhDs because they graduate

00:28:48.308 --> 00:28:52.548
with 50k debt and by the time they
finish their PhDs there'll be a 70k debt

00:28:52.578 --> 00:28:54.398
with the interest rates that we have.

00:28:54.728 --> 00:28:59.198
So that's the first thing, let's make sure
there are enough home PhD students around

00:28:59.468 --> 00:29:04.558
to create the intellectual capital to be
the researchers within such a company.

00:29:05.088 --> 00:29:08.358
Secondly, we have to make sure
that there's funding available

00:29:08.718 --> 00:29:10.598
to create the companies.

00:29:10.858 --> 00:29:13.018
So these are proof of concept funds.

00:29:13.018 --> 00:29:16.838
In the budget of the previous week,
the Chancellor had announced 40 million

00:29:17.288 --> 00:29:19.098
pounds for proof of concept fund.

00:29:19.448 --> 00:29:20.458
I said that was great.

00:29:20.518 --> 00:29:22.518
It shouldn't go to Oxford,
Cambridge and Imperial because

00:29:23.028 --> 00:29:24.318
those funds already exist.

00:29:24.638 --> 00:29:26.488
In Oxford, we're incredibly privileged.

00:29:26.718 --> 00:29:30.788
We have Oxford Science Enterprises, who
I know have already featured on these

00:29:30.798 --> 00:29:35.183
podcasts, so I don't need to explain,
but it's a fantastic opportunity to set

00:29:35.183 --> 00:29:41.713
up new companies in all sorts of fields,
deep tech, biotech, AI, et cetera.

00:29:42.173 --> 00:29:45.151
So that's the second
required step that you have.

00:29:45.151 --> 00:29:51.928
The third is scale up funding and really
one of those is maybe the mansion has

00:29:51.978 --> 00:29:57.388
compact delivering by 2030 on the promise
that five percent of pension funds will

00:29:57.388 --> 00:30:03.848
invest in these UK high tech companies
and then the fourth one and I'm not an

00:30:03.848 --> 00:30:08.268
expert in this at all, but there's also
been a report done which the House of

00:30:08.318 --> 00:30:12.678
Lords had looked at by Lord Harrington
for Jamie Hunt on foreign direct

00:30:12.688 --> 00:30:16.341
investment and again, this is all part
of the ecosystem because it so happened

00:30:16.341 --> 00:30:20.136
that Richard Harrington was in the year
below me at Keeble as an undergraduate.

00:30:20.696 --> 00:30:24.156
There's a lot in there and the
variables are many, but the prize

00:30:24.156 --> 00:30:27.506
is huge, so we can only hope that
some of those things come to bear.

00:30:28.571 --> 00:30:33.651
Taking a small step back, you spoke
about a few individuals there who have

00:30:33.651 --> 00:30:38.001
won Nobel Prizes and who have done
brilliant things for this country,

00:30:38.001 --> 00:30:42.641
but sadly, more latterly perhaps
under the auspices of a US company.

00:30:43.021 --> 00:30:47.261
William Hague published a comment
article earlier in the year commenting

00:30:47.311 --> 00:30:51.456
on Novo Nordisk and how they've
moved their national GDP by, I think,

00:30:51.456 --> 00:30:55.766
two to three percent and very much
focused on very few people will

00:30:55.766 --> 00:30:57.876
change the outcomes for very many.

00:30:58.296 --> 00:31:01.626
He is now Chancellor of Oxford University.

00:31:01.876 --> 00:31:06.136
How do you see his leadership in that
role potentially impacting some of

00:31:06.156 --> 00:31:07.666
these themes that you're talking about?

00:31:08.166 --> 00:31:12.286
So again, I think in the
Chancellor election we had six

00:31:12.356 --> 00:31:13.946
and then five in the final round.

00:31:14.056 --> 00:31:17.406
The six other ones voted for in the
first round, five of them ended up

00:31:17.736 --> 00:31:19.446
in the final round as the final five.

00:31:19.736 --> 00:31:23.216
I thought all five were
outstanding candidates and William

00:31:23.216 --> 00:31:24.406
Hague is a very clear winner.

00:31:24.776 --> 00:31:29.156
We used the alternative vote
system and every round William

00:31:29.156 --> 00:31:30.916
Hague, Lord Hague was the winner.

00:31:30.916 --> 00:31:34.503
So he has a very strong mandate, he's
highly respected at the university,

00:31:34.843 --> 00:31:38.693
he's been very good working with
his college, Magdalen, I know one or

00:31:38.693 --> 00:31:42.653
two Magdalen fellows that speak very
highly of him, very much involved,

00:31:42.823 --> 00:31:46.153
not just with the fellows, but also
with the students and I think he'll

00:31:46.163 --> 00:31:47.663
be the right kind of ambassador.

00:31:48.448 --> 00:31:50.008
He has the right gravitas.

00:31:50.148 --> 00:31:54.908
I happen to listen quite regularly on
Tuesday morning on Times Radio when

00:31:54.908 --> 00:32:00.818
he features between 9 and 9:30 and I
think he says very sensible things.

00:32:00.978 --> 00:32:05.988
He has worked very closely with Tony Blair
and the Tony Blair Institute on topics

00:32:05.988 --> 00:32:11.138
that we've discussed in this podcast,
sovereign data assets, making use of the

00:32:11.208 --> 00:32:17.275
tremendous NHS data assets that we have,
but also thinking about how, because

00:32:17.575 --> 00:32:22.721
behind the US and China, the UK is very
much in third place and there's not

00:32:22.961 --> 00:32:25.411
much doubt that it is in third place.

00:32:25.421 --> 00:32:30.221
So we've got this fantastic human capital
assets in terms of the AI research

00:32:30.711 --> 00:32:32.921
across Russell Group universities.

00:32:33.701 --> 00:32:38.781
It's often said that we have 12
universities in the top 100 in the world.

00:32:39.611 --> 00:32:43.801
What we need to do, what Oxford needs
to do, but not just Oxford, any of

00:32:43.801 --> 00:32:49.741
these other 12 universities need to
do, is to go from being the best in

00:32:49.741 --> 00:32:54.811
the top 100 universities in the world,
to being in the top 100 companies

00:32:54.881 --> 00:32:57.795
in these new technologies as well,
you not necessarily a dozen, but

00:32:57.825 --> 00:32:59.595
half a dozen would be good enough.

00:32:59.845 --> 00:33:01.235
Now that is quite difficult.

00:33:01.645 --> 00:33:04.275
People don't realise
sometimes why it is difficult.

00:33:04.805 --> 00:33:08.095
It is difficult, for example, in
healthcare, where I've got some

00:33:08.095 --> 00:33:12.915
expertise, is because if you
develop AI for healthcare or a

00:33:12.915 --> 00:33:14.575
new piece of medtech and so on.

00:33:14.995 --> 00:33:19.265
If you go to the US, you've got
50 states with roughly the same

00:33:19.495 --> 00:33:24.235
reimbursement rule, each of which is
roughly the size of a European country.

00:33:24.245 --> 00:33:26.525
So there's a huge
advantage there with scale.

00:33:26.545 --> 00:33:31.078
It is quite hard to scale up in
the UK because, we are a small

00:33:31.108 --> 00:33:34.498
country, smaller than Texas, for
example, or smaller than Florida.

00:33:34.858 --> 00:33:36.638
So how do you scale up?

00:33:36.988 --> 00:33:40.670
We've talked about in the podcast that,
we've got the ideas, we've got the

00:33:40.680 --> 00:33:46.530
companies being spun out, we have the seed
funding, I think we have Series A funding.

00:33:46.750 --> 00:33:50.800
I think what people are really beginning
to talk about is the scale up funding.

00:33:50.960 --> 00:33:55.910
That is missing and in Oxford, we need
to do that in partnership with OSE that

00:33:55.910 --> 00:34:00.520
provides seed funding, provides Series
A, contributes sometimes to Series

00:34:00.520 --> 00:34:06.580
B but if we want to do B and C, we
really need to have the funds available

00:34:06.770 --> 00:34:11.880
and that is where, at the moment,
we are failing and we're not seeing,

00:34:11.920 --> 00:34:16.474
not just the data, but some of these
companies as almost sovereign assets.

00:34:16.734 --> 00:34:21.214
I think, personally, it is a pity
that Arm is no longer a UK company.

00:34:21.689 --> 00:34:26.039
I happen to know the person who took
it from an SME to a world leader,

00:34:26.049 --> 00:34:30.339
Warren East, who then went on to be
Chief Executive of Rolls Royce, but

00:34:30.339 --> 00:34:34.769
when he was Chief Executive of Arm, he
developed this amazing licensing model.

00:34:35.409 --> 00:34:38.499
He's a graduate of my department,
he's an Oxford engineer.

00:34:38.499 --> 00:34:44.029
Arm, I think if the right capital had
been available, could have remained a UK

00:34:44.059 --> 00:34:48.759
company instead of being bought out by
SoftBank, and it's the Arms of this world

00:34:49.049 --> 00:34:53.749
that we need to grow and to go back to
the exam question, as we always say to

00:34:53.749 --> 00:34:58.329
the undergraduates, somebody like William
Hague, with the stature, the gravitas,

00:34:58.349 --> 00:35:04.399
the expertise, the people that he knows
and his presence on the world stage can

00:35:04.439 --> 00:35:08.359
help us and that's not the only thing that
the Chancellor will do, of course, but

00:35:08.389 --> 00:35:14.819
can help us in this particular domain,
really scale up as a university, to help

00:35:14.829 --> 00:35:21.129
the companies we've spun-out to go through
these stages beyond Series B, Series C,

00:35:21.179 --> 00:35:25.329
and really becoming, first of all, billion
dollar companies, then 10 billion, then

00:35:25.349 --> 00:35:28.179
100 billion, for the benefit of the UK.

00:35:28.314 --> 00:35:32.784
It's very exciting and you spoke there
about William Haig and Tony Blair having

00:35:32.784 --> 00:35:36.764
worked together and one of the things
that they've also co published papers on

00:35:37.114 --> 00:35:40.904
is pension capital reform, which we're
now seeing obviously the Mansion House

00:35:40.904 --> 00:35:45.674
side but also Labour looking into local
government pension schemes, all of which

00:35:45.834 --> 00:35:50.839
is very positive mood music, and we just
need to see it translate in these ways.

00:35:50.849 --> 00:35:54.969
So I too am full of optimism
for what that might do.

00:35:55.389 --> 00:35:59.619
There is the other half of that, which
nobody ever seems to want to talk about,

00:35:59.619 --> 00:36:04.009
which you've raised, which is, we still
don't have a natural buyer for many of

00:36:04.039 --> 00:36:06.259
these things at the scale that the US has.

00:36:06.357 --> 00:36:07.337
Thank you, Lionel.

00:36:07.337 --> 00:36:11.657
This has been a fabulous conversation
and I'm very grateful for your

00:36:11.667 --> 00:36:14.927
views across your domain areas,
we could have gone on for hours.

00:36:15.487 --> 00:36:20.307
Just finally, I would love to get
a sense of what you would most

00:36:20.317 --> 00:36:24.087
like to see developing here at
Oxford in the next five years.

00:36:24.612 --> 00:36:29.892
Well, we're sitting in Reuben College, so
I think I need to mention Reuben College.

00:36:30.192 --> 00:36:34.467
It's sometimes I describe it
as my fifth startup and it's

00:36:34.467 --> 00:36:39.617
wonderful how the university can
be flexible and do new things.

00:36:39.617 --> 00:36:43.557
We're the 39th college in, above
900 years, we don't know exactly

00:36:43.557 --> 00:36:44.797
how old the university is.

00:36:44.847 --> 00:36:50.357
What is remarkable about Oxford is
that at times, the bureaucracy can seem

00:36:50.387 --> 00:36:55.407
overwhelming, but the people at the
helm, the Vice Chancellor and now the

00:36:55.747 --> 00:37:00.467
Chancellor, William Hague and so on, are
going to leave the door open for Oxford to

00:37:00.467 --> 00:37:03.297
be a full participant in the 21st century.

00:37:04.157 --> 00:37:07.177
I think that there are possibly
one or two other colleges in the

00:37:07.187 --> 00:37:08.867
pipeline, it'd be great to see.

00:37:08.867 --> 00:37:12.577
I would welcome a 40th and a 41st
college, it'd be great no longer to be

00:37:12.577 --> 00:37:14.797
the newbie, and I think that is important.

00:37:15.232 --> 00:37:17.032
The university continues to evolve.

00:37:17.462 --> 00:37:23.032
I see the Ellison Institute of Technology
as part of that, but I do think that we

00:37:23.032 --> 00:37:29.022
do need to think about the human capital
that we're creating, that we align

00:37:29.362 --> 00:37:32.212
and I'm talking about a postgraduate
level, but of course, when I was head

00:37:32.212 --> 00:37:36.452
of engineering, I was also equally
concerned about the undergraduate

00:37:36.452 --> 00:37:41.077
level and there's this tension between
home and international students.

00:37:41.097 --> 00:37:45.597
Yes, we should educate international
students, it's part of Oxford's

00:37:45.597 --> 00:37:47.537
mission, it's a global enterprise.

00:37:47.977 --> 00:37:53.727
But I do worry that we're in danger
of not creating enough human capital.

00:37:53.967 --> 00:37:56.757
So, for these enterprises of the future.

00:37:57.337 --> 00:38:01.277
One of the interesting things that
is happening, I think, at the moment,

00:38:01.912 --> 00:38:05.132
is the revival of the humanities as
a result of what we talked about.

00:38:05.782 --> 00:38:09.682
So when people are saying that
artificial general intelligence is

00:38:09.682 --> 00:38:14.662
just around the corner, really, what
that makes us think about is, what is

00:38:14.662 --> 00:38:20.132
it that makes us different as human
beings from a piece of silicon on which

00:38:20.142 --> 00:38:22.372
a very intelligent algorithm runs?

00:38:22.762 --> 00:38:29.142
And I think the answers lie not just in
computer science, but also in philosophy,

00:38:29.422 --> 00:38:36.203
in theology and that's why Oxford is
a fantastic place to be able to bring

00:38:36.533 --> 00:38:40.423
those subjects together to discuss
exactly those things and I think we are

00:38:40.423 --> 00:38:43.930
going to be discussing these topics even
possibly one day on an Oxford+ podcast,

00:38:44.907 --> 00:38:51.097
and what I hope will happen is actually
not just what we talked about, companies

00:38:51.097 --> 00:38:56.757
to flourish at Series B, Series C, but
also graduate education, undergraduate

00:38:56.817 --> 00:39:01.417
education, but not just in the subjects
that we require to develop these

00:39:01.417 --> 00:39:05.197
algorithms, but also in the humanities
and so on and my parting shot would be

00:39:05.197 --> 00:39:09.837
to say, here at Reuben College, we're
right in the middle of the science area.

00:39:09.837 --> 00:39:13.247
Above us, we've still got one floor
of the science library, the college

00:39:13.247 --> 00:39:17.507
occupies the middle floor and the
grand floor and I don't want it

00:39:17.507 --> 00:39:19.077
to be known as a science college.

00:39:19.467 --> 00:39:23.977
It should be, like every other college,
a multidisciplinary college, and Oxford

00:39:23.977 --> 00:39:29.407
will be a fantastic place for these
conversations about where does AI go,

00:39:29.647 --> 00:39:33.587
how should we develop the right kind of
ethical framework when we're developing

00:39:33.587 --> 00:39:39.207
not just AI, but AGI, Artificial General
Intelligence, and also these incredibly

00:39:39.217 --> 00:39:43.697
important conversations as to what
the difference may be between AGI.

00:39:44.147 --> 00:39:49.302
A is in the clue, artificial, and the
intelligence, the consciousness, whatever

00:39:49.302 --> 00:39:54.392
it is, the human soul, if you will
and here I reflect my kind of Eastern

00:39:54.392 --> 00:39:58.622
European background, that is going to
be even more important to have those

00:39:58.652 --> 00:40:03.852
conversations and Oxford Colleges and
the university at large is the place and

00:40:03.852 --> 00:40:08.962
so my greatest wish will be that Oxford
continues to thrive, not unbalanced, but

00:40:08.982 --> 00:40:13.602
balanced across all of these subjects
for it to be a place where these really

00:40:13.612 --> 00:40:15.422
important conversations will take place.

00:40:15.587 --> 00:40:16.187
Wonderful.

00:40:16.290 --> 00:40:17.390
Thank you very much, Lionel.

00:40:18.023 --> 00:40:20.863
Thanks for listening to this
episode of Oxford+ presented

00:40:20.883 --> 00:40:22.203
by me, Susannah de Jager.

00:40:22.433 --> 00:40:26.573
If you want to stay up to date with all
things Oxford+, please visit our website,

00:40:26.653 --> 00:40:31.463
oxfordplus.co.uk and sign up for our
newsletter so you never miss an update.

00:40:31.873 --> 00:40:35.133
Oxford+ was made in partnership
with Mishcon de Reya and is produced

00:40:35.133 --> 00:40:36.983
and edited by Story Ninety-Four.