WEBVTT

NOTE
This file was generated by Descript 

00:00:01.686 --> 00:00:03.916
Susannah de Jager: Welcome to
Ethics and Innovation by Oxford+,

00:00:04.626 --> 00:00:08.706
a special miniseries hosted by me,
Susannah de Jager and sponsored by

00:00:08.706 --> 00:00:10.986
Equinox, Equitable Innovation Oxford.

00:00:12.288 --> 00:00:16.608
Throughout this ethics and innovation
miniseries, we've explored many of the

00:00:16.608 --> 00:00:19.278
risks created by artificial intelligence.

00:00:19.638 --> 00:00:23.988
But also an extraordinary counterpoint,
what becomes possible when we

00:00:23.988 --> 00:00:28.038
deliberately point these tools at
problems society has struggled to solve?

00:00:28.848 --> 00:00:34.608
Juan M Lavista Ferres is Microsoft's
Corporate Vice President and Chief Data

00:00:34.608 --> 00:00:38.358
Scientist and leads its AI for Good Lab.

00:00:39.138 --> 00:00:43.938
Since Juan co-founded the lab in
2018, his team has worked on more

00:00:43.938 --> 00:00:49.638
than 350 projects around the world,
applying AI and data science to

00:00:49.638 --> 00:00:55.458
problems spanning health, humanitarian
response, sustainability, accessibility,

00:00:55.818 --> 00:00:57.768
and fundamental human rights.

00:00:58.368 --> 00:01:01.568
Their collaborators range from
the United Nations and American

00:01:01.568 --> 00:01:05.278
Red Cross to universities, medical
researchers, and the Oxford

00:01:05.278 --> 00:01:07.498
Institute of Technology and Justice.

00:01:08.308 --> 00:01:10.348
The origins of that work are revealing.

00:01:10.818 --> 00:01:15.701
Juan became involved in research into
Sudden Infant Death, asking whether

00:01:15.701 --> 00:01:20.471
analysing data at scale, previously
impossible, might reveal patterns

00:01:20.501 --> 00:01:22.391
conventional research had missed.

00:01:23.231 --> 00:01:28.384
In this final episode of Ethics and
Innovation, we ask Juan what happens when

00:01:28.384 --> 00:01:33.274
you give some of the world's most powerful
analytical tools to scientists, lawyers,

00:01:33.754 --> 00:01:38.884
and humanitarian organisations tackling
some of its hardest problems, and what

00:01:38.884 --> 00:01:41.224
responsibilities come with that power?

00:01:42.167 --> 00:01:44.207
Juan, thank you so much for joining today.

00:01:45.007 --> 00:01:49.687
So before you founded the AI for
Good Lab, you were already a data

00:01:49.687 --> 00:01:53.917
scientist, and you were looking
at societal impact of big data.

00:01:54.067 --> 00:02:02.407
What made you switch from data to AI
enabled data looking at societal problems?

00:02:03.007 --> 00:02:05.890
Juan M Lavista Ferres: I actually started
my career long time ago working for

00:02:05.890 --> 00:02:09.084
the InterAmerican Development Bank, and
this was similar to the World Bank is

00:02:09.084 --> 00:02:13.054
an organisation that works helping the
region in this particular case, Latin

00:02:13.114 --> 00:02:18.764
America and the Caribbean with development
in general and I saw that the technology

00:02:18.764 --> 00:02:24.984
could be a great force for good and for a
long time, I, even working at Microsoft,

00:02:24.984 --> 00:02:30.074
I would work on sometimes on the weekends
to help some organisations around Seattle.

00:02:30.214 --> 00:02:34.164
One of my colleagues had lost a child
to SIDS, since the Sudden Infant Death

00:02:34.164 --> 00:02:38.274
Syndrome is the main cause of death
on children from between one month

00:02:38.274 --> 00:02:43.614
and one year old and through that
exercise, it was very clear to me that

00:02:43.614 --> 00:02:48.204
we could use AI and machine learning
to help a lot of these researchers

00:02:48.204 --> 00:02:50.057
learn more  about their problems.

00:02:50.687 --> 00:02:52.397
And that's how basically I started these.

00:02:52.481 --> 00:02:55.554
Later on I shared these with some
of the leadership here at Microsoft.

00:02:55.604 --> 00:03:00.314
That basically help us create like, "Hey,
can we actually do this as a full-time

00:03:00.314 --> 00:03:02.234
job?" It really was a dream come true.

00:03:02.234 --> 00:03:05.944
I'm very thankful to Microsoft for
that opportunity and that basically

00:03:05.994 --> 00:03:07.374
started the AI for Good Lab.

00:03:08.005 --> 00:03:12.064
Susannah de Jager: And just go into a
bit of detail because you spoke there

00:03:12.064 --> 00:03:16.504
about Sudden Infant Death Syndrome, and
most people will be familiar with it.

00:03:17.284 --> 00:03:22.204
But what you did and the volume of
data that you took in and analysed

00:03:22.444 --> 00:03:25.924
was completely unprecedented and
that's really the point here, is

00:03:25.924 --> 00:03:31.465
it's not something that could ever
have even been approached without AI.

00:03:31.948 --> 00:03:37.598
And so in that particular case, you
analysed 20 million births and 19,000

00:03:37.888 --> 00:03:43.698
cases of sudden unexpected infant
deaths and I just want to go into

00:03:43.698 --> 00:03:45.588
it because I found it fascinating.

00:03:45.798 --> 00:03:49.068
But give us a little bit
of the findings from that.

00:03:49.468 --> 00:03:51.118
Juan M Lavista Ferres: That was
our first research that we need

00:03:51.118 --> 00:03:52.738
that was related to smoking.

00:03:52.738 --> 00:03:56.474
So it was very well known and
established that maternal smoking

00:03:56.474 --> 00:04:00.798
would increase the chances of
having a child die of SIDS, correct?

00:04:01.188 --> 00:04:02.448
That was well established.

00:04:02.978 --> 00:04:05.348
What wasn't well established
at that point were two things

00:04:05.348 --> 00:04:06.428
that we wanted to understand.

00:04:06.428 --> 00:04:10.910
First was that what was the impact
of any single cigarette, correct?

00:04:10.910 --> 00:04:15.320
So it was the same smoking 10
cigarettes or 20 or one cigarette a day?

00:04:15.720 --> 00:04:17.100
There wasn't clarity on that.

00:04:17.580 --> 00:04:21.930
And not only that, the fact that we
could understand that 10 cigarettes was

00:04:21.930 --> 00:04:25.440
more than five, and we actually, what
we found from that research was that

00:04:25.440 --> 00:04:29.580
there was almost a perfect relationship,
like every single cigarette count,

00:04:29.890 --> 00:04:34.120
and not only that, even one cigarette
a day would double the amount of the

00:04:34.120 --> 00:04:36.370
chances of having a child die of SIDS.

00:04:36.710 --> 00:04:38.490
It wasn't something they knew.

00:04:38.950 --> 00:04:42.895
It wasn't something they knew that
even pre-pregnancy smoking in the last

00:04:42.895 --> 00:04:47.165
trimester before getting pregnant would
also increase the chances of SIDS.

00:04:47.528 --> 00:04:49.497
In order to answer all these
questions, you needed that

00:04:49.497 --> 00:04:50.967
amount of data that we had.

00:04:51.407 --> 00:04:55.931
It could have been approached, but what
was difficult was the researchers that

00:04:55.931 --> 00:04:59.381
were working on this, these are amazing
researchers that have dedicated their

00:04:59.381 --> 00:05:02.527
lives to help and understand SIDS.

00:05:03.004 --> 00:05:06.197
They didn't have the knowledge or
capacity to deal with this amount of data.

00:05:06.534 --> 00:05:11.274
That was our contribution to the problem,
was helping them ask questions of

00:05:11.504 --> 00:05:12.644
data, answer questions from the data.

00:05:13.054 --> 00:05:16.684
That is something that is critical is in
order for us, and that's something that

00:05:16.684 --> 00:05:20.974
we repeat in every project we have, in
order for us to have an impact, we always

00:05:20.974 --> 00:05:23.194
need to partner with an expert, right?

00:05:23.244 --> 00:05:26.774
What is not recommended at all is to
say, "Hey, I have the data about this

00:05:26.774 --> 00:05:30.074
particular problem, I can work with
this problem without understanding

00:05:30.074 --> 00:05:33.164
the context, without understanding
how the data was collected, without

00:05:33.164 --> 00:05:36.884
understanding the potential biases of
the data." This is why it's so critical

00:05:37.184 --> 00:05:41.144
that for every project we have, we always
partner with an expert in the problem.

00:05:41.144 --> 00:05:43.364
We're not experts in the
problems that we work.

00:05:43.994 --> 00:05:48.464
We are working from pancreatic
cancer to wildfires to detection

00:05:48.464 --> 00:05:49.954
of plastic in the ocean.

00:05:50.224 --> 00:05:51.574
We are not expert in those problems.

00:05:51.604 --> 00:05:54.724
Our expertise is AI and machine
learning on computer vision.

00:05:54.904 --> 00:05:58.684
We partner with experts and we
work on those problems together.

00:05:59.072 --> 00:06:00.752
Susannah de Jager: So you're not
just predicting something, you're

00:06:00.752 --> 00:06:04.442
actually changing the question that
we can ask and the granularity.

00:06:05.081 --> 00:06:09.864
You already broadly brushed across
some of the really kind of wild breadth

00:06:09.864 --> 00:06:13.914
that you are able to apply yourselves
to when working with other experts.

00:06:14.484 --> 00:06:19.644
So what helps you narrow down what's
an AI for Good problem and what

00:06:19.644 --> 00:06:20.996
you actually apply yourselves to?

00:06:21.446 --> 00:06:23.396
Juan M Lavista Ferres: One of the
great things about the space that

00:06:23.396 --> 00:06:27.386
we are is that for majority of the
problems out there in the world,

00:06:27.786 --> 00:06:30.186
behind the scenes, you have data.

00:06:30.486 --> 00:06:34.836
And as long as there's data and you
have the expertise, usually the methods

00:06:34.836 --> 00:06:40.176
that we use, the AI models that we
use, could be used to help solve.

00:06:40.296 --> 00:06:44.346
And this is something that is particularly
interesting because a lot of programmes

00:06:44.346 --> 00:06:46.056
for us look the same, correct?

00:06:46.106 --> 00:06:47.636
And I'm gonna give you an example.

00:06:47.636 --> 00:06:51.116
So one of the projects that we have is
that whenever there's a natural disaster,

00:06:51.386 --> 00:06:55.639
like the one that just happened in
Colombia or in Venezuela, we partner with

00:06:55.639 --> 00:06:56.989
an organisation that have satellites.

00:06:56.989 --> 00:07:00.229
We put a satellite on top of a
natural disaster, we take pictures,

00:07:00.319 --> 00:07:03.589
and then we use AI mode to build what
is called damage assessment maps.

00:07:04.039 --> 00:07:07.639
These are maps that show every single
house and which ones were affected or not.

00:07:07.909 --> 00:07:13.129
These are maps that are critical for
people on the ground to help save lives.

00:07:13.629 --> 00:07:17.079
When you look at that problem, and you
look at the very different problem,

00:07:17.079 --> 00:07:20.229
I mentioned pancreatic cancer, that
is basically trying to find early

00:07:20.229 --> 00:07:24.399
indications of pancreatic cancer from a
CT scan, even though these two problems

00:07:24.399 --> 00:07:30.069
cannot be further apart, for us, those
problems are basically the same problem.

00:07:30.069 --> 00:07:33.849
You have a big image, whether that's
a satellite image or a CT scan that

00:07:33.849 --> 00:07:38.709
looks like an image to us, and what
you need to do is find a pattern in

00:07:38.709 --> 00:07:42.459
that image, whether that's a house
that was destroyed or something

00:07:42.459 --> 00:07:44.019
that is different in your pancreas.

00:07:44.529 --> 00:07:47.919
For us, from a machine learning
perspective, from an AI perspective,

00:07:48.339 --> 00:07:52.127
these problems are basically the
same problem and that happens in

00:07:52.127 --> 00:07:53.777
majority of the problems that we have.

00:07:53.987 --> 00:07:58.077
A lot of these problems are very different
problems from a societal perspective.

00:07:58.437 --> 00:08:01.587
When you work with some of these
researchers, some of them which

00:08:01.617 --> 00:08:06.597
are using some of this technology
already, they are not a, like at

00:08:06.597 --> 00:08:08.337
all in communication with the rest.

00:08:08.337 --> 00:08:12.927
Like the people that are working on
satellite data problems are solving very

00:08:12.978 --> 00:08:17.167
similar problems are the ones that are
solving a medical imaging on CT scans.

00:08:17.617 --> 00:08:21.307
But these two groups of people don't
talk to each other and one of the beauty

00:08:21.307 --> 00:08:25.667
things for us is to try to come up with
things that  are solving the satellite

00:08:25.667 --> 00:08:29.267
data world and apply to the medical
imaging world, because again, from a

00:08:29.267 --> 00:08:30.487
data perspective, they're the same.

00:08:30.907 --> 00:08:36.857
So the beauty of what we have is the power
to solve problems because these problems

00:08:36.857 --> 00:08:38.717
are very similar from a data perspective.

00:08:39.400 --> 00:08:41.807
Susannah de Jager: So given that
similarity and that you're not actually

00:08:41.807 --> 00:08:45.701
looking for a specific kind of data,
given that there's such a broad

00:08:45.791 --> 00:08:50.424
applicability of the data, how are you
narrowing down what you do focus on?

00:08:50.454 --> 00:08:52.447
Because it must be tyranny of choice.

00:08:53.094 --> 00:08:55.070
Juan M Lavista Ferres: Yes,
it's exactly the right.

00:08:55.070 --> 00:08:57.680
I haven't used that word before,
but I'm gonna start using it from

00:08:57.680 --> 00:08:59.000
now on because I actually like it.

00:08:59.130 --> 00:09:02.825
We have a lot of problems on having
a lot of options and part of that

00:09:02.825 --> 00:09:04.295
challenge is the focus, correct?

00:09:04.295 --> 00:09:08.124
So every Monday we get together with
a significant portion of my team.

00:09:08.292 --> 00:09:09.492
We review the request.

00:09:09.492 --> 00:09:13.332
We get requests from multiple places
in the world, from other colleagues,

00:09:13.332 --> 00:09:15.132
from people that knows of our work.

00:09:15.132 --> 00:09:18.912
Sometimes we do open calls, and we
review every one of these requests.

00:09:19.552 --> 00:09:24.042
In the basis of first, they need to have
an impact on society and impact is a

00:09:24.042 --> 00:09:30.432
bit different from places to places like
helping save the giraffes in Tanzania is

00:09:30.432 --> 00:09:34.762
very different to helping understand a
particular case of breast cancer, right?

00:09:34.762 --> 00:09:39.442
So we try to look for a portfolio
approach and then from these

00:09:39.442 --> 00:09:41.242
problems, we start asking questions.

00:09:41.242 --> 00:09:42.352
We go through a funnel.

00:09:42.532 --> 00:09:47.982
The first question is if we
have the data, do we think

00:09:47.982 --> 00:09:50.202
that AI can solve this problem?

00:09:50.442 --> 00:09:52.482
And that's something that
come from our experience.

00:09:52.972 --> 00:09:54.112
There's a lot of problems out there.

00:09:54.142 --> 00:09:57.492
Not all of them can be solved
using AI or computer science.

00:09:57.862 --> 00:09:59.752
That's basically the
first question we ask.

00:09:59.782 --> 00:10:03.888
The second question we ask is there
data out there that was collected

00:10:03.888 --> 00:10:05.328
that could help us solve the problem?

00:10:05.868 --> 00:10:10.135
A lot of problems out there, I
would say 70, 80% of the projects

00:10:10.195 --> 00:10:13.585
that could potentially be solved
using AI usually don't have data.

00:10:13.585 --> 00:10:17.035
So our first problem is usually a
lot of times is that there's no data.

00:10:17.835 --> 00:10:21.919
Once you have problems that have
data, the next question is do

00:10:21.924 --> 00:10:25.448
we think that we will be able to
get access to the data, correct?

00:10:25.448 --> 00:10:29.358
So for very good reasons, privacy
being one of them, there's a lot of

00:10:29.358 --> 00:10:33.408
projects out there where you could
potentially solve it using AI, but

00:10:33.408 --> 00:10:36.665
it's not possible to solve it because
there's no way for us to get access

00:10:36.665 --> 00:10:38.375
to data for all the reasons, correct?

00:10:38.604 --> 00:10:41.754
That's something that, again, we
can go through a whole podcast

00:10:41.754 --> 00:10:43.134
about that particular problem.

00:10:43.134 --> 00:10:46.224
But there's very good reasons
sometimes that, yes, the data exists,

00:10:46.614 --> 00:10:49.804
but it's not easy for us or it
might be actually impossible for us.

00:10:50.194 --> 00:10:55.084
For some projects, a lot of projects
will require what is called label data,

00:10:55.144 --> 00:11:00.062
which means that you have, for example,
in the case of like a wildfire, you

00:11:00.062 --> 00:11:04.892
have an expert that is, in this case, a
firefighter that will look at a video of

00:11:04.892 --> 00:11:09.122
a particular fire and will say, "Here,
what you're seeing here is the start of

00:11:09.122 --> 00:11:13.652
a fire." And that person will go to a
tool and will actually, what is called

00:11:13.652 --> 00:11:18.902
label, basically indicate which of these
pixels are from the beginning of fire.

00:11:19.142 --> 00:11:21.272
The same happened, for example,
in the case of pancreatic

00:11:21.272 --> 00:11:22.532
cancer, like I mentioned before.

00:11:22.832 --> 00:11:27.182
You have an expert that is looking at the
CT scan and is able to go and label that

00:11:27.182 --> 00:11:29.168
particular case is pancreatic cancer.

00:11:29.721 --> 00:11:31.471
This is what you call label data.

00:11:32.021 --> 00:11:34.871
A lot of problems in
AI require label data.

00:11:35.291 --> 00:11:38.834
Once we go through that, the other
question we have that is likely one

00:11:38.834 --> 00:11:42.944
of the most important is, do we have
the right expertise as a counterpart?

00:11:43.371 --> 00:11:46.911
Sometimes we get amazing projects, but
these projects are coming not from the

00:11:46.911 --> 00:11:50.841
experts, are coming from people that
basically, were affected by the programme.

00:11:51.381 --> 00:11:54.471
But in order for us to solve the
problem, we need to be working

00:11:54.471 --> 00:11:56.751
with a counterpart that is an
expert because we're not experts.

00:11:56.751 --> 00:11:58.101
We need an expert on the other side.

00:11:58.801 --> 00:12:01.561
So for example, I mentioned, again,
pancreatic cancer, we are working with

00:12:01.561 --> 00:12:05.911
some of the best doctors, these are
John Hopkins University doctors, that

00:12:05.911 --> 00:12:09.121
they have spent their lives trying
to understand pancreatic cancer.

00:12:09.121 --> 00:12:12.224
That clearly for us is
an amazing partnership.

00:12:12.554 --> 00:12:14.684
So we always need to
partner with an expert.

00:12:15.254 --> 00:12:20.324
Once we have that then the other question
we have is, if we solve this problem

00:12:20.804 --> 00:12:25.004
do we believe that we have the right
counterpart where we can do a knowledge

00:12:25.004 --> 00:12:31.304
transfer to them, and then they can
actually work and we continue on this.

00:12:31.704 --> 00:12:36.354
That one is likely the last part
of the funnel, and it's one that is

00:12:36.390 --> 00:12:38.370
usually the one that restricts more.

00:12:38.800 --> 00:12:41.470
So sometimes you have the right
partnership, you have the right data,

00:12:42.160 --> 00:12:45.670
but you ask yourself, "If we can solve
this problem, do we really believe

00:12:45.670 --> 00:12:49.060
that the organisation that we are
partnering will make it to production?

00:12:49.540 --> 00:12:55.190
Will mean, that if we solve it, can
they actually use these models to help?"

00:12:55.590 --> 00:12:59.170
Once you go through the whole funnel,
you could be starting with 100 projects,

00:12:59.650 --> 00:13:01.270
you will end up with two or three.

00:13:02.044 --> 00:13:07.197
From those two or three projects, we get
together and we say, we try to address

00:13:07.197 --> 00:13:11.757
this from a portfolio approach in the
sense that sometimes our projects are

00:13:11.757 --> 00:13:15.297
there that are extremely difficult,
we already have solved these problems,

00:13:15.297 --> 00:13:18.417
but if we can solve it, then that
will be a huge impact on society.

00:13:18.987 --> 00:13:22.167
You will have projects that maybe
they're not that impactful, but we have

00:13:22.167 --> 00:13:25.377
solved those similar projects before
where you know that you can solve it.

00:13:25.707 --> 00:13:28.977
So we try to look for a portfolio
approach where you have very difficult

00:13:28.977 --> 00:13:32.217
projects, you have projects that you
know you're gonna solve, you have

00:13:32.217 --> 00:13:33.867
also projects in different areas.

00:13:33.867 --> 00:13:38.607
I mentioned biodiversity, I mention
healthcare, I mention disaster response.

00:13:39.027 --> 00:13:41.680
We also look at a portfolio
approach from that.

00:13:42.302 --> 00:13:43.660
It's a very interesting meeting.

00:13:43.822 --> 00:13:47.180
It's a very difficult meeting in the
sense that you, like you said, you have

00:13:47.180 --> 00:13:50.150
a lot of options, you try to look for
the best options to have an impact.

00:13:50.550 --> 00:13:53.160
Susannah de Jager: And you're talking
there about portfolio and the language

00:13:53.160 --> 00:13:56.250
sounds very close to investing
and of course you are investing.

00:13:56.250 --> 00:13:59.430
You're investing your team's time
and indeed other people's time.

00:13:59.970 --> 00:14:03.960
Out of interest, the intellectual
property is it completely freely

00:14:03.960 --> 00:14:05.880
available once you've developed something?

00:14:06.580 --> 00:14:09.034
Juan M Lavista Ferres: We partner with
organisations and we don't take anything

00:14:09.034 --> 00:14:11.029
meaning that there's no IP for Microsoft.

00:14:11.299 --> 00:14:13.879
Everything we do, we
open source everything.

00:14:13.939 --> 00:14:16.189
Ideally, we also open source
the data when possible.

00:14:16.189 --> 00:14:19.549
Sometimes it's not possible to
open source data, but we publish

00:14:19.549 --> 00:14:21.578
the results in academic journals.

00:14:21.998 --> 00:14:24.578
So our contribution to society is that.

00:14:24.774 --> 00:14:27.774
I would say the big majority
of the organisations out there

00:14:28.074 --> 00:14:30.864
have some philanthropic efforts
within the organisation.

00:14:31.104 --> 00:14:36.154
What we realised was, if we could
actually donate our time, like Microsoft

00:14:36.154 --> 00:14:39.614
still does philanthropic efforts outside
our team, but we will say, look, can

00:14:39.614 --> 00:14:43.097
we carve part of that philanthropic
effort that we have and dedicate it,

00:14:43.097 --> 00:14:46.937
instead of donating, for example,
money, we would donate our efforts.

00:14:47.051 --> 00:14:50.771
The whole hypothesis, the whole premise
here, is that we believe that we can have

00:14:50.771 --> 00:14:55.391
a bigger return of investment for societal
perspective than making the donations to

00:14:55.391 --> 00:15:00.181
those organisations and the reason that
is key is because we believe that even

00:15:00.181 --> 00:15:03.501
if we sometimes we could donate some
of these organisations do not have the

00:15:03.501 --> 00:15:07.281
structure or capacity to hire the talent
that is needed to solve this problem.

00:15:07.281 --> 00:15:08.421
So that's the whole idea.

00:15:08.731 --> 00:15:09.751
We don't take a fee.

00:15:09.781 --> 00:15:11.221
We open source everything.

00:15:11.511 --> 00:15:14.480
So yes, everything that we do
is contributing to society.

00:15:14.917 --> 00:15:17.155
Susannah de Jager: I would say
people definitely wouldn't be able

00:15:17.155 --> 00:15:20.065
to hire the kind of talent that
you've aggregated in your teams.

00:15:20.275 --> 00:15:23.935
You'd have to be in an enormous
organisation and even if you're within

00:15:23.935 --> 00:15:29.425
an academic setting with experts in,
to your example, pancreatic cancer, the

00:15:29.425 --> 00:15:35.155
chances that you're able to couple them
with a team of your experience and I'm

00:15:35.155 --> 00:15:37.407
sure breadth and depth, there's no way.

00:15:39.202 --> 00:15:44.232
This Oxford+ miniseries is brought to you
by Equinox, Equitable Innovation Oxford.

00:15:45.012 --> 00:15:48.732
Equinox is a major regional partnership
established by the University of

00:15:48.732 --> 00:15:52.872
Oxford to drive inclusive innovation
and sustainable economic growth.

00:15:53.502 --> 00:15:56.562
Bringing together universities,
local authorities, government,

00:15:56.562 --> 00:16:00.822
industry, investors, and community
organisations, Equinox provides

00:16:00.852 --> 00:16:04.812
a unified voice for Oxfordshire,
championing innovation that delivers

00:16:04.842 --> 00:16:07.572
economic impact alongside social value.

00:16:08.082 --> 00:16:11.742
More than 80 organisations have signed
the Equinox charter, committing to

00:16:11.782 --> 00:16:15.582
attract investment, remove barriers
to growth, and ensure the benefits

00:16:15.582 --> 00:16:17.562
of innovation are shared more widely.

00:16:18.072 --> 00:16:22.962
By aligning Oxfordshire's world-leading
research, enterprise, and talent, Equinox

00:16:22.962 --> 00:16:27.792
aims to unlock nationally significant
growth, reinforce the county's position

00:16:27.792 --> 00:16:32.142
as a global innovation hub, and ensure
that success creates skilled jobs,

00:16:32.412 --> 00:16:36.852
attracts investment, and delivers
lasting prosperity across the region.

00:16:41.079 --> 00:16:45.339
There's one example that I have read
about that I really enjoyed, and it

00:16:45.339 --> 00:16:47.049
doesn't fit within your framework.

00:16:47.482 --> 00:16:51.562
There was some data, but it was very
hard to get, and you actually helped

00:16:51.562 --> 00:16:53.992
pioneer a way of collecting more data.

00:16:54.382 --> 00:16:57.873
I'd love you to talk about that because,
I'm being a little bit tongue in cheek,

00:16:57.873 --> 00:16:59.763
it does fit within your other frameworks.

00:17:00.123 --> 00:17:06.683
But SPARROW, as you named it with the
acronym, looking at biodiversity, was

00:17:06.683 --> 00:17:11.453
very much focused on the data collection
problem that you observed while working.

00:17:12.113 --> 00:17:13.298
So can you tell us a
little bit about that?

00:17:14.098 --> 00:17:17.109
Juan M Lavista Ferres: Yeah, so we
work a lot in biodiversity where a lot

00:17:17.109 --> 00:17:21.729
of bioconservationists will work and
collect data and then what we would

00:17:21.729 --> 00:17:26.199
help them were we would basically have
AI algorithms that process the data.

00:17:26.219 --> 00:17:31.043
For example, if you have like a person
that was studying spider monkeys in

00:17:31.043 --> 00:17:34.433
the Amazon, they would collect a lot
of the data from spider monkeys, and

00:17:34.433 --> 00:17:36.353
we would help them classify the data.

00:17:36.353 --> 00:17:41.633
For example, they would have 10,000
pictures of things and then we

00:17:41.633 --> 00:17:45.533
would build an algorithm that could
detect whether a picture had a

00:17:45.713 --> 00:17:47.393
spider monkey and would help them.

00:17:48.139 --> 00:17:51.941
Three years ago, I was in a meeting
and the person was explaining the

00:17:51.941 --> 00:17:56.471
hassle that they needed to go through
to collect the data and I realised

00:17:56.471 --> 00:18:02.176
that we were certainly helping solve
a very small part of the problems.

00:18:02.246 --> 00:18:06.373
When you hear the stories that these
conservationists, they need to go to the

00:18:06.373 --> 00:18:11.193
middle of the Amazon, it would sometimes
take them three, four days to put a camera

00:18:11.193 --> 00:18:16.233
trap or a bioacoustic recording devices in
the middle of the Amazon, then come back.

00:18:16.533 --> 00:18:19.463
That, basically, they will lose
sometimes a full week from the

00:18:19.463 --> 00:18:24.753
going to put this and they need
to go back three weeks after that.

00:18:25.393 --> 00:18:29.143
They're doing the same trip that
in many ways, sometimes it's very

00:18:29.143 --> 00:18:33.403
dangerous, and to go and change the
batteries and get the memory cards.

00:18:33.403 --> 00:18:36.343
And once they got the memory card,
they need to go through back,

00:18:36.353 --> 00:18:39.793
and we, if they were lucky, they
had the data collected rightly.

00:18:40.233 --> 00:18:44.493
From the time that they started the
project and putting the camera to

00:18:44.496 --> 00:18:47.801
the time that they would analyse
the data, sometimes that whole

00:18:47.801 --> 00:18:51.197
process was like, it could be a
year, because of all the hassle.

00:18:51.377 --> 00:18:55.247
So it became evident for us that
we were solving the wrong problem.

00:18:55.277 --> 00:18:59.207
We wanted to help conservationists
in the world, we needed to help them

00:18:59.207 --> 00:19:03.347
collect data and there's roughly
200,000 conservationists in the world.

00:19:03.347 --> 00:19:06.137
All of them share one thing in common,
to do their job, they need data.

00:19:06.677 --> 00:19:10.487
And in many ways conservation
moves at the speed of that data.

00:19:10.567 --> 00:19:12.547
That's the problem that
we started with SPARROW.

00:19:12.547 --> 00:19:14.947
It's like, can we actually make it
easier for them to collect data?

00:19:15.487 --> 00:19:19.087
SPARROW is a small device.

00:19:19.100 --> 00:19:23.464
It's a small computer that lives on the
edge, meaning that they go to the middle

00:19:23.464 --> 00:19:25.294
of the Amazon and they instal this device.

00:19:25.744 --> 00:19:26.974
The device is solar power.

00:19:26.974 --> 00:19:28.384
It has a solar panel on top.

00:19:28.874 --> 00:19:29.664
It has a small GPU.

00:19:30.062 --> 00:19:32.954
GPU is what is used to
run these AI algorithms.

00:19:33.444 --> 00:19:36.177
It acts in the middle of nature
connecting to camera traps,

00:19:36.177 --> 00:19:38.067
connecting to bioacoustic devices.

00:19:38.547 --> 00:19:42.417
It process the information there on
the edge, and it sends the result back

00:19:42.897 --> 00:19:45.357
using a satellite up link to the cloud.

00:19:45.867 --> 00:19:48.647
And once it's in the cloud,
the conservation is can

00:19:48.647 --> 00:19:50.813
access real time that data.

00:19:51.443 --> 00:19:53.513
Once you install the device,
you no longer need to go back.

00:19:53.513 --> 00:19:57.053
So we have some of these devices that
right now been running for over a year

00:19:57.053 --> 00:20:01.083
in the middle of the Amazon, for example,
and our plan is to continue making this

00:20:01.083 --> 00:20:06.113
as simple as possible for conservationists
to collect data and to process data.

00:20:06.530 --> 00:20:09.432
Susannah de Jager: And it brings me onto
another example of your work that I've

00:20:09.432 --> 00:20:13.272
already had the pleasure of learning a
little bit about, which is with the Oxford

00:20:13.272 --> 00:20:15.432
Institute of Technology and Justice.

00:20:15.822 --> 00:20:19.872
And when I was interviewing Phillipa
Webb, she was talking about the apps that

00:20:19.872 --> 00:20:25.352
you're co-developing for Malawi, helping
helping women and girls who are subject

00:20:25.382 --> 00:20:28.532
either to abuse or forced marriages.

00:20:29.028 --> 00:20:34.392
And I think that what struck me in
that conversation was how many of the

00:20:34.392 --> 00:20:40.365
things that you were building in were
very simple and most importantly,

00:20:40.365 --> 00:20:42.285
were informed by people on the ground.

00:20:42.285 --> 00:20:47.375
So the fact that you could build something
incredibly complex, but if it doesn't work

00:20:47.375 --> 00:20:50.720
on 2G, it wouldn't work for that cohort.

00:20:51.050 --> 00:20:54.215
Juan M Lavista Ferres: Yes, this is one
of our best, most amazing projects we

00:20:54.215 --> 00:20:58.955
have collaboration with Phillipa and with
Amal and with her teams have been amazing.

00:20:59.375 --> 00:21:02.931
Particularly the fact that we work,
like you said, one of the key things

00:21:02.931 --> 00:21:07.091
that we have is that the co-development
of working with people in Malawi.

00:21:07.395 --> 00:21:11.711
So, the project that we have is Malawi
is a country that has roughly 22 million

00:21:11.711 --> 00:21:14.591
people and have around 800 lawyers.

00:21:15.041 --> 00:21:18.453
To have a notion like a lot of the
law firms out there in England or

00:21:18.453 --> 00:21:22.343
in US, we have significantly more
lawyers than the whole country.

00:21:22.523 --> 00:21:25.433
And the challenge with only had 800
lawyers is that there's not enough

00:21:25.433 --> 00:21:29.903
lawyers of course to help majority of
the people, particularly the people that

00:21:29.903 --> 00:21:33.493
are living in rural areas, particularly
the people that will be very difficult

00:21:33.493 --> 00:21:37.603
for them to even afford a lawyer,
impossible for them to afford a lawyer.

00:21:38.093 --> 00:21:42.657
These are women and children that
used to be suffering from either

00:21:42.657 --> 00:21:46.337
child marriage or domestic violence.

00:21:46.877 --> 00:21:49.910
They need help and a lot of times
they're not informed, they don't

00:21:49.910 --> 00:21:51.530
have the right legal information.

00:21:52.220 --> 00:21:55.147
And we've been working on these,
the collaboration is with the

00:21:55.147 --> 00:21:57.557
WLA team, this is the Women
Lawyers Association in Malawi.

00:21:57.557 --> 00:22:01.887
This is a group of women lawyers
that are doing an amazing job help

00:22:01.887 --> 00:22:03.447
these children, help these women.

00:22:03.447 --> 00:22:04.957
But of course, it doesn't scale.

00:22:05.137 --> 00:22:07.087
They're a very small group of lawyers.

00:22:07.527 --> 00:22:11.956
For us, what we want to do when we're
working with them is to try to make

00:22:12.496 --> 00:22:17.620
these lawyers and make these process as
productive as possible using AI, using

00:22:18.450 --> 00:22:22.740
large language models, to provide the
people that they have, the volunteers

00:22:22.740 --> 00:22:25.140
that they have, with the best information.

00:22:25.140 --> 00:22:30.794
So a person from using a like you
mentioned, even a 2G, WhatsApp

00:22:30.794 --> 00:22:33.554
conversation is able to communicate.

00:22:33.944 --> 00:22:35.744
Really we want it even in Chichewa.

00:22:35.744 --> 00:22:40.327
Right now the project still works mainly
in English, but we are working to have

00:22:40.327 --> 00:22:44.847
this ability to speak in Chichewa,
have the ability to ask questions,

00:22:44.847 --> 00:22:51.397
get legal information, and help those
people on the ground, provide the best

00:22:51.397 --> 00:22:53.707
information to these women and children.

00:22:54.587 --> 00:22:55.877
It's an amazing project.

00:22:55.977 --> 00:22:59.047
Like you mentioned, what if you let
the group of engineers solve a problem,

00:22:59.047 --> 00:23:00.697
they might solve it completely wrong.

00:23:00.697 --> 00:23:03.067
For us, what is critical
is to work with the expert.

00:23:03.067 --> 00:23:07.987
This is people like Phillipa and like
Amal and like their team and work with

00:23:07.987 --> 00:23:10.927
the people that are working with the
lawyers, and these are the people in

00:23:10.927 --> 00:23:12.847
Malawi, and work all of us together.

00:23:13.627 --> 00:23:15.457
It has been an amazing experience.

00:23:15.550 --> 00:23:17.792
Very excited, very happy
about that project.

00:23:18.002 --> 00:23:18.782
Susannah de Jager: It's amazing.

00:23:19.205 --> 00:23:23.435
You have already spoken, so it's a little
bit unfair about a portfolio and that

00:23:23.435 --> 00:23:25.505
there's clearly so many amazing impacts.

00:23:25.505 --> 00:23:28.745
But is there one area where you are.

00:23:29.015 --> 00:23:31.475
I don't even want to say where it's
your favourite area, but where you're

00:23:31.475 --> 00:23:36.035
more excited about the potential
impacts of what your team is able to do?

00:23:36.920 --> 00:23:38.821
Juan M Lavista Ferres: It's like
asking which is your favourite child.

00:23:39.241 --> 00:23:40.850
I don't think I have a one favourite.

00:23:41.340 --> 00:23:45.526
The project that we have in Malawi
is one that can scale extremely

00:23:45.526 --> 00:23:49.876
well, not only to Malawi, but this
is a problem that also can be in the

00:23:49.876 --> 00:23:52.867
Philippines, in Bolivia, in Cameroon.

00:23:53.107 --> 00:23:56.437
It's a project that could
potentially be working worldwide.

00:23:56.617 --> 00:23:59.527
I like projects that have that magnitude.

00:24:00.187 --> 00:24:04.457
My background is the intersection
between AI and healthcare.

00:24:04.557 --> 00:24:10.482
I always have an amazing appetite
to help on that intersection between

00:24:10.482 --> 00:24:13.362
healthcare and health in general and AI.

00:24:13.712 --> 00:24:16.922
But it's one of the most challenging
to solve what is called problems in

00:24:16.922 --> 00:24:21.222
production because you need to go
through a lot of effort to make sure

00:24:21.222 --> 00:24:25.752
that eventually, for all good reasons,
to make sure that the these solutions can

00:24:25.752 --> 00:24:27.912
be used in a production setting, correct?

00:24:28.502 --> 00:24:32.049
But for example, there's a project
that we've been working now for the

00:24:32.049 --> 00:24:35.507
last four or five years, a project
that I started myself as an individual

00:24:35.507 --> 00:24:39.257
contributor in this project that is
called Retinopathy of Prematurity.

00:24:39.497 --> 00:24:42.647
Retinopathy of Prematurity is the
leading cause of blindness in children

00:24:43.157 --> 00:24:44.897
in a lot of countries around the world.

00:24:45.467 --> 00:24:49.157
This is a disease that didn't exist a
few decades ago and now we're seeing

00:24:49.157 --> 00:24:50.957
a huge increase across countries.

00:24:51.617 --> 00:24:55.577
And interesting enough was one of my
aha moments because I asked how is

00:24:55.607 --> 00:24:58.487
it both in that we have a problem now
that didn't exist a few decades ago?

00:24:59.147 --> 00:25:04.337
And Retinopathy of Prematurity affects
very small premature babies and

00:25:04.487 --> 00:25:06.707
these babies wouldn't survive before.

00:25:07.367 --> 00:25:10.208
So now more and more improvements
in health and healthcare, more and

00:25:10.208 --> 00:25:11.867
more of these babies are surviving.

00:25:12.527 --> 00:25:15.107
But the challenge is that a lot
of these babies are not ready

00:25:15.107 --> 00:25:16.307
to live on the planet yet.

00:25:16.307 --> 00:25:20.847
So the retinas are not fully developed
and some of them will suffer from ROP.

00:25:21.147 --> 00:25:22.989
This is a Retinopathy of Prematurity.

00:25:23.329 --> 00:25:26.466
If they are diagnosed and they
are treated the blindness through

00:25:26.466 --> 00:25:27.906
these is completely preventable.

00:25:28.206 --> 00:25:32.256
The challenge is that they have a
very small window of time, roughly

00:25:32.256 --> 00:25:36.126
between 24 to 36 hours, that if they
are not treated during that time,

00:25:36.576 --> 00:25:38.196
that baby will be blind forever.

00:25:38.646 --> 00:25:42.546
But the challenge is that you only have
in the world 200,000 ophthalmologists.

00:25:42.996 --> 00:25:44.526
These are not paediatric ophthalmologists.

00:25:44.526 --> 00:25:47.406
From these 200,000 roughly 10,000
are paediatric ophthalmologists,

00:25:48.006 --> 00:25:49.176
which is physically impossible.

00:25:49.176 --> 00:25:51.996
Like you don't have enough
ophthalmologists to even

00:25:51.996 --> 00:25:53.256
diagnose these disease.

00:25:53.346 --> 00:25:55.629
Every one of these babies
needs to be screened.

00:25:56.266 --> 00:25:57.256
It's physically impossible.

00:25:57.316 --> 00:26:01.296
So we have done AI models that
can show that can diagnose this.

00:26:01.316 --> 00:26:04.826
We have them running on a smartphone
and we are working with some

00:26:04.826 --> 00:26:07.316
of these amazing doctors that
they know that they don't scale.

00:26:07.316 --> 00:26:11.986
They have dedicated lives to help make
sure that there are no blind babies

00:26:11.996 --> 00:26:13.659
but there's no way for them to scale.

00:26:14.276 --> 00:26:17.886
That is one of the projects that I
feel most passionate about because I

00:26:17.886 --> 00:26:21.666
see that this is a problem that not
only AI is a solution, but it's the

00:26:21.666 --> 00:26:23.966
only solution we have as a society.

00:26:24.386 --> 00:26:27.646
So we've been working a lot on
the programme and we have hope

00:26:27.676 --> 00:26:30.826
that eventually we'll have that
impact in society in production.

00:26:31.113 --> 00:26:31.193
Susannah de Jager: So.

00:26:31.277 --> 00:26:34.817
In both of those examples that we've
just discussed, and I'm by no means

00:26:34.847 --> 00:26:38.717
extrapolating that this is true of
all of them, but it's very much that

00:26:38.717 --> 00:26:43.937
there's a sort of supply bottleneck
that AI is able to solve for.

00:26:44.527 --> 00:26:48.697
And I can see that's got a huge
amount of value in so many areas.

00:26:49.497 --> 00:26:52.797
You've spoken a little bit
about what perhaps doesn't work.

00:26:53.597 --> 00:26:58.247
You work with AI all the time and
you've got a team who are pointing

00:26:58.247 --> 00:26:59.730
it in the right directions.

00:27:00.320 --> 00:27:04.580
When you look at these issues
and its power, what worries you

00:27:04.760 --> 00:27:06.727
about the application of AI?

00:27:07.527 --> 00:27:10.470
Juan M Lavista Ferres: Every project
where we have we'll go through, what

00:27:10.470 --> 00:27:12.240
is called, a responsible AI framework.

00:27:12.270 --> 00:27:15.120
Microsoft was one of the first
companies that actually have a

00:27:15.120 --> 00:27:16.680
Chief Responsible AI Officer.

00:27:17.310 --> 00:27:18.930
There are things that can go wrong.

00:27:19.227 --> 00:27:23.847
Not by my team, but other teams
in the world were using AI

00:27:23.847 --> 00:27:27.417
to help diagnose skin cancer.

00:27:28.125 --> 00:27:33.696
One of the things that happened was that,
a lot of the, the data that was collected

00:27:33.696 --> 00:27:37.386
for skin cancer was from Caucasians,
because Caucasians in the US usually

00:27:37.386 --> 00:27:39.006
are the ones that go to dermatologist.

00:27:39.006 --> 00:27:42.066
There's more chances for them to
have skin cancer to begin with.

00:27:42.666 --> 00:27:47.586
But the challenge was these models will
work well in people that are Caucasians.

00:27:47.586 --> 00:27:51.439
They will not train on data from
Latinos or for Asians or for African

00:27:51.439 --> 00:27:56.327
Americans, which means that these models
might not even work for those places.

00:27:56.687 --> 00:27:59.927
Which means that if they don't work
on the wrong side, they might tell

00:27:59.927 --> 00:28:03.617
you that either you have cancer where
you don't, or worse, that you don't

00:28:03.617 --> 00:28:05.177
have cancer where you have, correct?

00:28:05.207 --> 00:28:09.680
This is a case that the model has
been trained on a subsample of

00:28:09.685 --> 00:28:13.160
the data that represents society
which means that it only will work

00:28:13.160 --> 00:28:15.350
well in that subsample of data.

00:28:16.040 --> 00:28:19.760
Every project that we have go through
a process to understand potential

00:28:19.760 --> 00:28:23.250
biases, potential areas where
we should not be using the data.

00:28:23.700 --> 00:28:28.770
Also potential bad cases of technology
in that you can build a technology to try

00:28:28.770 --> 00:28:34.940
to solve a problem, but what if people
use the same technology to make it worse?

00:28:34.990 --> 00:28:37.270
And SPARROW is a great example of that.

00:28:37.540 --> 00:28:39.910
One of the only indications
that we have, we don't want this

00:28:39.915 --> 00:28:44.470
technology to be used by people that
are dedicated to hunting, right?

00:28:44.650 --> 00:28:46.390
This would be actually useful for them.

00:28:46.810 --> 00:28:49.990
So we have clearly terms of
service on that describe that.

00:28:49.990 --> 00:28:53.260
But not only that, we only
partner and we only allow

00:28:53.500 --> 00:28:57.070
organisations that are biodiversity
organisations to work with us.

00:28:57.510 --> 00:29:00.876
We also need to build these responsibly,
make sure that we have the right

00:29:00.906 --> 00:29:04.566
questions, we partner with the right
people, we understand potential biases, we

00:29:04.566 --> 00:29:06.336
understand the impact of potential biases.

00:29:06.586 --> 00:29:11.546
Every tool that we build could
potentially be used as a weapon and

00:29:11.546 --> 00:29:15.326
I think it's important for us as a
process to make sure that we maximise

00:29:15.326 --> 00:29:18.956
the use of this tool, but minimise
the potential be using as a weapon.

00:29:19.564 --> 00:29:21.981
Susannah de Jager: And that
obviously speaks to, as one would

00:29:21.981 --> 00:29:26.961
hope, what your team is doing both
inside a large corporate, but also

00:29:26.991 --> 00:29:31.971
given the principles around which
your kind of entity was founded.

00:29:32.651 --> 00:29:38.501
Do you feel confident outside when
you look at the evolution of AI more

00:29:38.501 --> 00:29:43.074
broadly, that other people are gonna
have similar handbrakes in place?

00:29:43.541 --> 00:29:45.371
Juan M Lavista Ferres: When you
read the news, a lot of people are

00:29:45.371 --> 00:29:48.201
very concerned and sometimes they're
concerned for the right reasons.

00:29:48.201 --> 00:29:49.611
I'm not questioning that.

00:29:50.078 --> 00:29:54.388
The reason I'm usually optimistic is
that if you look for a random sampler,

00:29:54.388 --> 00:29:59.328
let's say a thousand people in the
world, almost everybody will usually

00:29:59.328 --> 00:30:01.158
want to live in a better world.

00:30:01.208 --> 00:30:03.971
As humans, we want the best in others.

00:30:03.971 --> 00:30:06.094
We want the best in society.

00:30:06.194 --> 00:30:08.684
And of course, I'm not saying
that everybody's like that, that

00:30:08.684 --> 00:30:11.894
you will find some bad cases.

00:30:12.384 --> 00:30:16.674
But in general, I would say majority
of the people in society want to

00:30:16.674 --> 00:30:20.454
use these for right reasons and
want to use these to help and

00:30:20.454 --> 00:30:22.224
that's why I'm usually optimistic.

00:30:22.934 --> 00:30:24.314
Susannah de Jager: I'm
pleased to hear you say that.

00:30:24.314 --> 00:30:28.364
I often say this to my son who's
become very stranger danger aware.

00:30:28.604 --> 00:30:31.334
And I keep saying to him, I said,
yes, my love, you must be careful.

00:30:31.334 --> 00:30:32.924
But most people are good people.

00:30:33.565 --> 00:30:34.795
Thank you, Juan.

00:30:34.795 --> 00:30:40.925
I've really enjoyed this and I'm so
pleased that AI in your hands is looking

00:30:40.925 --> 00:30:43.475
to tackle so many amazing problems.

00:30:43.475 --> 00:30:45.185
I can't wait to see more
of what you put out.

00:30:45.638 --> 00:30:46.028
Juan M Lavista Ferres: Thank you.

00:30:46.028 --> 00:30:47.468
Thank you very much for the interview.

00:30:48.638 --> 00:30:52.148
Susannah de Jager: Thank you for listening
to this episode of Oxford+, hosted by

00:30:52.158 --> 00:30:56.398
me, Susannah de Jager If you want to
keep up with all things Oxford+, visit

00:30:56.398 --> 00:31:01.418
our website, oxfordplus.co.uk or sign
up for our newsletter on Substack.

00:31:01.918 --> 00:31:04.998
Oxford+ is a podcast produced
by Story Ninety-Four.