0:01

Jonah: Too Long, Didn't Read. Brought to you by the Alan Turing Institute.

0:05

The National Institute of Data Science and AI.

0:10

Hello, welcome to a Too Long, Didn't Read special.

0:13

I'm Jonah, a content producer here at the Alan Turing Institute.

0:17

Smera: And I'm Smera, a Research Assistant in Data Justice

0:19

and Global Ethical Futures. Jonah: Ahead of our second season of Too Long, Didn't Read, we wanted to ease you

0:25

in with this special episode all about something close to our hearts, or at

0:28

least something that has been consuming our workouts for the last few months.

0:32

AI UK, the national showcase of data science and AI brought to

0:36

you by the Alan Turing Institute. On this episode, we're going to explore how drag shows can use

0:42

AI to explore society's biases.

0:44

How the UK could be the best place to further AI in healthcare and how children

0:49

are helping shape the AI of the future. Smera: AIUK is our annual two day event and this year it took place in

0:56

March at the QEII Centre in London.

0:58

It was packed to the brim. Brim with interactive demonstrations, workshops, and talks from some of the

1:04

leading minds in data science and AI.

1:06

You can imagine the setting Jonah, you're in Westminster, houses of

1:10

parliament across the street, British think tanks lining the streets of

1:13

White Hall, the crisp spring weather, and most importantly, data science.

1:18

Jonah: You make it sound very romantic. But, uh, we were kind of running around like headless chickens trying

1:22

to get interviews and things like that. So ARUK is open to the public and the entire event was streamed online,

1:29

free for all, but the audience does tend to be predominantly researchers,

1:33

business and government people. So since there was so much good and relevant stuff, we decided

1:38

to dedicate an entire episode to what we learned and who we met.

1:46

Smera: So to kick us off, we are joined by the one and only Lily

1:49

Hughes, the producer behind ARUK.

1:53

Welcome, Lily. I imagine AIUK was a lot of work.

1:57

Have you recovered? Hi, Samira. Lilian: Thanks so much for having me.

2:00

I don't think anyone ever really recovers from AIUK, but I am doing much better now.

2:04

I've rested, the sun's shining, back to my normal work hours.

2:08

It's all good. Jonah: Well, congrats on your work, Lily.

2:11

Um, what is AIUK for our listeners and why is it relevant?

2:15

Lilian: What AIUK is, it's really an opportunity to bring all these

2:18

different people from across the AI ecosystem in the UK together.

2:22

So Turing works with. academics, with researchers, with universities across the UK.

2:28

We work with industries, with corporate partners.

2:30

We work with government, policy makers, civil servants, and we

2:33

work with the third sector as well. And very rarely do events bring all of those different people together to talk

2:40

to each other and learn from each other. And that's what AIUK does.

2:44

Great. Yeah. Meeting of minds. A meeting of minds.

2:47

That's really cool. Thank you. Smera: Well, amongst the minds that were met, how is AIUK actually put together?

2:52

Who decides who, you know, which minds receive the showcase to

2:56

talk, discuss, and demonstrate some of the work they're doing?

2:59

Lilian: AIUK is a work of so many people.

3:03

So there's a couple of different groups that I work with the most,

3:05

I would say the most important of which is the program advisory group.

3:09

So this is a group that volunteers their time from Turing.

3:12

It's both business leaders at Turing and researchers at Turing, uh,

3:15

from all different levels across, across our, across the Institute.

3:19

They come together. They think about what we should be platforming at AIUK.

3:23

And then we go from there. What's possible, what isn't possible, how do we want to shape it, working together

3:29

to build those ideas, develop them, and eventually showcase them at AIUK.

3:33

And a showcase it Smera: was. Jonah: Yes, it was, uh, a lot of people involved and.

3:37

We were all some of them. Smera: So you've seen at least two AIUKs through.

3:42

So what's different about this year? What was your favorite bits?

3:45

Well, what are some main highlights that you came across and what

3:48

was different about AIUK 2024?

3:51

Lilian: I don't know if I can ever choose sort of favorite sessions.

3:54

They're all brilliant in their own way, but there were some really

3:57

fun things that we did this year. I worked, as I said, with the program advisory group to come up with

4:01

some content that's a little bit different from what we'd done before.

4:04

One of the things that I really enjoyed doing this year

4:07

was the opening provocation. This was a project spearheaded by Drew Hemnett, uh, called The New Real.

4:13

So we worked with Jake Elwes and collaborators to present a kind

4:18

of positive utopia of AI, a bit different from the, the fear you

4:25

sometimes hear about existential risk.

4:27

We were really thinking about how humans and AI and society, all of

4:32

this is going to come together. And there is a, there's an opportunity here for positive change.

4:37

And I wanted to encourage everyone at the event to be curious about

4:40

AI, to think about it in a way they might not have approached it before.

4:43

So we did this as the opening provocation to really start off on

4:47

that foot, celebrating AI, celebrating artists and celebrating being

4:51

Smera: curious. That's brilliant. I think that was a phenomenal show.

4:54

We did manage to catch up with me, the drag queen and the project

4:58

creator, Jake Elwes backstage at AIUK.

5:02

And we asked them why drag is a good vehicle to explore AI.

5:06

Me the Drag Queen: We're throwing rhinestones and wigs and big makeup

5:09

on scary tech so that people can understand that there are amazing ways

5:12

to use it as well as insidious ones. Jake: Yeah, and drag is a great way of doing that, because I guess AI systems

5:16

have a lot of bias towards normativity. So for us, if we're injecting drag kings and drag queens and drag things, gender

5:24

non conformity into those systems, it's a wonderful way of exploring, kind

5:29

of, biasing the AI towards otherness rather than normativity, and breaking

5:33

it down, and seeing when it glitches, and finding poetry, and when it fails.

5:37

I think Me the Drag Queen: I need technology, any, anything that impacts society in

5:42

such a way has to represent and reflect. all of society.

5:46

You can't just have it modeled on the majority because the rest of us do exist.

5:50

We're here. There may not be many of us, but we're here and we're kind of more

5:54

fabulous than you, so get us involved. I guess for me, it ain't a party if the queer's aren't there, darling.

5:59

So Smera: me, the drag queen and Jake, they also told us about how

6:03

AI has influenced their own work. I think this is important because we see now why representation is

6:09

important, especially in places where queerness is often on the margins,

6:13

if it is even included at all. Jake: I've been working, yeah I guess I've been working with like AI machine learning

6:18

for like coming up to 10 years now. Um, so right in the early days I was like AI in its infancy, the

6:25

earliest generative adversarial networks after Ian Goodfellow's paper.

6:29

I was in the basement of my art school like programming these systems and

6:32

hacking with them and and kind of finding poetry and when they failed and broke

6:37

down and back then it was like could only generate these tiny little images.

6:41

And I guess my thinking around AI has changed a lot, like I used to very much be

6:45

interested in, in these kind of questions of agency, uh, and consciousness and

6:49

how, how much kind of agency can I give the computer as an art making machine.

6:55

But then, you know, As I kind of carried on researching in this field and realizing

6:59

that questions about who's building these systems and who are they building

7:02

them for became more the forefront.

7:05

Looking at bias, looking at far more pragmatic, what's in the

7:07

data set, whose data set is it? Is it my data set?

7:10

I think early on it was about appropriating large data sets and kind

7:13

of making them fail and pointing out sometimes political things in that sense.

7:17

But then it became far less about the sort of metaphysical questions around

7:21

AI and more about the kind of pragmatic, How is this affecting people right now?

7:25

And how can we, yeah, offer alternate visions of kind of AI features?

7:32

Jonah: It's interesting, isn't it? This, it sort of shows how art is one of the first methods to take new

7:36

technologies apart and like question the ethical side of how they're used.

7:40

Maribeth Rohr, an artist and research engineer at Google DeepMind, also

7:44

fresh off the stage, told us about how AI relies so much on categorization

7:49

and how queerness can encourage us to think beyond categories.

7:52

Maribeth: Queer representation is important in AI because it's, um,

7:56

one of those spaces that breaks down the binaries that we have and the,

8:00

the fixed categories that we have. I think queerness pushes the boundaries of categorization

8:05

into spaces that are more fluid.

8:07

And in AI, we often are putting things into categories and boxes.

8:12

Classification was a, is still a very like common application of

8:15

AI and queerness challenges that. And it's important to be.

8:19

critically interrogating our technical systems. And so queerness brings that perspective and that's really important for building

8:24

like societally responsible and just also better performing AI systems.

8:30

Jonah: Such a cool way to start AI UK, Lily.

8:33

Um, I imagine it challenged a lot of people's preconceptions

8:37

that opening provocation. Um, we'll link.

8:41

Jake's work in the show notes, so be sure to check that out.

8:45

So that was something new for this year's ARUK.

8:47

Something else that was new for 2024 that wasn't there in 2023 were workshops.

8:52

Am I right with that? Lilian: We actually had some workshops in 2023, but they were only an hour long.

8:57

They were very limited in scope. This year we really expanded it.

9:01

We had seven two hour workshops throughout the two days, and

9:05

they were really designed to be. These very interactive, very intensive sessions, uh, where we could, they

9:14

were designed to solve problems. If that makes sense, I didn't want the workshop to be another panel.

9:18

We had plenty of space for panels on the stages.

9:21

So what is a workshop? What are you doing there? Are you building a community?

9:24

Are you finding new projects? Are you solving a problem?

9:28

Are you getting into data into research?

9:31

All these things, I was like, that's all on the table.

9:33

And there were some hugely creative workshops at AIUK.

9:36

I'm really, really proud of this part of the program, actually.

9:39

Jonah: Yeah. I thought they were really cool. Like I was running around with my camera and I spent quite a lot of time in,

9:43

um, the Lego play workshop where, uh, they were sort of building things to

9:48

discuss cyber defense and it was so good.

9:50

There was so much interaction. Lilian: And I think one of the things about events like this, and

9:54

actually this is, it calls back to what Jake was doing as well.

9:57

That, you know, these. events, these showcases, they're opportunities for us to experiment

10:02

and play outside of the everyday.

10:05

We can try things we haven't done before. We can meet people we haven't met before.

10:08

And I think if you're all around a table, all playing with Lego or all brainstorming

10:12

ideas with, you know, with strangers, with new friends, if you're watching a draft,

10:17

a deep fake drag queen, you know, it's going to challenge what you want to do.

10:21

What you think and what you encounter in your every day.

10:24

And it's so important to do that work. It's so important to experiment and to play, especially in a field like AI.

10:30

So I think the workshops were really trying to encourage that.

10:33

Jonah: During another workshop, I spoke to Rebecca Cosgriff, who is the deputy

10:37

director for the data for research and development program at NHS England.

10:42

Um, Rebecca was a lead. on a workshop about unlocking healthcare data for safe, transparent, and fair AI.

10:48

And she told me what she was learning from having this interactive sort of session.

10:51

Rebecca: Yeah, I think one of the key learnings for me today was that there

10:54

was actually really significant consensus across England, Scotland, and Wales, which

10:59

were represented on the panel, but also on some of the table discussions across

11:03

industry, academia, and the NHS on some of the key enablers of AI, including

11:08

things like the proactive curation of data before it's provided out to researchers.

11:13

Um, it's really important that organizations like the Alan Turing

11:16

Institute have rapid access to granular, multimodal healthcare

11:20

data generated by the NHS and other data sources to answer some of our

11:24

really key imperative questions on how we improve care for patients,

11:29

support the NHS and drive innovation.

11:31

Smera: Speaking about healthcare data, there was also another panel

11:35

on improving disease detection.

11:37

Chair of the Board of the Turing and Director of the Natural History

11:41

Museum, Doug Gurr was on this panel and he told us why the UK is not only

11:45

best place, but probably the only place where such advances can happen.

11:49

Doug: So healthcare data is probably the most sensitive area.

11:52

You've got to be able to bring patients with you.

11:54

You've got to bring trust. And that's why you need a regulatory environment that can reassure.

11:59

that we're going to do this in an ethically sound, sensitive, safe

12:02

way, but at the same time a way that doesn't constrain that innovation.

12:06

And the UK is, I would say probably, but actually I'm going to go for

12:10

certainly, the best place in the world to do this, because only really in the

12:14

UK do you have those amazing data sets. And So with the opportunity to bring together the data science talent, the

12:20

clinical talent, get the government involved and actually bring everybody

12:23

around the table so that for the first place in the world, we can truly reap the

12:28

benefits of what AI can do for healthcare.

12:30

Smera: The element of trust raised by drug is very crucial with the

12:33

technology we are confronting. Almost daily we see reports on AI for good, especially in

12:38

healthcare, but we also see how AI has the capacity to disrupt labor

12:43

markets, influence the economy. Giving a lot of people reasons to be skeptical.

12:47

Hence, I think that trust factor should be at the center of developing an

12:51

inclusive and equitable path forward, particularly in using AI for healthcare.

12:56

Jonah: Yeah, definitely. So when I was exploring the demonstration area, I caught up with a team who are

13:01

developing digital twins of hearts.

13:03

So you can have a digital copy of your own heart with.

13:06

all the accurate medical data, and then they can run simulations of, say,

13:11

different drugs or stresses on your heart and see how it would respond.

13:15

That's pretty cool, isn't it? Smera: I think my own heart may be too broken for a digital twin.

13:20

Ah, Samara! Anyway, this is phenomenal work.

13:23

It was really exciting to see them 3D printing these hearts right

13:27

in front of us on the expo floor.

13:30

But beyond just the excitement of seeing a 3D printed heart, the future

13:33

of this work could revolutionize health care for a very particular group.

13:38

Can you guess which one? Jonah: Is it a vulnerable group?

13:40

It is a vulnerable group. Do you want to get specific?

13:43

Is it, um, children?

13:47

Children. Smera: So we can use digital twins to advance children's health care without

13:52

actually involving real human children.

13:55

Children have very different bodies that are constantly changing, and

13:58

we cannot merely copy paste adult healthcare responses to that of a child.

14:03

Moreover, I think the ethics of, you know, trialing and testing these

14:06

healthcare responses on a vulnerable group like children is at the

14:10

forefront of the concerns that are faced by regulators and governments.

14:14

Of course, at the heart of all of this is data and the Turing has been working

14:19

very closely with children to better understand how to safely use that data.

14:24

From the, what can children teach us about AI session?

14:27

We grabbed Turing fellow, Veri Aitken, and Steph Wright from the Scottish AI Alliance

14:33

to talk about their project, which puts children's voices at center stage.

14:37

Steph: Well, in Scotland's AI strategy, we had a commitment to adopt, uh, the

14:42

UN's policy guidance on AI and children.

14:45

And we wanted to explore how we can engage with children to get their

14:50

input into our shared AI futures.

14:53

Uh, it just so happened at that time, Vari and her team at Alan Turing

14:56

Institute were also interested in that. And, um, I thought, what better organizations to bring together

15:02

than the Academic Excellence of ATI with the children's rights based

15:08

approach of the Children's Parliament.

15:10

Mhairi: When we think about child centered approaches to AI, it's important

15:12

that we're not just thinking about safeguarding children from the risks.

15:15

Of course, that's one really important dimension of it, but often there can be

15:20

kind of overly paternalistic approaches, which are all about identifying from an

15:24

adult's perspective what the risks are and safeguarding or protecting children

15:27

from those risks or perceived risks AI.

15:30

But if we don't actually speak to children and understand from children's

15:33

perspectives what their experiences are, what their interests are, what

15:36

their concerns are, we might miss some really important aspects of this.

15:40

Um, and it's also important that this isn't just about identifying risks

15:43

or safeguarding children from risks. It's also about finding ways that we can maximize the value and

15:48

maximize the benefits of technology and innovation for children.

15:51

When we speak to children about AI, the big themes that come out, the kind of

15:54

central areas that, that they really. Yeah. I want to focus on discussing, um, uh, quite consistently around themes

16:00

of fairness, um, and particularly how these technologies might work

16:04

differently for different children. Uh, and I think the, uh, the children that we've spoken to certainly seem

16:09

to really kind of intuitively, uh, gravitate towards the concept of fairness.

16:13

It wasn't something that we introduced. It wasn't something that we planned to have as a, as a central theme of the

16:17

engagement, but it was really what, what the children wanted to talk about.

16:20

Um, and they grasped very quickly that, that AI might have different outcomes

16:24

for different groups of children. Um, and that they were really wanting to understand more about how we could

16:29

develop these systems to make them fairer, to make sure that they, uh,

16:32

had, you know, equitable benefits for, for different groups of children.

16:35

That's part of the value of having children in these conversations,

16:38

because, well, as an adult, we might, a fairness is maybe a kind of a

16:41

concept, you know, an abstract thing. And something that we know is important, but adults often make

16:45

kind of, uh, justifications.

16:48

Oh, well, that might not be fair, but it's because of this, this, this, and this.

16:51

Whereas children will say, that's not fair. That's not okay.

16:53

You know, we need to do something about it. We need to make that fair. And actually that's sort of the value of bringing that children's

16:58

perspective into these discussions. Yeah. I love it when the plan comes together.

17:01

So obviously the collaboration kicked off. We're now approaching the end of phase two.

17:06

The first year was, um, Uh, led by the Turing Institute, uh, to

17:11

explore children's rights and AI.

17:13

The second year, which we've just come to end, is about exploring how to dig

17:18

deeper into operationalizing some of the, um, findings in phase one of the

17:25

project, especially around, you know, safety, uh, AI in education and, um, And,

17:32

and bias, which were all these concerns that children expressed, they were

17:35

particularly interested in exploring. So phase two was about, you know, partnering up with actual organizations

17:41

with actual projects or policies they were developing that the

17:45

children can, you know, meaningfully,

17:49

Jonah: That's really interesting. Maybe this will begin to fix the trend of future generations having

17:53

to fix past generations mistakes.

17:56

Talking of futures, I was at AIUK last year and heard a lot of

18:00

predictions for the year ahead. And a lot of them were about how generative AI like ChatGPT was going to

18:05

go stratospheric and how we will start to encounter more misuse of those tools.

18:13

This year I caught up with Micheal Wooldridge who chaired a session all about large language models and asked for his thoughts. Mike: So I think there's two things I think are really interesting to

18:18

keep an eye on in terms of risks. So the first is about misinformation and disinformation, particularly in elections.

18:25

And as I'm talking now, within the next year, we're going to

18:27

have more than a billion people worldwide going into elections.

18:30

We've got elections in the UK, we've got elections in the US, elections in

18:34

India, the world's biggest democracies.

18:36

And the fear that was just beginning to be voiced a year ago was that

18:40

AI was going to be used to generate disinformation on an industrial scale.

18:45

Now the worrying thing is we are beginning to see the signs of that happening.

18:50

We're beginning to see fake news stories.

18:52

And actually, interestingly, we're beginning to see news stories where people

18:56

are claiming that it's AI generated, even though it's actually original, which

19:00

is not something that we anticipated.

19:03

So I'm Keeping, I'm looking at that nervously, um, the government's announced

19:07

initiatives to try to deal with that. Let's hope that they get it right.

19:10

And I think the Turing will be, will be front and center in those discussions

19:14

about, about how to get that right.

19:16

The other thing is the age old question of AI and employment.

19:21

And again, a year ago, we were looking and contemplating whether large

19:27

language models were going to lead to unemployment on a, on a large scale.

19:31

For example. We're just beginning to see the first signs in some sectors of the

19:36

impact of large language models. So we're beginning to get, for the first time, uh, some understanding

19:43

of how this technology is going to affect the workplace.

19:46

And I think this is going to be a crucial year.

19:49

At the end of this year, I think there's a real chance that we will have seen

19:54

some really significant signs of how AI in general, but large language models in

19:59

particular are affecting the workplace.

20:01

So I think that is something that we should really keep

20:03

an eye on over the next year. Jonah: Lily, are there any sessions that we haven't touched

20:09

on that you want to cover? I'm probably drawn to some of the, the big headlines, um, some of the more

20:15

accessible sessions, um, Uh, but I'm very aware that there's loads of stuff that's

20:19

been talked about that will be interesting to lots of different audiences.

20:23

Lilian: Yeah, absolutely. So there were about 50 sessions, all in all, at AIUK.

20:27

So there are so many I wish I could talk about, and they really do range in topic.

20:32

We had sessions on AGI and LLMs.

20:34

There were sessions on productivity that.

20:36

It might not seem sexy, but that's the stuff that's going to impact my life.

20:41

That's going to make it easier for me to fill out forms, for me to

20:45

interact with the state, that's going to make my life better as a citizen.

20:49

And I think that stuff's really important and really interesting.

20:52

It's really important to have these conversations that we might not be

20:55

having in other spaces, or we are having in siloed conversations where

20:59

just experts are talking to each other, but at ARUK, Like I said, everyone's

21:04

there, policy makers, experts from across different fields, researchers,

21:08

students, professors, industry leaders.

21:11

So you're getting all sorts of people together to talk about things

21:14

that they might not have previously talked about or heard about.

21:16

Defense and security, for example. I think it's really important to platform those issues at AIUK and

21:22

have a discussion outside of the defense and security ecosystem.

21:26

Obviously, the people who are interested in it. They're welcome to come, they learn, and they can contribute to that conversation.

21:31

But people who might not be familiar with that field, it's great for them

21:35

to see that content as well, I think. Smera: No, I fully agree, Lily.

21:37

I think with the recent developments, we can see how AI and tech is being

21:42

positioned as a very important tool in a nation's defense arsenal.

21:47

Throughout history, we've seen how advances in tech have been driven

21:50

by investments in research and development by departments of defense.

21:54

For instance, you know, I remember in our episode on chip wars, We saw how defence

21:59

investment was critical to improving the microchip architecture and capacity.

22:05

Jonah: I do. Yes. That was a good one. Chip Wars.

22:07

You can check it out on Series One. So there was a session at AIUK called The Secret Session, which was a chat

22:13

between Tim Watson from the Alan Turing Institute and Stephen Mears,

22:16

who works for the Defence Science and Technology Laboratory, DSTL, the

22:20

research arm of the Ministry of Defence. We spoke to him about the impact AI is having on defense and security.

22:25

Steve: So I think, uh, defense and AI, obviously a topic that a lot of people

22:30

feel concerned about, but for me as a scientist working in the defense area,

22:36

I really see this as a transformational technology that can really support

22:41

our armed forces in the difficult role that we undertake, that they undertake.

22:46

So, um, everything from things like commander control,

22:50

where they have to make. difficult decisions, how can we use AI to help get them the best

22:55

information, to help them make the best possible decision in the

22:59

different environments that they're in? Intelligence, surveillance and reconnaissance, how can we use AI to

23:05

help make sense of massive amounts of data and help them understand

23:09

what's going on around them? And then perhaps closer to home, how might we be able to use AI to counter

23:15

disinformation and misinformation and help ensure that people can really

23:21

understand and the authenticity and provenance of the information

23:25

they're seeing on the internet. Jonah: Lily, thank you so much for joining us for this, uh,

23:32

Too Long Didn't Read special. Uh, and congratulations to you and all your colleagues.

23:37

And that also obviously includes us.

23:39

Congratulations to everybody involved in AIUK.

23:42

Lilian: Thank you so much, Jonah. It's been great to be here and I'm so glad we can re watch it on YouTube as well.

23:47

Yes, Jonah: all the sessions, cut downs, lots of exciting content will be on our

23:50

YouTube, but we'll probably talk about that when it actually appears there.

23:54

Speaker 9: Thanks, Mera. Nice to see you again after all this time.

23:57

And hopefully I'll see you in a month for a brand new season.

24:02

Jonah: Yep, and uh, thank you of course to Jesse, who's in the

24:05

background furiously scribbling away notes, that is, not just drawing a

24:09

picture, um, and you for listening.

24:12

See you soon for series two.

24:15

Toodaloo!