The Alan Turing Institute

My name is Mona and I'm Associate Professor of Battery Systems in Warwick Manufacturing Group, University of Warwick.

I'm also a data science and AI fellow at Alan Turing Institution, and also associated with Faraday Institution through one of these projects called Next Generation Electrodes for Batteries, and I'm also an applied AI guest lecturer at the University of Southampton.

So a little bit about my background to see that where I'm coming from to this problem.

So my background is system modeling and control engineering.

And so we have this global academic network of identical interoperable platforms.

And so this has been developed since 2014, starting at UCL, then at Alan Turing Institute, now at Loughborough and Cambridge, but also with our collaborators across countries.

So you can see the map here.

We need to add Peru to this.

We talk a lot about sovereign AI and the fact that...

We just talked about Arm, and, you know, probably DeepMind, Alan Turing Institute.

Beyond that, though, it's pretty hard to think of other companies.

This is an amazing fund.

You're the director of respond- the Responsible AI Academy at the Arts Council in England.

You're a previous Turing Fellow at the Alan Turing Institute, and so basically everything we're interested to hear, you're right at the epicenter of it.

So I'm so excited to chat with you.

Thanks for coming on the podcast.

Yeah, it's interesting.

I think we, you know, I think in the UK we do have a really good, you know, assurance, um, governance, responsible ecosystem around these technologies.

So I think, you know, if we look at, you know, institutions like the Alan Turing Institute, if we look at the Ada Lovelace Institute, if we look at the work of Braid, um, another project I'd, I'd s- flag is, um, Public Voices of AI.

Um-

So people are revisiting this idea of how easy it is to plant data inside of a training set, which you can do simply by putting information out onto the open internet, because that's the place where all of this starts is If you're training a new model, you're sending it out into the internet to scan everything that's been put out there.

Now, Anthropix's alignment science team with the UK AI Security Institute and the Alan Turing Institute, they ended up doing this huge test.

It was the largest data poisoning investigation ever.

And the results that came out showed that you can plant a mere 250 malicious documents inside of an LLM's training data to get this result.

leading the Responsible AI Academy at the Arts Council of England, alongside her academic work as senior lecturer at Goldsmiths at the University of London.

She's also a former Turing Fellow at the Alan Turing Institute.

She talks to us about how arts organizations, museums, and creative companies can adopt AI thoughtfully, ethically, and with purpose.

Her central idea is beautifully simple.

So people are revisiting this idea of how easy it is to plant data inside of a training set, which you can do simply by putting information out onto the open internet, because that's the place where all of this starts is if you're training a new model, You're sending it out into the internet to scan everything that's been put out there.

Now, Anthropix's alignment science team with the UK AI Security Institute and the Alan Turing Institute, they ended up doing this huge test.

It was the largest data poisoning investigation ever.

And the results that came out showed that you can plant a mere 250 malicious documents inside of an LLM's training data to get this result.

But we need to also be very clear that their continued degradation and objectification of women is having also a tremendous impact.

A study from the Alan Turing Institute in the UK was looking at content on a video-based social media platform.

And they were examining popular hashtags that were being used to share Lux Maxi content.

And they found that just under half, so about 44% of the Lux Maxi videos they identified, used hashtags related to the most extreme end of the incel spectrum.

And how do all the institutes that are all interconnected,

how do they work with each other? There's the Alan Turing Institute as well.

How do you all work together?

There's a big ecosystem of organizations that are working in this area.

Alan Turing Institute and Ada Lovelace Institute were obviously particularly interesting because they're named for the two big thinkers over the last 200 years.

And of course, Alan Turing really looked at Ada Lovelace's work very hard and Ada Lovelace's put forward some provocations that he took into his work and really informed how he thought about the Turing test.

So there's a very direct connection there.

The Alan Turing Institute, in practical terms, is a completely different organization to us.

After the Future Cities Catapult mo- uh, funding ended, there was a couple of years gap where we just kept it going in-house.

And then the Turing Institute funded it until last year, basically.

And we're now in the process of hoping and thinking that, uh, we need some more funding for this, basically.

But the whole funding thing in Britain, as I'm sure you know, is a, is a real game these days.