Verity Harding: How to Build Trust in AI

0:05Hey, everyone, and welcome to Generative Now. I am Michael Mignano and I'm a partner at Lightspeed. AI has obviously exploded over the past few years. We have new startups, new products, new models, you name it. The world of AI feels very much like the wild, wild west right now, including how it's aligned with our goals. This week's podcast guest has a thoughtful perspective on what needs to happen to make AI safe. Verity Harding is the author of the newly published book, AI Needs You. It's a nuanced perspective on building policy around AI, encouraging public trust and looking at past points in history, like the space race, the development of IDF and the internet as potential inspiration for AI alignment and safety.

0:48Time Magazine named Verity one of the 100 most influential people in AI. She's the director of the AI and geopolitics project at the Bennett Institute of Public Policy at Cambridge. And she also runs the tech consultancy named Formation Advisory. Previously, Verity worked in UK politics before transitioning to working in tech and directing policy for Google's DeepMind. So take a listen to this conversation I had with Verity Harding about her book, AI Needs You, how we can change AI's future and save our own. Hey, Verity. So good to see you. Thanks so much for doing this. Oh, thank you for having me.

1:24It's great to be here. So there's so much I want to talk about with you. Obviously, I want to talk about your book, AI Needs You. Congratulations. But before we get into that, I thought it'd be really helpful to go through your background. You have an incredibly impressive background. You've been working in AI long before many of us probably ever really thought about it or started building in it. So, you know, why don't you just take us back to the beginning and give us the story of Verity Harding? Sure. Yeah. Well, you're right. I have been in AI for a really long time. So it's been fun to see people kind of discovering it and getting excited about it because that's really what drew me to working on it in the first place.

2:03My background, actually, I started off in politics. I began my career working as a political advisor to then Deputy Prime Minister Nick Clegg, who's obviously now president at Meta. And I worked on a bunch of different issues, everything from equal marriage and same-sex marriage to issues of national security. And actually, national security really is the link that ended up making me want to go and work more deeply in the tech community. And this is just before Snowden. And it was really clear to me there was this big gulf, really, between what kind of technologists in the tech community knew versus what the political classes and the sort of democratically elected representatives who were supposed to be overseeing and scrutinizing and regulating that technology understood.

2:55So I wanted to be kind of, I guess, a mediator or an interpreter between those two worlds, explaining politics to tech people and tech to politics people. So I was at Alphabet close to a decade in total, but the vast bulk of that was with DeepMind, who I'm sure all of your listeners will know now, but was a little known British AI lab at the time. And I was at Google when Google acquired them and got to know Demis Hassabis, the founder and CEO, through that process. And was very inspired by him personally, of course, and also the work that he was doing and his team was doing. And it became really clear to me that all of the issues that I was interested about and wanted to work on when it came to technology, so technology for, you know, improving the world and making it better, also technology for sort of supporting and defending civil rights and human rights rather than undermining them, was gonna be made immeasurably better or worse by AI.

4:03And it was clear, you know, Dennis and his teams wanted to make it better, and I was keen to be part of that. So I went to DeepMind when it was, I think, under... just under 200 employees. And I set up all of the company's sort of policy work as global head of policy and also co-founded the ethics and impact and society teams with social scientist research, as well as our civil society and academic outreach. And while I was there, also co-founded the partnership on AI. So it's been a long journey in the subject. Yeah, such a fascinating background. And especially in hindsight, like how incredible that you were able to be at DeepMind in this really transformative and sort of pivotal moment in the history of AI.

4:51Looking back, did you all at DeepMind know at the time that the work coming out of this team would be so important to not only Google, but obviously the broader AI landscape? Yes, definitely. And I mean, I certainly did. We had too much work to do, so we weren't really sat around talking about, you know, how important are we in the space? But for me, being a sort of externally facing person, it was really clear to me how important it was. We're in this big sort of AI hype cycle at the moment, right? Driven by ChatGPT. But again, because I was kind of dealing with people on the outside, I very acutely remember the big hype cycle that happened in AI around sort of 2016, 17.

5:37and it felt like being at the center of something very dramatic. I've said to sort of friends that are open AI now, like, you know, I know how it feels, you know, when you're in the center of it, because AlphaGo was such a huge moment. And I think it put AI, certainly the kind of latest iteration of AI, which DeepMind had pioneered or popularized or both. It really put that on the map and it certainly put on the political map. But this, I think, was the first time that people really started to engage with it seriously. And we also started to see big global powers engage with it seriously. So you had the outgoing Obama administration were doing some work on AI, which I was involved with.

6:26Then you had Xi Jinping in China in 2017 do his big declaration. of China as an AI superpower by 2030. We had Canada and France with back-to-back G7 presidencies putting AI on the map. So yes, it definitely felt at the time like a big shift and very important and that it was going to have a big ripple effect. Do you think at the time, like, did you all know kind of what the impact of AI would be on the world in the way that I think kind of everyone or at least everyone in tech understands now in sort of this post chat gpt world like did you all know it would get to that point and did you know it would get to that point so much sooner than i think lots of other people realized it's why i think demis was so supportive and keen that we set up what was at the time called deep mind ethics and society and i i think it's probably the name is probably changed now but my team looking at really deeply engaging with the societal impacts of the work, because we knew, and Emmys and others knew, that it would have a huge societal impact, that it was going to potentially affect everything, you know, if we were successful.

7:43And so I think there was this kind of deep sense of responsibility towards that. I was very proud, you know, to build a team that was, you know, giving out grants to help grow the ecosystem of groups, you know, looking and scrutinizing and understanding this technology because we were really clear that it wasn't going to be something that we just did all by our, you know, all on our own, you know, that it shouldn't be and that it couldn't be. But definitely from the early stages, you know, from the earliest days of my working there, it was all of this stuff that has taken everybody by surprise now.

8:21is none of it has taken me by surprise because we kind of knew what was coming. Perfect time to transition into talking about the book. Congratulations on the book, AI Needs You. Thank you. Incredible accomplishment. Also a really, really great title. Tell us a bit about that title and how did you land on that? I think I had written the book before we came up with the title. And I think we both felt, my editor and I, that it really symbolized what I was trying to do with it, which was to some extent demystify AI. It's not a manual and it's not a deep technical breakdown because people had already written those.

9:00But it is, I think, an attempt to say to people that AI is for you. It's for everyone. And everybody has just as much of a right to have an opinion and an instinct about what they want to see from AI as the next person. You don't actually have to be a deep technical expert or in the community to have your voice be important and your opinion be valid. So while the bulk of the book is really a political history of science, saying, you know, what have we done when it comes to transformative technology before and what can that teach us for AI? I think the overarching message for people is that, you know, AI is a technology, and throughout history, we have seen that science and technology are imbued with the values of that sort of culture and society at the time.

9:58So, you know, you see that the prevailing culture and the prevailing community and the prevailing politics shapes science and technology as much as science and technology shape society and culture. It's more typical to think about the second of those, right? to think about how the Industrial Revolution changed our society and changed our working patterns. And that's true. But also, the technology that gets built, that gets funded, that gets prioritized, that gets regulated, that gets widely disseminated, that is also hugely influenced by what's going on in society at the time. And I think you can take a really positive message away from that, which is, you know, human input and control, our ability to shape, to some extent, the forces that technology brings to the forefront.

10:49And if we can see that we do have this role, that we can shape the technology, then what kind of values do we want to shape it with? And what kind of society do we want to see? What sort of society do we want to build? I think it gives you agency. And a concern I've had in the debate from the past couple of years is, I think we've taken a step backwards in terms of frightening people when it came to AI. And if people are scared, then I think they're disengaged. And I think if they're disengaged, then we both get a worse outcome because there's less diverse viewpoints involved in shaping things.

11:25But also, there's less trust in the technology. So it's kind of bad for builders and innovators to have a suspicious society that's rejecting the technology as much as it is for society at large in terms of making sure that that technology is appropriately sort of shaped and guided. This point about people being scared, what it makes me think of is the fact that the world in general, maybe partially due to the internet and social media, is very, very polarized right now. It seems like every issue, you know, you've got this tribalism of people on one side and people on another side, and maybe that's what leads to the fear.

12:02In the book, you talk about these sort of historical points of reference of innovation, which we'll get into. And one of the things I was thinking about as I was reading it was just like, I wonder if that same level of fear and tribalism existed during these other historical points of innovation because we didn't have the internet, we didn't have social media. How do you think about that? Do you feel like as a society, we're gonna be able to have the same types of pragmatic debates that maybe we had during the space race or IVF that we're having today on Twitter? Yeah, look, I think you're very perceptive and right to say that we're a very divided society at the moment and that reflects not only in our conversations about AI, by the way, but in AI itself.

12:46Sure. The thing that's really clear to me after all of this historical research is that, as I said, given the fact that the political culture of the time affects the technology so deeply that there's no wonder really that the AI community feels very divided at the moment. Of course it is. It's just reflecting back what is happening in society at large. Given that AI is just us, right? It's just us. It's us building it. It's us using it. It's us guiding it. Of course, it's reflecting back to us. It's holding a mirror up and we're seeing ourselves, which is a very divided, polarized society, a very mistrustful society, particularly in our institutions and, you know, a very unequal society.

13:36So I think we have to be really mindful of that and really aware of that if we want to try and wrestle it back. Right. You know, we need to listen to each other and talk to each other and communicate and trust each other and all those kinds of things to ensure that this, like, goes in the right way, ultimately. Um, in terms of, you know, whether societies were as polarized back then in these other historical examples I use? The answer is yes, definitely. I mean, you're right, we didn't have social media. But even without that, very, very divided. I mean, the space race, which is the first example, you know, comes out of society kind of in the US, riven with division because of the, you know, the Red Scare, the House Committee for Un-American Activities.

14:27You know, this Cold War-induced fight against communism when people were, you know, shopping in their colleagues and neighbors and, you know, accusing people of being communists. And it was, you know, very, very fraught. In the second example, which is IVF and human embryology research in the UK particularly, this all happens under the Thatcher government. And the Thatcher government was very controversial. And there were huge divisions in UK society at the time based around economic decisions that the Thatcher government had made. So I think we definitely have been in this position before. and it doesn't make things easy by any stretch, but it doesn't mean that sort of thoughtful, committed people can't reach out across those divides and still try and make progress.

15:18I thought the historical points of innovation that you cite in the book, so the space race, IVF, you also talk a lot about the internet as one of the examples, were really, really interesting. And I think looking back, obviously these are hugely inspiring moments in history, right? In the history of innovation and humanity in general. And the space race, as you highlight, the motivations early on were more about war than peace. I'm curious, the historical moment that I feel like a lot of people do talk about now, you even see this in pop culture through films like Oppenheimer, is the development of the atomic bomb.

15:54I think one of the reasons that people often make the comparison to the development of the atomic bomb is that they believe that there's sort of this global AI arms race right now. if you really think through the game theory of it all, at least somebody is always going to be accelerating here and thus everyone must be accelerating. Why is that right or not right in your mind? I think you're right about that. There's this sense that AI is somehow like the atomic bomb because the only narrative that we can sort of impose it onto is one of a Cold War arms race, essentially, or a wartime arms race of some kind.

16:34I think it's an unhealthy analogy to choose. I think it's not very relevant to AI today. I mean, AI can be used in weaponry, but it's not a weapon. I think it's an example that makes people switch off because it just seems so sort of frightening and outlandish. I think it distracts from the very issues that we have to think about when it comes to AI and control and regulation by making it seem that it's this issue that can only be controlled in some sort of, you know, international arms control type forum. So I think actually the idea that there's this AI arms race has gained a lot of traction over the past few years.

17:19And I have this new project at Cambridge called the AI and Geopolitics Project, which is trying to provide alternative frameworks for AI on a geopolitical level. So, you know, as you say, if we talk about AI in a geopolitical context at all at the moment, it's one of competition and nationalism and antagonism and, you know, arms. And of course, when people talk about AI arms race, they're not always talking about arms themselves. you know, weapons, they're talking about just a sort of competition. But I think if you combine that with this sense that the atomic bomb is the only relevant example, or even is a relevant example at all, I think that's obviously where people end up going.

18:06But in actual fact, you know, AI has hugely exciting potential to sort of uplift and inspire and change the world in really meaningful, exciting ways. And, you know, it gives us a great deal of hope, You know, at DeepMind, we were always thinking about, you know, AI for science, AI for the scientific endeavor. How can it support scientists in the innovation that they're doing? How can we use AI to do things like AlphaFold, which is this great program from DeepMind that has helped unlock a hugely complex and time-consuming part of their process for biologists? What can we do for the climate crisis with AI?

18:47There was a paper written years ago by a bunch of really esteemed people in the community about how AI could help the climate crisis. But I'm yet to see anybody, really, and certainly no national governments, talking about AI in the geopolitical context for climate. You know, something that ends up being a huge effort, a moonshot, if you will, a cooperation and collaboration between nations and companies. So I think it's a really unhelpful framing this arms race narrative, which, if you accept it as true, then a bunch of other things stem from that, which I think can be quite unhelpful. One of them being, you can't introduce any regulation because you might slow down companies building.

19:31Now, look, I'm actually not, even though I work in policy and politics, I don't want to see technology regulated too quickly or over-regulated by any stretch. But again, this is why I think the lessons from history are so important. What I really took away from the examples in the book, and I write about this a lot around the second example of human embryology research, that actually some limits, some guardrails, some very carefully crafted regulation can actually enable companies and innovators to flourish and to thrive within kind of clear boundaries, essentially. So I'm hoping the work that me and the team will do at Cambridge will start to bring forth other narratives that are, and ideas and approaches that are grounded not so much in competition and militarism, but more in sort of cooperation and essentially a more positive vision for what the future of the world looks like with AI rather than one that's so negative as it is currently.

20:35I think a lot of builders, entrepreneurs, VCs, frankly, like, I think my natural inclination is to be like this as well as a former entrepreneur and now VC, is to not slow down and to keep building and, you know, sort of an eye towards technological progress. I guess, what advice would you have for people like me or builders who feel that we shouldn't be decelerating from both a technology and geopolitical standpoints? Yeah, I don't want to see deceleration either. The reason I got into technology was because I've always believed in science and tech as sort of this incredible way that we progress as a society as we move forward.

21:12I mean, I wouldn't want to live 100 years ago. One of the reasons I wouldn't have the internet, and I can't go five minutes without the internet. But all these incredible advances that make our lives today so much better than they were in the past, huge numbers of them come from science and technology. and I think we need it to help us continue to move forward and make progress. The way to do it, I think, is just with intentionality and with purpose. This is really what comes through in the final example around the early internet, that it sometimes can be to the detriment of progress if you try and not have other viewpoints involved, you try to keep, e.g., the government, as far away as possible.

21:55Of course, no one is building in a vacuum. That's not possible. You're building in a context. You're building in a society. And being very aware of that and very alive to that, I think, makes the building ultimately better for a bunch of reasons. One, trust. You know, it's really important that people trust the people building the technology and the technology itself. You know, you probably saw this, but I wasn't there, but I saw that at South by Southwest recently, through an advert for Gen AI, and it was booed by the crowd. And this is a South by Southwest crowd. It's people like me who care about technology and want to see it be the best of itself.

22:41I think you only do that by being really alive to the context in which you're building. So I think that's one really important reason. I think the other is that it's much easier to build and to grow if you're doing that on strong foundations, stable foundations. And actually, this second example in the book around IVF and human embryology research showed that some limited, targeted, thoughtful regulation really enabled a flourishing, multibillion-pound life sciences sector to grow in the UK. That technology went from being hugely controversial to being sort of completely depoliticized and non-controversial.

23:25Whereas you look at the US where the same issues became, you know, were not dealt with in the same way. Legislation was not brought in. There wasn't that same sort of deliberative discussion. And that type of life sciences research remains, you know, controversial in the US to this day. George W. Bush banned stem cell research, which I don't think is good for science. Obama then unbans it. but then you're making something which should be an issue about science for people's health and well-being becomes partisan politics, which I don't think is good. So the other reason I think for builders to care about the cultural context is because I think it also ultimately creates a more stable environment for them in which to build.

24:11So when it comes to AI governance, who are the specific constituents that you feel need to have a seat at the table in the conversation? I think the technologists, the people building themselves absolutely need a seat at the table. I was running the policy team at DeepMind and at Google. I want the scientists, the builders there, you know, in those conversations for sure. It's just that I also want a very diverse set of views to be at the table because I think we'll get to a better solution for all of us if that's the case. So if we're looking at the broad umbrella term of AI, right, there's a bunch of things that come underneath that.

24:47you know, Gen AI is one, but let's also look at the way automated processes have affected people's lives. I mean, no one wants to talk about this as AI anymore, and I'm guilty of this too. Sort of something becomes integrated and it's like, well, that's just, you know, algorithmic decision making or, you know, everything used to be called big data. And now all of it is seen as just kind of normal. And now when we talk about AI, we're only talking about the future stuff. But if we talk about, you know, automated decision making processes that are happening, changes in society that are happening through things like live facial recognition.

25:21You know, these are discussions that need to be representative. And actually, the people who are often best placed to weigh in on those discussions are the people already being affected. So I often say to people, talk to the people affected by this. You know, talk to somebody who's maybe had their benefits or their health insurance stopped because they got trapped in algorithmic decision making and they couldn't speak to somebody. It talked to a delivery driver who is monitored with cameras and sensors the entire time that they're working so that the types of freedoms that those in more privileged, higher-status jobs take for granted are basically impossible.

26:07Your breaks are timed. You know, the fact that you put your foot too hard on the pedal to stop or accelerate is monitored. Monitoring your facial expressions all the time. Every single delivery that you make has to happen within a window. And these types of things, I think, can feel very removed. Sometimes, I felt, from the tech community, I felt like we were, you know, often, not always, but often a pretty privileged bunch insulated from some of these harms or some of these trends. And I think that leads you into a false sense of security, because it might be those people today, but it might be more people in future.

26:54And if that's not the type of society that you want to see, then learning about that and educating yourself with those stories, understanding that more deeply, and indeed consulting with those people. Participatory design processes can actually really help improve the outcome and lead to sort of avoiding unintended consequences. So at DeepMind, we sort of brought in social scientists, as well as kind of traditional AI research scientists. And what we found is the collaborations and the people working together really were fantastic in terms of getting us to a better outcome at the end. The more voices, the better.

27:32Sounds like a kind of glib, easy statement to say. But to some extent, the greater diversity of viewpoints, the better, both for getting a better product or a better outcome or ensuring that you can align more clearly with your purpose and with your intention when you're building something. But also, I think, in terms of avoiding potentially adverse outcomes that are ultimately damaging to your brand or your ability to continue to build in the way that you want to. It feels like there are many, many, many topics that will need to be discussed, many design conversations, as you put it, to be had from obvious things like, you know, facial recognition, which you mentioned, to the fair use conversation around training data and all the way to like things that may happen far into the future, like job displacement and universal basic income and things like that.

28:26How do we even go about prioritizing this list of conversations to be had? Where does it start? Well, the good news is we have to fix it all if I go. I think sometimes it feels really daunting if you think about AI as all of those conversations that you mentioned, and we have to fix all of them. And in actual fact, it will be different groups of people at different times who focus on different parts of that conversation. So in the second chapter in the book that focuses on IVF and human embryology research, this was a new technology which, after initial excitement, people became very nervous about.

29:07I mean, it's almost exactly like we're seeing with AI today. You know, the first baby was born using IVF techniques in 1978 in the UK. Everybody was super excited to start with. And then in the media and in broader discussions in politics and societal awareness, there was suddenly this concern. You know, what does this mean to be human? What will this mean in the future? What does this mean we might be capable of? And people coming up with these kind of Frankenstein-esque type nightmares. I mean, it feels very similar to the AI conversation we're having now, this kind of deep philosophical conversation as much as a practical one.

29:45What the government said was, look, this is new. We don't know. We can't just regulate straight away. So we're going to kind of set up a kind of commission to look at it, a multi-stakeholder commission, which was led actually by a philosopher. And social workers, legal scholars, representatives of religious community, biologists, scientists, all together on this commission, consulted really widely up and down the country about things that were important, concerns that people had, potential exciting things that might come from this, the huge advances in tackling hereditary genetic diseases, for example, and produced what, after a few years of debate and discussion, became basically the gold standard globally when it comes to regulation of this space.