AI Special 4#: Alex Kendall, Co-Founder and CEO of Wayve: AI Automobiles and the Future of Driving

26:59So I don't want to put out a wave. We don't put out specific dates at wave because we want to avoid that height until we build the confidence internally around seeing this. So, you know, it's going to be here next year that you see many, many folks in the industry saying. I think that's the thing because you've also got to be very careful about how you define it's here. So you have a definition of surpassing human drivers in the UK. Yeah. So just to be clear, I'm not talking about driver assistance or partial driving automation. I'm talking about fully driving automation. You know what I mean?

27:29full automation, the sleeping in the back scenario. How far away are we from that passing the amount of people on the roads? Do you think, what, five years, 10 years, 20 years? What do you think? So if you took the approaches that are on the roads today at a superhuman level of safety in some of the US cities I spoke of, if you took them and brought them to the UK, with billions of dollars in a couple of years, you could replicate them here at extraordinary costs. I think it would be challenging to create a business case for that, but you could do it. I think where they would fall down is with some of the weather patterns, the rain, the fog, the snow that we see, as well as the complexity, for example, in London with roundabouts and merging traffic.

28:14You don't have grid-like streets that have wide lanes with clear markings. It's dynamic and chaotic. So I stand by that I think a different approach is needed to bring this technology to a city like London. And we certainly think that if we can get our technology to work in London, we should be able to bring it anywhere to the world, even a place like, for example, Mumbai. So it's not going to be here tomorrow or next year is what I'm trying to say. Having said that, the rate of progress we're seeing in AI and the rate of development that our team has been able to push, we're seeing traction in building this technology and even starting commercial trials that it's not something that I think you're going to have to wait 10 years to experience.

29:01So I'm going to leave it with those upper and lower bounds, but I won't throw out an uncertain date. I sent a career in politics after this, Karan, so I don't know what you think. I think so too. But it makes sense. It's a very sensible answer, by the way. I do want to, though, go back to something else you said, Alex, which I was very struck with. In addition to what you said about the people that you collaborated with at Cambridge, you talked a lot about the potential to solve some of the world's biggest problems and the social impact you could deliver. So just whizzing that right up to the future and taking where you are right now and where you want to, building on Gareth's point, where you want to take the future.

29:45What are the kind of big problems that you'd love to be able to solve using some of the technologies that you've built? Let me start with autonomous driving. I think if you think about the opportunity autonomous driving presents, a lot of people talk about on the safety side, and I think that's a really, really key part of it, being able to address the 4 ,000 deaths a day we see on the roads from vehicle accidents. I think there's a bit of nuance there, though, because the vast majority of those deaths do come from developing countries. And actually, you see that modern automotive technology can mitigate a lot of them, but we still see a large number of road deaths and life-changing accidents, even on UK roads.

30:26So regardless, there is a definite safety problem to be addressed. And look at the airline industry as an example that in 2019 had a year of no deaths in commercial aviation. And so I think that's something that we can and should push for as a mobility sector. But I also think it's interesting to think about the second order effects of autonomy, because in addition to safety, you've got a bunch of other really positive side effects. For example, it's estimated that road accidents and congestion cost the UK and US economies 2 % of their GDP. And I think through more efficient transportation, we can really unlock that level of growth, which is substantial.

31:06There's other ones that might not be immediately obvious, like we can turn car parks into green parks. In some cities, 10 % of their real estate is used for car parking. This is space that should be able to be opened up into green parks and to better utilisation. I have a friend who works in urban planning that they ran an experiment where they removed the parking outside shopfronts and saw that through increased foot traffic, through increased engagement between the street and shops, these shops saw 20 % increased turnover without the car parks in front of them. So we can create healthier streets, or even if you think about the level of resource or cost that we invest in our road policing, you know, with autonomous vehicles that are much more adherent to road rules, we can redeploy that in other areas of justice and policing systems.

31:53So I think there's a bunch of really, really positive benefits to society that autonomous vehicles will bring. One of my favorite things that I unearthed when we were doing research on yourself and Wave is that you described Wave a couple of years ago as a contrarian startup. So I just wanted to ask you to elaborate on what you meant by that. And also just maybe give us a little bit of an overview of how you would describe the culture that you were built there. Well, if you cast your mind back to 2017, when I co-founded the business, that was the time when many big technology giants in America were making billion dollar investments in autonomous driving.

32:30And they all said that AVs will be here a year away. And it was at that time where, you know, a couple of PhD students at Cambridge decided to take them on, raise a couple of million dollars of seed funding and go say, we're going to take on the space with a different approach. So you can probably understand why at the time we were laughed at by the industry. People made comments like, you're crazy, it's impossible. And certainly it wasn't obvious that AI systems would be able to scale. I think over the last five years, what's played out is that actually, while through share capital and brute force, these classical approaches to autonomy have been able to work in limited areas.

33:08While we've seen a real challenge from a cost and from a complexity perspective to scale them beyond that. Yet on the other side, what we've been able to do at Wave is actually deploy end-to-end deep learning technology, AI technology, on public roads, in commercial trials across over 10 cities in the UK to do autonomous driving and show that this system can work at a proof of concept level. Meanwhile, we've seen AI scale like a rocket ship, being able to capture the imaginations of the world. And I think we're starting to see the narrative shift where actually, I think the AI breakthroughs that we're seeing in the recent six or 12 months are the unlock that will make autonomous driving possible.

33:49And we're right at the forefront. Wow. Well, where is the UK, by the way, in the global pack in order in terms of innovation and maybe regulation as well in terms of the whole thing? You said it was probably a little bit behind the US. Where are we talking in the global stage? So AI was invented in the UK. It came out of some work from Alan Turing. And I think that speaks to the real strength the UK has in academic research and AI, if you think about the universities and the industrial research labs we have here, generate so many of the big breakthroughs that lead to these breakaway models. I also think that there's been a growing startup and early venture capital ecosystem in the UK that I feel fortunate to be a part of and has been able to finance companies like ours that are going after these big moonshots.

34:35It's not yet at where Silicon Valley is, but it's certainly an ecosystem that can facilitate this kind of growth. I think what we're yet to see in the UK is really a plethora of businesses that have gone from proof of concept or early product into worldwide impact. There's a lot of, when companies go from startup to scale up, a lot of acquisition by US big tech. There's a lot of ones that fail to get that funding because that growth stage capital isn't quite there yet. but I think we're starting to see that to change and I'd love for Wave to be an example that can really help UK and European ecosystem get through that next stage of growth and hit that that level of escape velocity we need to to bring this level of impact to the world.

35:18So what was the biggest engineering challenge Alex in terms of on this journey you've been on I'm sure you had a few which one stands out for you as a pivotal challenge that you overcome in order to you know to get this this technology working in the way that you you wanted it to work? Yeah look we started the company in a house office, in a residential house in Cambridge where we were living and sleeping and eating together in this house where the small bedroom was a server room, the main living room was our office space, our boardroom was the middle bedroom, and we had a car in the garage that we were hacking together.

35:53We had to do everything ourselves into getting off the ground. It was a real challenge where there was no precedent for what we'd done. Once we got it to drive around the block, though, we started to get some traction and people started to get interested in our approach and actually help support us, whether it was businesses like Admiral, the insurer, who helped back our trials from the very beginning, or Microsoft, who's helped provide us with the compute that we need to train our models. I mean, the support of partners like these has been really important. But the biggest engineering challenge for me, and maybe this is surprising, is that to build an AI company, you know, 99 % of the problem is in data and compute.

36:32I said before that the algorithmic breakthroughs for AI breakthroughs typically come 10 to 20 years before the actual breakthrough itself. The underpinnings of chat GPT and these language models and transformers and contrastive learning and PPO or a form of reinforcement learning that allows you to give feedback have actually been around for many years before the breakthroughs today. Or similarly for us, the basis of what we're doing, it's very similar technologies, but it's the access to data that's the real challenge. And autonomous driving, you can't just sit there in your big tech office and get it off the internet.

37:04You actually have to build fleet partnerships, deployments. For us, we've got our data collection devices on our partner fleets like DPD, Ocado Group and Asda to be able to get training data to support our AI. And so getting this data, cleaning it, making it accessible and to be able to train, that's the real challenge. For the models we train, they're based on video primarily. and the level of compute you need to train on video is extraordinary, greater than, say, the text tokens you train large language models on. Instead of kilobytes of text that's needed for each learning cycle, you need to train on gigabytes of video.

37:41And this is where the challenge lies. So for me, the most exciting thing and the thing that gets us to scale is that engineering platform of data. Yeah, fantastic. Well, let's shift gears maybe then. I want to move the attention towards artificial intelligence at large on a macro level. One of the main concerns about AI that people generally are talking about at the moment is that it's going to replace jobs. It's going to have an enormous impact on the labor market. Have you got an opinion on this, on how this is going to play out, how this is going to evolve over the next, say, five, 10 years?

38:14Look, I think the opportunity from artificial intelligence technology is so enormous that we have an imperative, a society to go after in a responsible way. But I think it's important to understand the level of disruption that it will bring, and we should be eyes wide open about that. From a macro perspective, I think the things that are interesting to me about this disruption is that I think it will occur over a matter of perhaps 30 years in terms of seeing AI become so pervasive in so many different job sectors. And the challenge that presents is if you compare it to, say, a similar level of disruption we might have seen in the Industrial Revolution in the 19th century, that happened over 80 to 90 years, right?

38:57And the key difference there is that 30 years is within one person's career. It's within one generation. 80 to 90 years is over a few generations and allows space for society to retool, reskill. the challenge that I think AI presents is that we're going to see disruption within perhaps severe disruption within one generation and that presents a challenge during this evolution where we're going to need to support the world to be able to shift to adopt this technology to make sure that it benefits the masses and has equitable access. I think that's the real challenge. And, you know, I guess, bring that back to how that affects WAVE or us being transparent and advocates around this, I think, is important.

39:46But ultimately, working with and partnering with government to help support this transition, I think that's something that, as an ecosystem, we need to be really careful about. And what are the kinds of skills that you think people will need to be developing to make the most of the opportunities that AI will unleash? If you think about the strengths that AI brings, I mean, ultimately, it is a far more powerful and advanced tool than we've ever seen before. So if you think about, I mean, this is tools of what set humanity apart from others and other species in the animal kingdom, right? That part of what makes us human.

40:21And AI is a rocket ship version of what, say, calculators are to accountants or x-ray machines to doctors. You could list many other examples of tools. And AI is going to be a similar accelerator, right? So it's going to let us, it's going to give us much more agency as a society. It's going to let us achieve a lot more if we build it in the right way. And I think that's the vision that we need to see with this technology. What is AI good at? Well, AI is fantastic at making clear data-driven decisions. It's fantastic at indexing and searching and presenting knowledge. So I think that that's where it really shines, as well as areas where there are significant safety concerns.

40:57I think that's where you need automation. And at school, what are the kinds of things that you think children should be learning in order to make the most of it? Yeah, look, I think it's really important to be curious about the technology, to be early adopters of it, whether that's just playing with it, understanding it. Some of the demos we've spoken about, whether it's stable diffusion or chat GPT, I mean, they're fairly accessible, free to access if you have an internet connection and a browser. So being curious, playing around with it, I think is really important. but then also challenging yourself around where can you benefit from it.

41:32I think these could be extraordinary tools to improve files at school. I'd love to have these tools to help write some of my essays and elevate my thinking to higher level structuring and narrative and strategy and avoid some of the lower level tasks that it comes to actually putting pen to paper. And so I guess all I can say is be curious, be an early adopter and challenge yourself where you might be able to benefit from these tools because I think they'll provide a competitive advantage. Yeah, it's a great tagline, isn't it? Yeah, the calculator analogy, I think, is one I basically heard a lot about this subject.

42:04I think some people are very conflicted on this. I know because I've spoken to a lot. Carenza and I have brought this exact topic up with a number of guests and whether, you know, the child handing in the essay, if they have used ChatGPT to write that essay, does it matter? Should we be concerned? Should we be more like valuing the fact they've interacted with the technology to create a great outcome? So in terms of ethics, then, let's just maybe quickly cover that. What ethical considerations, in terms of the widespread adoption of AI technology, what ethical considerations do you think this is going to throw at?

42:38I think the important things for the ethics of AI revolve around the use case that you're trying to apply it for. So automation is really important in safety-credible applications or in situations where there's a very clear and data-driven decision being able to make. I think the corollary to that is that humans and human capabilities can really thrive where there's a subjective decision to be made, where it's driven on human nature, human relationships, or where we want to take and have agency for a decision that's made on top of tooling or analysis enabled by AI. And so I think we should be clear on two things there.

43:17One is the expectations and two is the boundaries of an AI system. If we're clear on the expectations, it means that we won't overly rely on it or expect to do something that it's simply not capable of. And that comes with really being able to validate the capabilities and transparently understanding what it can and can't do. And then on the boundaries, understanding, being clear on where that limit lies and making sure that that's respected. And we still have agency to action or to leverage the output of that AI system. I think it also comes back to something you said before, Alex, around the quality and the volume of the data, because it's so important.

43:51I mean, people often talk about garbage in, garbage out. You have to be so scrupulous to make sure that biases aren't creeping in and that all of the data sources are as kind of pure and ethically sourced as possible to make sure that whatever comes out the other side and however that is then used by the algorithm, it's as ethically sound as possible and is then applied in a very responsible way. I agree. And I think it's also important to consider the application of the AI itself, because the risk profile and the value profiles are very different. If you have a search engine, for example, or a accountant or an AI doctor or an AI driver, you know, that value that it's providing is very different in each of those, but also the risk.

44:32And so having regulation or ethical values that you apply broadly to all AI is, you know, perhaps as a base level, like we have in human society, you know, a base level of ethics and morals that we should expect from one another. But, you know, more specifically, when it comes to the application, you really need to consider things on an application by application basis, because the way that we think about validation and the risk and say the medical space is different from some of the others we cited. We might even see in the medical space that if an AI system can give a better insight than a medical professional, then we may even see, say, insurers mandate that in order to facilitate the insurance claim that a decision needs to be made or checked by an AI system.

45:19And so I think where these systems can actually improve the ability of society, we do have an imperative to adopt them. So Alex, I started the interview by asking you about what excites you most about the evolution of AI technology. So I just wanted to flip that and focus on what you are most fearful about in relation to AI technology innovation. I feel concerned about what it's going to take to deploy AI systems at scale in the right regulatory environment to achieve this kind of future that we've been describing. It's a really complex challenge with so many different stakeholders. And so to create the environment that facilitates innovation, that allows us to move fast, allows us to iterate and improve, while also being responsible, making sure that we can have provable levels of acceptable safety and really build trust in these deployments.

46:12When you then layer on that these things are so different in each application, like we discussed, whether it's a medical, whether it's an application, whether it's a search engine or whether it's a social media or a self-driving car. That's a really complex set of challenges for regulators. And I think that's going to be a significant hurdle for us to jump through to get this technology to the level of impact we need. Yeah, I think keeping up with the innovation, I think, is absolutely right. I think it's going to be interesting to see how that plays out. What advice do you give to people who are fearful about the evolution of AI and think we are heading towards this Terminator 2 Skynet situation?

46:53What advice do you give to the cynics, Alex? Well, it sounds like they should use ChatGPT to go write a sci-fi book.

47:03No, but more seriously, I think there's so many possible futures. If they're genuinely worried about it, then I encourage them to join the industry and actually take part in helping to make sure we don't build that future. If they're worried about it from an armchair observer's perspective, then I think it's important that we talk about these issues, that we debate them, but also understand the realities of where we are at today. And, you know, I'm optimistic that the trajectory we're on today and what we as an industry and WAVE and other actors are building is certainly building a much more effective future like the vision we've been describing in this conversation.

47:39Take part or make sure you're grounded and engage in constructive conversation. Yeah, fantastic advice. So I know we're coming towards the end now, Alex, and we've covered so much ground here that I could literally talk to you all day. I'm sure it feels the same because I totally geek out on this stuff. And I know you're a young man, okay? But I'm just thinking, looking back on your career to date, if you could have a conversation with, let's say, 21-year-old Alex, what would you tell that guy? What advice would you give that guy? oh gosh meeting 21 year old Alex would be uh would be crazy days and I also um am aware that I'm very privileged to be able to to say this but really prioritizing learning and experiences over say money or material things is is something that I think it pays absolute dividends in your life yeah so that's been really really big for me as well as you know just just to instill a sense of confidence in myself as as you know as everyone goes through that journey of finding their own identity and being proud to be contrarian and chase your dreams that's what comes to mind.

48:44I love that I'm a contrarian at heart as well I question everything to my wife's annoyance but yeah I love that brilliant. I was just wondering what he would have told the kind of 11 year old version of himself the one that was playing with video you know inventing video games. To my 11 year old self I'd really encourage myself to to chase my dreams and it comes back to that confidence point first. It's to be okay with failure, to be okay with making mistakes and to just try and learn while you have that privilege and to really save that time. I'm coming back to the same statement. It's all about time and experiences.

49:24Sure. Probably more emphasis on gaming when you were 11 though, yeah? Do more gaming, yeah? Oh, I wish I had time for that these days. Yeah. You're an AI expert, obviously. What AI tools, browser extensions, what tools have you used recently which you benefited from from a productivity standpoint? And what other productivity advice can you give the listeners? Oh, look, as I said before, I think being curious and an early adopter in this technology will be a competitive advantage as these systems just get better and better over time. I've been a big advocate for our team to adopt this internally at Wave, whether it's GitHub's co-pilot that accelerates your ability to write software, whether it's ChatGPT to help provide advice or restructure writing, or Notion's AI system to help with writing our internal documentation.

50:14These are some of the tools we've evolved. I've even been using Stable Diffusion to generate graphics for some of the presentations I give. I think that learning how to write prompts and engage these systems, it takes a bit of time and investment to actually make it a value add and not just a gimmick, but prompts into ChatGPT like you are a chief of staff at a technology company. Answer concisely. Please help me by summarizing these statements and putting text in or helping to reformat things or even asking for advice. It can be a remarkable coach. having said that going back to the awareness of its limitations you know there is a lot of hallucination and bias still present in these systems and so i'm not yet at the point where i've let them run autonomously on my communications but as a you know writing guide and aid uh i've found them to be remarkably effective yeah fantastic i mean the the act as a prompt was a game changer for me anyway in terms of gchat gb2 yes it's a brilliant prompt but can you tell Tell us a little bit about Alex outside of work, what you do to chill out.

51:17And what's next for Wave? What exciting stuff you've got coming up? Gosh, I mean, Wave is such a hobby for me. It's consuming in my life, and I love it. It's nothing else I'd rather spend time on doing than the work and the people that I feel so privileged to work with. For me, outside of work, I love adventure. I love sports. I love friends and family. I love reading and learning. I think these are some of the things that really are real core values of my life. To the second part of the question, what's next for WAVE? Well, we're excited to take this thing to scale. As I said, we've got a proof of concept system that shows that a next generation approach with end-to-end deep learning is possible to solve the self-driving problem.

51:57We want to scale this. We have all the ingredients we need from it, from the investment, the operations, the data, the compute, partners, and we just need to build this to bigger and better and ultimately get this to a point where society can trust it and we can deploy it in a safe and accessible way to power the future of urban mobility. So watch this space. I think embodied AI and robotics is about to go through that inflection point that we've seen in internet text and video. Yeah, robotics is robotic. The age of autonomy is on its way. And what an appropriate note to end on. That was a fantastic line at the end there, Alex.

52:37Thank you so much. I've really, really enjoyed this one. It's been fantastic. And congratulations, Alex, on all you've achieved and good luck for the future. Thank you, Trenza. Thank you, Derek.

52:50I mean, I think, you know, definitely a superstar of British business within the FAI space, at least. And I think we're going to hear a lot more about Wave. And also autonomous driving systems and driverless cars are just going to be everywhere soon. This is inevitable, isn't it? So I'm privileged to get Alex on and talk about some of the amazing topics and things we discussed. So he said, what scares him most about AI? I think it was a very different answer to this question. I did ask a couple of the guests this as well, and they didn't really bring this up. And I was quite surprised by this, especially coming from him.

53:26he basically said that he thinks AI has the propensity to maybe become too complicated for humans to actually sort of build on basically I suspect what he's getting at is that we may lose control a little bit possibly and I think for him who obviously understands that the technical element of artificial intelligence I think that is quite a profound statement did you have any thoughts on what he said his answer to that question? Yeah Yeah, it just struck me that he's someone who's thinking incredibly deeply about it. And because he is a data scientist himself and an engineer himself, he can sort of almost horizon scan a future gaze and think about what the potential impacts could be of the future.

54:15So I think we're in safe hands with somebody like him because he's worrying about it. It's the same, I mean, here at BT, we've got the most incredible data science lead in Zeri Webster. And she spends a lot of her time thinking about the risks and the challenges. And I think it's sensible that the engineers are thinking all of this through so that we can ensure that we safeguard the world for the safety of human beings, the environment and all of us. And I think it's the sensible thing to do, to be thinking all these things through. Yeah, absolutely. You want your best people working on your difficult problems, don't you?

54:51So was there anything else that sprung to mind or anything else to add? Well, only that you missed the big one there, Gareth. It's like nearly wave is making waves.

55:09As you quite possibly have heard me mentioning in the past, we recall the mass majority of our episodes at an amazing studio facility here in Cardiff at Tramshed. tech. Tramshed Tech is a collaborative community of entrepreneurs and scaling businesses geared towards supporting growth in tech, digital, and creative industries across an ever-increasing collection of locations and partner locations, UK-wide and internationally. It really is the perfect place for your business to start up, scale up, accelerate, or innovate. Head over to tramshedtech.co.uk or just search Tramshed Tech on your favourite social media platform.