#134: Daniel Hulme, Global Chief AI Officer at WPP, on Machine Consciousness and the Future of AI Agents

23:21And then the third pillar, so the first one is enable people to innovate at the edge. Second is make sure that you're using AI talent to differentiate your business. And the third is work with the right partners to bring in solutions so you don't have to build yourself. What I'm seeing is employees trying to build marketing solutions, HR solutions, CRM solutions. Now, actually, arguably, some of those solutions can be built internally because they're just a shiny veneer that sits on top of data. But the marketing stack is not a shiny interface to some data. It requires machine learning, optimization, deep expertise in LLMs.

23:55And that's what I'm responsible for, is making sure that that intelligence layer is differentiated. So those are the three things I think an organization needs to survive and thrive in the world of AI.

24:06Daniel Hulme:Yeah, that's fantastic advice. So in terms of controls and governance, Daniel, do you get involved with helping organizations put the right controls in place? Because it's a minefield, isn't it, really, for most companies, within employees, putting data into these models and things. I've now set the AI strategy. we're executing that strategy from a marketing supply chain perspective. Now my big focus is agentifying WPP. And AI governance and safety has been my passion for almost two decades. Because what's happening or going to happen very soon is that companies are going to unleash an army of essentially intoxicated graduates in the form of agents across their organization.

24:50And forgive the technical term, but it's going to be a shit show because most of those AI agents are not going to be capable of doing their job. They're going to fail. People are going to upload the wrong data. It's going to be a real problem. And so we've actually arguably pioneered that governance process. And just so you're aware that we spun out a company from WPP that I lead called Consium. The North Star for Consium is to try to solve machine consciousness. The hypothesis is that a conscious superintelligence will be safer for humanity than a zombie superintelligence. But one of the products that we've launched three months ago as part of this moonshot is agent testing.

25:29So if anybody has ever built software, they'll know that at least 80 % of the energy of deploying software is testing. And what's going to happen is people are going to deploy software in the form of agents. They're not going to test them and it's going to fail. So this tool, which we call Verify AX, we believe is the world's most complete agent testing solution. and the reason why WPB has invested it, the reason why I'm putting my time into it is because we think there's going to be a huge opportunity in that space in the next few years.

25:58Daniel Hulme:A hundred percent. I mean, yeah, I completely agree. You were talking about this a couple of years ago and I thought that that opportunity and the periphery opportunities to what you're talking about was going to be here by now. It's taking a bit longer. Do you know what I mean? I think, and this is what I don't understand. I think people are, some people struggle to use these large language models, Agenda AI, to achieve goals. Do you know what I mean? I think it's the interaction between human and technology that's quite challenging for people. Do you know what I mean? Are you seeing that? I'm asking you about speed of adoption, basically, Daniel.

26:34Daniel Hulme:Are you seeing an increased acceleration in adoption from the businesses you're involved with, or is it quite stagnant at the moment? I don't think it's stagnant, but I think... So Daniel Ariely, a professor from Duke, or yeah, I can't remember, he's written books called Predictably Rational, he said that 10 years ago, he said big data is like teenage sex. Everybody thinks other people are doing it, people say they're doing it, but the reality is nobody's doing it. And we're still, I think, in the teenage sex phase of agents. Actually, I would argue that some people are doing it, they're just doing it badly.

27:11So I don't think it's... The idea that you're going to get all of your data together and somehow extract insights that are going to drive value in your business was nonsense. It was never going to work. It's a big data. But agents are absolutely transformational. What people are recognizing is deploying software at scale that works safely and responsibly is a lot harder than what they think. So I don't think it's stagnated. People just are starting to sober up in terms of the effort required to deploy agents and AI in the right way.

27:42Daniel Hulme:What can go wrong then for businesses, Daniel? What is worst case scenarios if you get this stuff wrong? I'm trying to edge into security and things like that. Worst case scenario is it causes an existential threat to your business. I can go and create an agent now. I can go and tell Claude to go and delete everybody in my address book. And he'll do it because he doesn't know that that's a stupid thing to do. We need to make sure that we're putting the right guardrails in place to make sure they don't do stupid things. Yeah, sorry. I probably asked you. It was probably a bit of a bad pass. What I meant was in terms of what is likely, do you know what I mean?

28:14Daniel Hulme:What mistakes are companies likely to make? Obviously, they're not going to do something stupid like that, but they could be drawn into making bad governance, not having governance in places of an obvious one. Well, I think the biggest pitfall is people placing them on bets. They're deploying resources, building solutions that can be bought at a fraction of the cost from a third party instead of identifying the problems across their supply chain that are going to actually differentiate their business. I often say to an organization, if you're going after quick wins and low-hanging fruit, you're going to make a massive mistake because quick wins and low-hanging fruit can be supplied by a third party.

28:51You need to focus on the problems that will differentiate your business and those are not quick and they're not easy.

28:55Daniel Hulme:So what use cases of agentic AI are you most excited about at the moment then in the business context? Operating workflows, I'm actually quite excited to see how people are using their personal agents to go and interact with systems and CRM is definitely an area that is ripe for being being agentified but again the reality is is that there's there's most people are just building wrappers around chatbots they're not actually getting agents to do agency to click on things and make decisions for them that will come when it's starting to come but I think beyond that once they do start to have agency do things they're going to realize they're going to make mistakes so they need to be tested but the real opportunity for me around agents, and this might be still a few years away, is a concept called multi-agent reasoning, which is having agents collaborate with each other to be able to learn new things about the world that they didn't previously know.

29:45So if I've got one agent over here that knows that Socrates is a man and another agent over here that knows that all men are mortal, by getting them to talk to each other, they can infer that Socrates is mortal. They can infer new knowledge that's never existed, that's not even on the internet. And so for me, the real power of agents is getting them to work together and with human beings to come up with new knowledge, new insights that are not available to any one model.

30:08Daniel Hulme:Let's bring this over to skills then. We talked about business context. What kind of skills, if you were to advise someone early on in their career or looking to have a career change, shall we say, what kind of skills are going to become high demand? But that may be a not high demand at the moment, do you think? What's going to emerge over the next couple of years, in your opinion, Daniel? I mean, it's so hard. I've been asked this question for many many many years and i used to say one word which was orchestration and what i mean by orchestration is the ability to to orchestrate people and inspire people and get them to collaborate around a common problem we also need to be able to figure out how to orchestrate complex ai agents and systems to solve a problem so actually having an engineering mindset having an architectural mindset i think is important but the reality is it's very hard to predict beyond the next three or four years.

30:59And so the best thing that people can do is to learn how to be as adaptive as possible. If you go back to my definition of intelligence, goal-directed adaptive behavior, the more adaptive you are, the more intelligent you are, which means learning how to be better at critical thinking, resilience, curiosity, creativity, building out those dimensions of yourself will allow people to adapt more rapidly to this changing world. But I think everybody should ground themselves in sort of an engineering mindset. I think we sort of learn humanities and learn about ourselves.

31:32Daniel Hulme:Yeah, there's a lot of change ahead, isn't there? Yeah. What about prompting? Do you think that that is a bit of an overrated skill in a potentially energetic AI world? You know what I mean? It's more of a generative AI skill, isn't it? What do you think of that? I think there's two types of prompting. There's one is sort of how you construct a prompt. And I think that's becoming less and less important. And AI can take what you've asked and it can structure it in a more effective way. But what we found, for example, with some of our creatives in WPP is those people that have interesting, diverse backgrounds, those people that have learned anthropology, art history, photography, so that lighting, aperture, camera angles, they're able to ask much better questions of the AI than I will because they're able to refer to some tribal painting in some corner of history that I will have never heard of.

32:21So it's not better prompting in the sense of how you construct a sentence. It's just that they're able to refer to things that AI knows about that other people can't. And that requires actually having a breadth of knowledge. It becomes being a bit of a polymath and also, I guess, understanding the humanities. So yes, being able to have a breadth of knowledge to be able to ask better questions will typically get better answers.

32:47Daniel Hulme:So let's get a bit philosophical then, okay? So I know Elon Musk did an interview recently with The Economist, I think it was. And he was talking about how we're going to get to a point where there's so much sort of abundance created by AI technology, economic abundance that he doesn't think people are going to have to work. And we're just going to be in a world that's very different to what it's like today. So the question I've got for you is, what do you think the biggest societal impact of AI technology is going to be in the next five to 10 years? So first of all, I would be very cautious about what you believe Elon Musk to say a few years ago.

33:25And I have this thing that probably will get me into trouble, which is you only have to be incrementally smarter than most people to accumulate a disproportionate amount of wealth. You don't have to be just disproportionately. And the problem is that people think that people like Elon Musk are disproportionately smarter than other people, and they're not. And then so we must caution ourselves to believe that because you've got lots of money, that somehow you're an expert at stuff.

33:51Daniel Hulme:I think it's his appetite for risk that sets him apart from most people, Danny, as opposed to intelligence, isn't it? Again, I actually have a controversial view of that. I don't think entrepreneurs are risk takers. Either you're not asking the right questions, in which case you're going into things blindly and you're not aware of the risks, or you have a clear idea about how you're going to execute all that kind of stuff, you have the right funding, and therefore you're still not taking on any risk. Yes, of course you do. I'll take on a little bit, but I don't think that people take on risk blindly.

34:22I see what you mean.

34:23Daniel Hulme:You're calculating risk-reward. Naive, or you're calculated, exactly. Anyway, that's a different conversation. But in terms of the future, I actually argue that there are seven singularities. So you might have heard the word singularity comes from physics. It's a point in time that we can't see beyond. and I've spoken a lot about this and I've written a lot about this but first of all, beyond five years nobody knows what they're talking about so if anybody can tell you what the thing the world will look like beyond five years they're talking nonsense but for me there are some interesting things that we're facing as a species everything from curing death which there are scientists that believe there are people alive today that don't have to die to building a super intelligence which might happen in the next three or four years to people building ascribing consciousness to machines that are not conscious or even the risk of building machines that are conscious which then changes our position as a species all the way through to creating a post-truth world or AI is getting very, very good at persuading people.

35:18So will it create surveillance capitalism? But what you've alluded to is what is called the economic singularity, which was coined by a very good friend of mine called Callum Chase. And by the way, it's people like that I think you should be listening to and not Elon Musk. So Callum Chase has been writing. Callum Chase, he's a good friend of mine. He coined the term economic singularity, which is the point in time where we automate the majority of human labor. And let me give you the...

35:41Daniel Hulme:I thought it was Mark Andreessen who did that, but I... No, this is the problem. The people with the loudest voices tend to get attributed to these things. And the argument that Elon Musk is trying to make is, well, first of all, the capitalistic model to reduce costs, increase profits, means that when we can free up whole jobs, we probably will stop hiring or move people. And if that happens very quickly, then our economies won't be able to rebalance fast enough and it could lead to social unrest. And I know lots of organizations whose goal over the next three years is to half their workforce.

36:16And I think that governments have a duty of care to try to mitigate the risk of mass technological unemployment, three-day working week, whatever, UBI. The counter-argument of this, and this is what Elon Musk is trying to say, is that by removing friction, which usually means human labor, by using AI to remove friction from the creation and dissemination of goods like food, healthcare, energy transport, we can bring the cost of those goods down so much that they become free, right? So what would the world look like if we've automated the majority of things that humans need to thrive and survive?

36:51Now, I ask people, what would you do if you were economically free? And most people say they'll improve their golf handicap and they'll travel and they'll indulge in their hobbies, they'll spend time with their friends and family. If you keep pushing people hard enough, if food, healthcare, energy, transport, education if it was all free then people say I'll do all of those things that make me happy but I'll also want to do something that makes the world a bit better and I actually think that we are all learning born into economic constraints preventing us from living our true humanity which is not just living for ourselves but living for other people and by the way if you want a gold-plated Lamborghini you probably also need a therapist most people I know they've become economically free they don't care about having more stuff.

37:36They care about meaningful connections. Purpose, exactly. That ceiling is not as high as what people think. They've done studies about this. Getting to that point is not as high as what people think. What matters then are all of these other things we've talked about. So I don't care about these people making billions of dollars. In fact, most of them are not motivated to make billion dollars. It just so happens that companies are a nice vehicle to create value. And that value is often translated as wealth. But what I care about are the people that are living on a dollar a day. And actually, there are leading indicators to suggest that those people are getting more access to economic freedom, food, health care, education.

38:12And so the promise of AI, if we get the timing right, is to elevate everybody to the point where they don't have to care about surviving.

38:20Daniel Hulme:Yeah, there's an interesting theory, which I'm sure you've come across, post-labor economics. I think David Shapiro talks a lot about it, and it's the sort of shift from wages to capital. so income moves away from hourly pay and salary to owning a share of the automated tool, the stock, a public wealth fund, or whatever it is. But I don't know if you've come across David Shapiro. Yeah, yeah. No, absolutely. Yeah, yeah. I think that what's interesting is that AI can now give everybody the ability to create innovations extremely quickly. Now, of course, they're digital innovations, but at some point it will allow you to create algorithms that make real-world goods more efficient and more effective.

38:58So it's a very interesting time to be alive.

39:01Daniel Hulme:It is indeed. I've got to get this one in, Daniel, because I was looking forward to asking you this. Is AI conscious, in your opinion? Okay, so in my opinion, the answer is no. But what is your definition of conscious, though? Okay, so there's a... Consum is Latin for consciousness. So we've managed to attract some of the world's leading thinkers in this field. One person, one of the academics will say, we'll never create machines that are conscious. Another academic will say, we're going to have machines that are conscious in the next few years. And the reality is that people are going to be arguing about consciousness for like 2 ,000 years.

39:39And let me explain why I think they're getting things wrong. And that's going to sound massively arrogant, but I think I've got now my head around consciousness. So, okay. So by the way, I've got a book coming out in the next few months called Perspectives on Machine Consciousness, where we brought together 30 of the world's leading thinkers is to provide chapters. But let's go back to my definition of intelligence.

39:59Daniel Hulme:We'll put that in the show notes, by the way. Yeah, great, great, great. So often people confuse and conflate consciousness with human-level consciousness and also intelligence. If you ask people what is consciousness, they'll say things like, oh, it's a voice inside my head or it's my ability to know that I'm a thing in the world, it's my ability to feel or to empathize or to do long-term planning, whatever. They refer to features. And those features actually can be mapped to my definition of intelligence. They're features that either allow you to make predictions and plan or learn whether those actions and plans worked or not.

40:31And so these features that we've developed to allow ourselves to be more intelligent, actually, if you imagine each feature is like a segment on a color wheel. And if you spin that color wheel, if you put all of those segments in motion, then what happens if you have all of the colors? white emerges and actually arguably consciousness is just a emergent property of the interaction of all of these systems and features that we've developed to allow ourselves to be more intelligent and when you stop the color wheel and then try and look for a consciousness it disappears which is why i think people won't ever find consciousness because it's the emergent property of all of these things that are in motion in our minds and body and so i actually think it's the wrong question to ask, I think the question we need to be asking is, what's the smallest size system, number of segments, and in what motion does something need to feel pain?

41:22That's the first question I'm interested in. Because that's actually the moral problem we have to solve for. How do we make sure that we're not creating machines that are mass suffering?

41:34Daniel Hulme:What is the application of artificial intelligence technology then as a force for good? What are you most excited about in terms of, you know, I don't know, addressing global poverty? What are you most excited about in terms of the application? Well, I do think that, you know, models right now are a little bit like intoxicated graduates. And the question you should be asking yourself is, I've got a task. Would I hire an intoxicated graduate to do this task? And if the answer is no, then don't hire an AI. But we are seeing now at Graduates, we're like a master's level where it could do rudimentary reasoning.

42:07We're starting to see now PhD level capability. So the ability to actually do science, I think in the next few years, we'll have a postdoc level so that something can actually apply complex scientific apparatus solving very complex problems. And by the end of this decade, we're going to have a professor in our pocket. So what I'm excited about is when we get AIs to actually be able to do real science, it will push the boundaries and knowledge of humanity.

42:29Daniel Hulme:Yeah, through the roof, I suppose. And I think, yeah, that's really exciting. Daniel, looking back now, in your armchair with your cigar, looking back over your career, and I know you've got a long way to go. what advice would you give to your 21 year old self? Oh gosh, I usually give this advice to everybody, which is if a genie gives you three wishes, what do you do? You ask for more wishes, right? So the first thing I would do is get as much advice as you can. The problem is that my 21 year old self probably wouldn't listen to that advice. And the other thing I guess what I've learned is that some people sometimes get a break.

43:04I've never had a lucky break. I've had to work really hard. And you should assume that you're never going to get a lucky break. You just have to work really hard. The harder you work, the luckier you get.

43:15Daniel Hulme:Yeah, absolutely. That's fantastic. And in terms of AI adoption and things like that, what type of tools and what do you point people towards to upskill themselves to get better at integrating AI technology into their daily life? Do you know what I mean? To upskill themselves to use the technology? Yeah, I'm a big fan of Lord and Anthropic and Co-Work. And so I use those tools regularly, but I can just get stuck in, get an appreciation about what these technologies are good at, what they're not good at, the trajectory that they're going and get stuck in. Absolutely. Great. Great answer. Can you tell us a little bit about your upcoming book, Daniel, and where people can find you, reach out to you and keep an eye on what you're doing?

43:57Daniel Hulme:I know you've got your own website, haven't you? Yeah, I've got my website is hume.ai. It's my last name.ai. And there's loads of blog articles on there for everything from consciousness to how to adopt AIs. The book that's coming out in a few months is called Perspectives on Machine Consciousness, and it'll be available most likely probably on Amazon. I've got another book coming out next year, which I can't talk about right now, but that will be helping people understand and get their head around these technologies more effectively. But just reach out to me on LinkedIn, connect with me on LinkedIn.

44:22I'm very happy to keep the conversation going. Fantastic.

44:25Daniel Hulme:How many books have you got now then? Three, is it? I don't know. I'll just keep them coming until somebody tells me to say stop. Okay, fair enough. Well, look, Daniel, it's been great catching up. I mean, some really great bits of information and tips and anecdotes and all sorts. It was a fantastic conversation. I really enjoyed chatting to you. But yeah, thanks for coming back on the Tech Leaders Podcast. We've gone full circle. It's a pleasure. Let's see you again in seven years.

45:01Daniel Hulme:I love that. That was amazing. That was Dr. Daniel Hume. As always, plenty to take away from that conversation. Obviously, Daniel has got a very philosophical take on the proliferation of AI, which I love. And I could talk to him all day. But I think there's so many things I could draw attention to. But I think what stood out was Daniel's point about where organizations should really be focusing their efforts when it comes to AI adoption. I think experimentation and quick wins are important. but he was talking about that they're unlikely to be where the real competitive advantage lies in terms of advice for IT leaders.

45:40Daniel Hulme:I think if something is relatively easy and inexpensive to do, the chances are it will quickly become available to everyone. So the bigger opportunity is in tackling the harder problems that are specific to your organization. And obviously, that requires the right combination of domain expertise, data, and technology. But I thought that was absolutely spot on. I think it's about knowing where AI can really give you an advantage. Daniel went on to talk about the importance of knowing what to build internally, what to buy, and where to bring in specialist expertise in terms of building AI capability in your organization, at least.

46:16Daniel Hulme:This is obviously going to become increasingly important for IT leaders. So I thought that was a really useful perspective, especially at a time when so many organizations are still working out where AI can genuinely make a difference in their organization. But yeah, I mean, there's so many other things I could draw attention to. But I think that was what really stood out in the context of the conversation. Again, it was great to have Daniel back on. It was a fantastic conversation. I enjoyed it so much for obvious reasons. I got a bit sentimental. I hope you enjoy it as much as I did almost seven years from when we did the first one.

46:50Daniel Hulme:But look, thank you so much for listening. Don't forget to subscribe. Give us a like. Share us with a friend. I mean, it helped. Every little thing helps us. But thank you so much for your support. We've got some incredible guests coming up for the remainder of 2026. So keep an eye on the Tech Leaders Podcast social media channels, and I look forward to seeing you on the next one.

47:17Daniel Hulme:This episode was brought to you by Be Digital. Be Digital support leadership teams to optimize cost and get more out of technology investments. BDigital and the team have unrivaled expertise with technology license management and data remediation and are therefore perfectly positioned to help prepare organizations for AI technology capability. And on the last point, BDigital have developed a cutting edge AI readiness assessment, which provides tech leaders with a platform they need to make well-informed decisions about AI adoption strategy. Go to Be Digital UK to find out more and get in touch.