Transcript of AI, Social Media & The Future of Humanity | James Poulter (Former LEGO Innovation Officer) & Shekhar Natarajan New

Brands recognize that in a world of AI slop and in a world of mass-generated content, you need to reach audiences through the lens of other humans that people connect with.

I'm really scared about basically like, you know, I'm not scared as much about like deepfake as much as I'm, you know, scared about fraud. I think we are going to see in the next 3 years is the biggest transformation of the software industry we've ever seen. So there's going to be such big unemployment problems that we are creating for ourselves.

And I think this is the biggest crisis we have to deal with, is the crisis of meaning, right?

Welcome to another episode of Tomorrow Today with Shekhar Natrajan. It's my honor to actually have the next guest, J.P. Poulter. Former innovation officer for LEGO, author, AI ethicist, and a guy who knows a lot about AI and an amazing soul. So I'm incredibly honored to have him on the show. Looking forward to having a deep dialog with you, JP. I know this has been long time coming for us. You know, we've been talking about it for a while. Without any delay, I would love to jump right into getting to know JP, who he is, what he's done, your worldviews. You know, like, I've enjoyed every discussion I had with you. So let's get started. So JP, like, you know, if people were to ask who you are, how would you describe yourself?

I mean, I spent most of my career and most of my waking hours at the moment living at the intersection of technology and communications, and particularly the way that technology shapes the way that we communicate, and then conversely the way that we communicate is shaping the technology that we have around us, and that's led us to this moment in, in the world of AI. Professionally, I spend most of my time Speaking, writing, creating content, and consulting with companies and organizations of all different kinds to try and help them navigate this moment that we're living in, which is characterized by massive change, and most of that is catalyzed by what's happening in artificial intelligence, both in terms of the frontier models as well as also just the everyday consumer adoption of these tools that we've all learned live with or are learning to live alongside. And so that's where I spend most of my time.

Very good. So that's your professional life. So let's go back to the beginnings. Yes. Like, who's JP? Like, you know, like, where are you from? Like, what is your upbringing? Like, what did you do? You know, the transition of your life. I would love to get to know you more as a person.

I grew up in southwest London, in Kingston upon Thames, which is a leafy suburban part of the city in which we find ourselves. Which I've actually recently returned to, living there with my wife and our two kids. And I grew up, yeah, in a very, you know, kind of modest middle-class family, to parents, one of whom was an entrepreneur and business leader, and still is to this day, my father, and my mother, who was mostly a housewife and worked at school and helped bring me and my brother up. And I was shaped a lot by the men in my family, in particular my grandfather Reg, who was a BBC cameraman for his entire career, who worked on the earliest seasons of some of the most popular BBC television shows of the '70s and '80s, Only Fools and Horses, Top of the Pops, Doctor Who, the early seasons of that, and many, many others. So yeah, you can find him on IMDb, Reginald Poulter, unfortunately no longer with us, but was a great man. And being around kind of the world of media and television and radio in the early days, I remember on many occasions when we would stay at my grandparents' house who lived a couple of miles away from us and still do, my grandmother who's 96, currently in hospital but doing okay.

We would be put to bed listening to cassette tapes for those listening who don't know what those were. These were like these little plastic things that you put into machines to listen to radio shows. And mostly they were tapes of things like Monty Python and The Goon Show and some of these kind of early radio comedies. From BBC Radio 4 and the Home Service originally. And that just fascinated me that like you could turn on this machine and like voices would come out of it and they would make you laugh. And so I've always been interested in stand-up comedy and in radio and like the just ability for voices and conversation to delight us and entertain us sometimes without visuals. And so because of that, I think that put an early interest in me in what we could do with the art of the voice and the art of the spoken word, whether that was, you know, kind of being a kind of lamenting, you know, kind of teenager writing poetry in my, you know, kind of university dorm rooms or—

To which university did you go?

I went to the University of Westminster here in the city and studied radio. That was my degree. So I studied radio production, radio broadcasting, because I was fascinated by this amazing technology that, you know, still persists today. It's still, I think, one of the most intimate mediums. I mean, and podcasting has obviously become its successor in many ways. But because I was really interested in technology, I spent most of the 3 years in my degree thinking about how technology was going to change radio. And so actually, I started one of the first podcasting production companies in the UK in 2006, 2007. In a response to what I had learned on my degree course, which was very sparing in terms of talking about the internet's effect on the radio. And so yeah, that's kind of where I got my professional start in life, was by starting to make podcasts actually, way before anyone really knew. I mean, like Steve Jobs only just kind of like declared the word podcast probably about 12 months beforehand, you know, which was at the description of like, he described it as kind of broadcasting and the iPod, and you've got podcasts.

Like, yeah, that's how they got it. And so that was my thing. And so I've been kind of working at the intersection of, as I say, technology and communications ever since.

Got it. So what is the role of your mom and dad in your life?

I think my dad showed me what it meant to be an entrepreneur, and he still does today. He's run his own business for for coming up on 3 decades now. And I think particularly in the kind of like early '90s when he started doing that, particularly here in Britain, entrepreneurship wasn't necessarily the logical path for most people. You know, this was at the kind of the height of the Blairite kind of era of politics, the Labor Party in power for the, you know, obviously they're back in now, but in that kind of time it was Tony Blair standing on the steps of Downing Street saying, "Education, education, education," right? The promise to the kind of the elder millennial generation of which I find myself was go to university, get a good job, join a profession, climb the ladder, right? But thankfully, even though my father had done some of that, although he had never been to university himself and been given an early start in life as a young clerk here in the legal district in central London, clerking for barristers, running papers up and down Fleet Street in boxes of, you know, kind of wrapped in pink ribbon, which they still somewhat do today.

I think his finding himself made redundant and out on his own and having to work out what to do in life rather than choosing just to go get another job, of which he could very competently have done so, chose to start something of his own and kind of follow that entrepreneurial journey. And myself and my brother, to his credit, have both kind of seen that, I think, as an example and always been slightly dissatisfied of being employed by other people. You know, you can, when you're shown that example early on in life that you can kind of make a way for yourself, then, you know, I think that puts that in you, and I've always been slightly dissatisfied with the idea that other people would decide what I did. And so, yeah, I think that's really shaped it, really shaped it.

But you ended up joining LEGO.

Yeah, I mean, LEGO was a really interesting chapter in my career.

What happened prior to LEGO? Like, where were you prior to that?

So my first kind of real, real job, so to speak, after the kind of the spasm of entrepreneurship in my kind of early 20s was working in the advertising industry with, by the grace of a chap that we've spoken about a little bit, and I'm sure many people watching will know, called Rory Sutherland, who, if you go, he's the ad guy on TikTok that you've probably seen wearing a flamboyant waistcoat and with two vapes around his neck. He's a great guy and still a mentor at a distance to me. But when I was doing that podcast production company in the early days, I was looking for guests for a show I was running for a video production company, um, all about branding and social media and technology. This was 2006, 2007, you know, Twitter had just come on the scene, Facebook was just opening up to brands, and I was like, I need to go find someone that can talk about this. And I came across Rory from his posts on Twitter, that was where I found him at the time, and I sent him a cold email I guessed his email address and sent him a cold email and I said, Rory, hi, I'm doing this show, it's about branding, can I come and talk to you about Facebook and like what's going to happen with Facebook?

Rory, then and now, is the vice president of Ogilvy and Mather, one of the most illustrious advertising agencies in the world, part of WPP. He's been in that position for over 25 years and he emailed me back like almost within like an hour. I was like, yeah, sure, like here's my assistant's email address.

This is like unlike Rory.

It's not unlike Rory at all actually. As I cut I've come to find out, but at the time I really didn't think that was the case. And so I emailed him and his assistant booked me a meeting, and I thought I'd go in and see him for 20 minutes. And I went to this amazing office at the top of— at the time that we were based in Cable Square in Canary Wharf, in the financial district here in London. And so I went up this kind of 20-story building to this, like, bright red office, which the Ogilvy kind of office was at the time. And sat down with this rather eccentric but wonderful guy, and he gave me an hour and a half of his time, and we just talked and talked and talked about tech for ages. And at the end of that, I was like, okay, this could be somewhere that I might want to do the next bit of what I'm going to try and achieve. And so a couple of weeks later, I actually emailed him back and I said, look, I enjoyed our time together. Any chance that I could find a job in amongst what you guys do at Ogilvy?

And to his credit, he set me up with 2 interviews, the first of which was with—

Why did you give up your entrepreneurial journey?

Well, I think I found at the time in 2007, running a podcast production company was not particularly commercially viable, shall we say. I was just too early. It's the hallmark of my career, I think, has been just been slightly too early all the time. But, you know, I was in my early 20s, right? Like, I needed to pay the bills. I was like, okay, if someone will pay me to go and do this, then that'd be interesting. And so that's kind of what I went for. So I went, he set me up with 2 interviews, and I went and saw the first guy, a chap called Bo Hellberg, who later became a colleague, and then after that a client of mine years later. But at the time, he rightfully said, "You have no idea what you're talking about, get out of my office." More or less what Bo had said. Bo is a very straight-to-the-point Swedish guy. But the second guy was, actually I came to find, was a friend of my late uncle Ricky. My uncle Simon Rickey was the guy, and he gave me a job in their broadcast PR team, basically working out what social media meant for radio and TV.

And within 2 weeks, I went from being an intern to having a job, and 2 weeks later, from being an intern to having a job, to then being asked to go to Washington, D.C. and train with their global team who were setting up this practice at the time called 360 Degree Digital Influence. Sexy name, which was basically all about the role that AI and— sorry, the role that social media was going to play in the landscape of advertising and marketing. And so I went to Washington and trained under a fantastic practitioner of social and media relations called John Bell, and then got shipped back to London and was like, basically try and set up a practice. And we did, me and a couple of other colleagues. And that kind of started my journey for the next 10 years of working in social media. In a variety of different agencies, Ogilvy most notably, and also Edelman, the PR company. And that ultimately led to an opportunity to go and join the team at LEGO, who were clients of ours at Edelman. But I came across a job that was being advertised, and I said, okay, that sounded like a bit of me, which was to—

So what were you doing in Ogilvy and Edelman? Like, what was your job day to day?

I was a social media strategist, you might be able to say. That is much more of a common type of thing that we might know of now, but in the early days, that meant basically looking on the web for tweets and blogs that were interesting and finding the people that wrote those tweets and wrote those blogs and seeing if they would do things with brands. Brands like IBM or Vodafone or Volvo.

Like a social influencer working for them?

Yeah, social influencers today, but I mean, we're talking about, yeah, kind of 2008, 2009 here where social influencer, I mean, you could have like 500 followers on Twitter and you were a big deal, right? So This was the early, early days of that, and there was always two scales of it, right? You had Obama, you know, kind of like doing 10 million followers, but basically everybody else was— I mean, if you were above 10,000, you were like really well known in those days. So we were reaching out to these people, seeing if we could get them involved with brands, getting them connected, and using some of the early tooling that existed back in the day before it was all acquired by ultimately Adobe and Salesforce to do that kind of work.. And so I spent most of my early career doing that.

And then scouting for like social influencers, connecting them to the brand.

Absolutely. And then, and then gradually—

so what is the commercial model like? Yeah, so I'm, I'm still like trying to understand like how this whole world works. So what is a commercial model that the social influencer has with a brand? Like, you know, what is their role? Like, how does that structure work?

It's changed significantly, um, these days. I think the, the role of the social influencer now is one that's much more driven by brand deals than it ever has been. In the earlier phases, you could rely much more so on YouTube revenues and then on the creator funds that came from Meta and then obviously TikTok more recently. But almost all of the revenue now is in deals with brands, which obviously sets up a very interesting paradigm for being—

So these guys are brand ambassadors, you would say?

Well, I mean, that's the real challenge, right? It's who takes the money from whom is the challenge. There are certain brands that want the reach, and if you've got the right audience, then maybe you will be an ambassador. But are you always a user? That depends a little bit. So I think that—

So are the social influencers endorsing a brand without actually using the brand?

I would think most of them with integrity would say no. I think if you talk to some of the big guys and the ones that really have reach, they are very protective of their authenticity and choosing who they're going to work with. Got it. But the model these days is hard, and so you see, you know, all of the big social influencers now branching out into their own products, their own services, their own digital twins. And obviously the role that AI is having on top of that is changing it drastically because you can now be an influencer without having to ever sit in front of a camera.. And so that changes the dynamic there as well. Certainly even more so in the world of the written word where, you know, you can train an AI model to talk and write like you. That's even easier than training a visual model to sound and look like you. So that whole dynamic is changing quite significantly now.

Yeah, that's why I wanted to stay on this topic, like, because I think it's going to be interesting in the world of AI how it's going to, like, play out. But, like, let's stick that, you know, stick on that for a while. So basically now So I'm, let's say like I'm Shekhar, like a social media influencer, and I'm working with a brand. So what's my cut?

Well, I mean, that is somewhat in your purview to kind of give. It's also up to the brand to dictate. I mean, I now regularly am kind of given RFPs to respond to by brands where you set your price and you see what comes back. Often this is on like a—

So you would be a social influencer that the brand would reach out and say, like, you know, promote my brand?

Yeah, absolutely. I've done a few of those kind of brand deals in my own time for myself personally. And I've also been on the other side of it, of creating those deals for influencers. And sometimes that's on a per piece of content basis. Other influencers will do it on a buyout deal for a certain period of time. Increasingly though, because we're able to see so deeply into the analytics and also with the power of AI, you might say, okay, my ad will run but only for the first 500,000 followers or for the first 500,000 views, uh, and then it will dynamically get removed. So the tools that are at your disposal now for basically every commercial model you can think of are, you know, just so plethoric that it's hard to kind of nail down that there is one strict way of getting it done these days. But the biggest thing is that brands recognize that in a world of AI slop and in a world of, you know, mass-generated content, you have to have cut-through and authenticity. And to get that, you need to, you know, reach audiences through the lens of other humans that people connect with.

And that's why brands keep doing it. We often talk at Elvis, one of the roles that I hold as the head of AI for an agency here in London, about the idea of like, we need real serious entertainment, we need brands to cut through because IP at the moment is the only thing that has any real currency anymore. You know, the production cost has gone to basically zero. Becoming an influencer is the thing that you can algorithmically begin to do, right? We've seen AI avatar influencers, we've seen entire characters fictitiously created. So the only thing that really is going to make the cut anymore is being authentic and having an audience. I think it was Stephen Bartlett that said on his Diary of a CEO podcast last week that AI is kind of like a cloud that's kind of rolling in at the moment, and it's like if your brand is already above the cloud, kind of, then you're okay, like, yeah, you're up there. But to break through that cloud now is becoming so difficult to be found amongst a massive sea of samey AI-generated content. And, and that's only really going to get worse before it gets better.

So I think, I think the shape of all of this is going to change quite significantly.

Got it. So, so I'm an influencer, I get the money, and, uh, and then like, how long does it last?

I think that's the real challenge. I don't think anyone can say, you know, again, it comes back to this thing. If you've been able to break through that cloud, if you've got a brand, if you're being recognized, if you're being seen as authentic If you listen, that is when you say like you, it's like you're talking about you being the social media influencer. Yes, you being the influencer in this case, then, then maybe you've got a chance of kind of making that long-term sustainable. But you even see, you know, the biggest guys in the world, MrBeast is the one that most people will obviously go to as the example. You know, he has multiple lines of business. I mean, yeah, Stephen Bartlett, I mentioned probably 8, 9, 10 different ways that revenue is coming into him as the brand of Stephen, the brand of Diary of a CEO. Um, and you know, long list of other AI influencers that are doing the same thing, then maybe you've got it. Maybe you could sustain it. I think for many of the smaller creators, content, being an influencer, is a part of the mix. It's not the sole thing that you're going to be doing.

And almost all of these guys are trying to create, whether it's trying to create IP through courses and materials, whether it's trying to sell digital assets, whether it's going into the latest crypto scam, whatever it might be. Yeah, everyone is looking to diversify right now because, like I say, the cost of creation of content is just heading to zero.

Who creates the content in this case?

Well, I think that will— it depends, right? We're creating content right now, so some people are going to be doing it the authentic way, the original way. Increasingly, the AIs are creating content on people's behalves, and you see this particularly through tools like ElevenLabs or HeyGen. And many of the creative tools that are coming out off the back of those large language models. And increasingly brands are going to have to create the content too. I think we are heading into the age where brands really are going to have to see themselves as full-scale entertainment companies if they want to have longevity. You know, it's the reason that Kodak could have seen it coming, right? Blockbuster could have seen it coming. And they didn't really kind of get there because they didn't think about the full life cycle of themselves as a brand and being part of people's lives as opposed to providing a product. And that's almost true of every brand that you can think of now, is that we look— all brands are looking for ways of finding an audience and deeper integrating themselves into people's lived experience, right? So I think the content is coming from both directions.

It's the individual creators the brands themselves, and increasingly there's a kind of middle ground where they meet and create some really interesting stuff.

But like, the world seems to react to memes, not to original content.

It does, and then at the same time—

Why is that?

There's enduring things that go beyond the memetic. You're right, like the world does react to memes, but the world doesn't stay with memes over time. The very definition of things being memetic is that they don't last very long, right? They catch the public consciousness and then they go again. And many brands are responsible for those memes too, right? But the things that really endure, you know, kind of cut through. I mean, we started off talking about kind of my time at LEGO, like why did I join that as a brand. I think it's an important part of this story is that there are some companies, some brands, some forms that persist. You know, LEGO's nearly 100 years old as a company.

1958, right?

That's right, yeah, absolutely.

And so tell me the history of LEGO, like, you know, it's a great company, like, you know, probably one of the most innovative companies out there.

Absolutely, and has almost been broken several times, right? And it's kind of revived itself. Yeah, so LEGO has been around, obviously, coming out, like I say, for nearly 100 years, I joined them, I think, around about the 75th or 80th anniversary, I think it was, so going back about 10 years now. And I think that one of the things that characterizes that as a company, one is that it's family-owned and family-operated, which I think is really fascinating about a brand that has that enduring legacy that can be passed down through the family. It was Ole Kirk Christiansen—

How was LEGO formed in the first place? Like, what is the story of LEGO?

So Ole Kirk Christiansen is a Danish toolmaker and then wooden toy maker at the Dawton of the kind of, well, in the kind of the post-Second World War kind of era. And he is living in Billund in the western Jutland of Denmark, a small town, probably about at the time no more than 500 to 600 people living there, mostly farmers raising cows. And he is building wooden toys to hand out to the families to kind of keep up their spirits during the wake of the Second World War. There in Denmark post-occupation. And he moves on from being a furniture maker to a toy maker. He goes through a couple of episodes where the family farm and also the barn where they make the toys burns down and has to be revived. But it's ultimately in the kind of '50s and '60s where his son Godtfred, who was the creator of what we know today as the LEGO System Brick, the little plastic brick that we know, discovers injection molding at a conference. I think it was in Germany. And comes back and tells old Kurt Christiansen, "Look, these wooden toys, we're never going to kind of make it big with wooden toys.

We have to go to plastic." And so they start building these injection molds to work out how to print these little plastic bricks. But really the biggest unlock is the creation of this System Brick in the '60s, which is the fact that it has this clutch power to it, these little bricks that you can push together and they don't instantly fall apart, they hold. And that unlocks a whole other way of playing with these little bricks, because you can build things that you can put on the shelf and they don't fall over, and that kids can play around with and they don't fall apart. And so all of it has really grown from that. The brick is still absolutely central to the proposition of the brand.

So how do they come up with the, like, what is a toy? Like, you know, like, what am I selling? Like, you know, like, what's the creative process involved in that?

I mean, now that process is a highly advanced thing, but in the early stages, I don't think it was that simple. Yeah, they tried all sorts of different things. And on a couple of occasions when the company, LEGO, went bankrupt twice was because they tried too many different things, from building playground furniture to still doing lines of wooden toys. They had to go at doing dolls, all sorts of things in the early, early stages. When they really solidified around using the brick as the system, that then really helped grow the brand and its awareness, and they began to see that people would want to come and understand the brick. They then branched out into their amusement parks. So we know Legoland today, that originally started as essentially a kind of show-and-tell area outside the factory in Billund to show toy buyers from around the world. They would fly into Denmark and Ole Kirk and Godtfred was like, we have to have a way of showing what the brick can do. So they started building model villages in Billund out of these little bricks and people would come and see them. And so they evolved from there, but then, you know, really the thing that became, I suppose, the popularity where it really kind of took off was the development of the different themes with inside of it.

You might remember the Castle theme in the early days or Space, and then obviously enduringly LEGO City, which is one of the longest running themes inside of it, which is policemen and fire trucks and things like that. And then later years, then building and then licensing IP from other brands. This is why I say it's such an important one because It shows that brand and IP is one of the most important things to kind of endure in culture, right? When they first started licensing the Star Wars brands from Lucasfilm, later the other brands from Disney, Harry Potter, which has been one of the most enduring from Warner Brothers, and then building their own IP through things like The Lego Movie, The Batman Movie, Ninjago, and a number of other themes.

And tell me, like, how the licensing business works, because I work for Disney Disney. Yeah, I work for the consumer goods. So, uh, explain to the audience how this works, like the licensing model, IP.

IP and licensing is shifting drastically at the moment because I think IP holders realize the power that they have, and IP buyers are the ones who are often on the back foot in this. LEGO is an interesting one because they have both the cachet of the brand themselves, and like everyone wants their brand to be immortalized in LEGO, right? And at the same time, they also are fighting against other competitors whether it's Mattel or Playmobil or other brands, to get the IP off the shelf, you know, whether you want to do the latest Peppa Pig or Snoopy thing, or you want to do Harry Potter, right? Other people want those brands too.

So those brands are creating the characters, the characters are sold as IP to LEGO, LEGO would go and print it, every time LEGO sells something, LEGO has a piece of the money.

There's a bit of money going back.

And then the money goes back to the—

And in some cases in both directions, right? So, I mean, The LEGO Movie is a fascinating kind of circular IP example where the movie is all about LEGO, right, the IP of LEGO, but it licenses and leverages all of the IP of all the different themes that they've created, including things like LEGO Batman and Wonder Woman and all these kind of characters. Those are all Warner Brothers characters showing up in a LEGO movie that also features characters from Disney, which you never usually see those two things in the same place. And then the movie itself is made by Warner Brothers, so the, the loop is absolutely complete there. These days though, I often talk about IP is that there's 3 models of IP. You can either buy it, you can build it, or you can broker it, right? Those are the 3 ways that you get to IP. And brands at the moment—

like, you're buying it in perpetuity?

Yeah, so you, you can buy your way in, right, to IP, right? And that might be particularly like we were talking about before, influencers or social media. If you are a boring brand, you probably have to buy some IP, right, somewhere along the line. If you want to be, you know, if you're, um, Home Depot in the US or B&Q here in the UK, you know, you sell DIY goods. Maybe you're not there, you sat on a storehouse of IP right there, but maybe if you go to, in the UK it would be like Homes Under the Hammer, or you go to Chip Joanna Gaines in the US and you say, hey, you guys have got some really interesting IP, I'm going to buy that in. That process I think is the one that lots of brands go to immediately. But what's changing, particularly because of the cost that AI is doing around the cost of building IP, is that increasingly we're seeing people broker or build IP. Which is that now every brand can be a content creator, every brand can have an entertainment approach. And so when they do that, they begin to build their own IP or they broker it.

They say, okay, well actually we're building a format here and maybe, yeah, you're gonna be the presenter. You know, Shaka, we're gonna have you come in to be the presenter of our, you know, kind of, I don't know, pick your boring CRM company of choice. We want you to be the presenter. You've got some IP there, but maybe we build a show and 6 months from now we don't need you as the presenter because we've now got our IP, now we're going to broker somebody else into that IP. And so I think that cycle is beginning to shift, particularly because the cost of production is coming down, you know, ideas are becoming quite cheap to create and execute on, but also because the competition is just so high. Everyone needs to, you know, break through in the algorithm, whether that is, you know, in 15 seconds in TikTok or whether or not it's long form on Netflix, you know, even Netflix, right, are buying in the podcasts of other people now. Because they recognize that they can't sustain their own IP generation necessarily. You know, you get a Black Mirror once in a while, or you get a House of Cards once in a while, but that's a very expensive endeavor.

And so even they, the big players, are beginning to look for brands or other companies or influencers to come in and brokering new IP deals as well. So that is also changing.

Got it. So what did you do in like LEGO?

Hmm. So I was brought in along with a small team at the time, I think it was about 7 or 8 of us, to help build a social network for children. This was in 2014-15. Social media was everywhere, like in its full frenzy mode, before we'd had the kind of—

When you say children, like what age are we talking?

The age of LEGO playing kids, so really from kind of sweet spot was probably like 7 through 11-year-olds. And, but what we really wanted to do was build something that was safe, that was fun, and that was appropriate for children of that age. Not something that was addictive, not something that was going to keep them on there all the time, but something that would allow them to experience what it's like to share their LEGO creations with one another. One of the great things about the product is that kids, they build models to the building instructions, but they also play with it just freeform. They make all sorts of stuff.

And then they take a picture and send it to someone?

Well, that typically was the behavior at the time, right? You take a picture and send it to Granny or Grandma on, you know, kind of text message, or maybe you'll post it on Facebook. But kids weren't posting on Facebook, they weren't allowed to, and still aren't today, at least officially. So we wanted to create something that was inside of the LEGO Play universe, something that was safe, something where kids could feel, and parents parents crucially could feel safe for their kids to share their creations with one another and they could get feedback.

And so that's what— So it's like a LEGO community?

Absolutely. And well, that's what we created. So it was an app called LEGO Life. It scaled, I think, to over about 10 million users in the end, in like loads of different countries. It was on every LEGO box.

How did it start? Like, you know, there should have been a cold start problem, right?

There should have been, but it was building off the back of something called the LEGO Club, which for a long time had existed inside of LEGO, which was a magazine that was published in physical print and distributed through the LEGO stores around the world, and also subscription and through the post, which allowed kids to kind of see the LEGO creations that other kids were creating through them literally mailing in pictures, or sooner or after.

LEGO Book or something.

Yeah, it was a little magazine that you could get. So it was built off the back of that, and we knew that that behavior was really present, which was that—

So you were tapping into all the existing, like, you know, brochure holders, so to say.

Exactly, yeah, to begin with.

To begin with.

But also the LEGO games and the LEGO social media presence that we had for the longest time, and still today LEGO is like one of the most followed brands across Facebook and Instagram. So we had like a very captive audience of parents, particularly parents that had grown up with that behavior themselves. You know, if you think about the parents that were becoming parents in 2015-ish, it's mostly millennials, and millennials were, you know, kind of LEGO players at the height of the IP, you know, kind of boom. And so they were looking at that behavior and going, well, actually, we're not as connected with one another. We want our kids to have this opportunity to share. So I think that's what really hit the zeitgeist. And so it really grew really fast. The thing that was so fascinating was we— this was pre-AI, right? We didn't have AI to kind of do this work for us. So we had to think about how would you moderate the content, how would you check? And so we had some of the earliest versions of moderation systems running on machine learning inside of the app, doing things like image detection, detecting for people's faces, detecting for kids' faces.

Even in, you know, things like reflections and in the mirror and in the background.

But who was annotating those images?

Well, at the time we worked with a number of third-party companies, the names of which escape me, but were providing services to safely do, you know, kind of human annotation of images. But of course we faced the same challenges then as we face today, which is that most of the content that is out there on the internet is still being annotated and viewed by humans to build, you know, kind of systems of human feedback, right? Whether those are repetitive or algorithmically driven. And so at the time, we were really reliant upon that technology. But we built those systems out and basically created a moderation pathway where real humans, both within LEGO and through some of our external providers, were making sure that the content was safe for kids to share, which broadly it was. We ran into some incidents along the way, like everybody does, but really allowed the community to grow and scale.

How do you moderate bad actors?

That was one of the challenges then as it is today, which is the tech now is beginning to get to the point where algorithmically we can have AIs that do this job for us. At the time, we had primitive versions of those things, mostly watching for keywords and pattern detection. But, you know, this was early. This was early stage stuff. Today we would have large language models running alongside other large language models and monitoring the output of agents monitoring other agents. But in those— that time, you know, 10 years ago, that was a real challenge. And so that was one of the first things we had to overcome working with some of those early vendors who at the time were mostly doing the same thing for lots of the online gaming communities. So things like Xbox Live and PlayStation Network, many of those types of services, Twitch, some of those early streaming services were using similar solutions at the time where they would be watching for keywords, they would have humans in the loop also being in the communities watching for users that were getting a lot of repeat behavior, those types of things. So yeah, it was a real challenge and still is today.

I mean, it's one of the biggest issues that we face is keeping, you know, kind of not just kids, but people of all—

Why do you think, why do you think like social media has that bad actor problem? What's the underlying problem?

I think the world has a bad actor problem. I think that's the— I think that's—

No, but like why couldn't they moderate it on day one? There's a— like, because Lego is like such a perfect example, right? I wouldn't assume that the Lego user is a bad actor.

No.

But like they still suffer from the bad actor problem. So what's missing in that? Like, is it information analysis or is it like credibility management or like what is in the platform that is missing that is actually driving that behavior? Or is it just like, you know, you're trying to cycle through the demand and the supply unregulated that causes like these, you know, actors to pop up?

I think the bad actor problem comes not really from a deficit of the technology, but probably a deficit of the human soul. There are always people that for whatever reason want to go and cause harm to others, to be nihilistic, to create chaos in the system, and they'll just go wherever it's easiest to kind of get in. I mean, inside of the walled gardens of things like those networks that we built, we try to make it as difficult as possible for them to do so, right, through password protection, through security layers, through moderation, through primitive at the time AI and now much more advanced AI. But every system that's built by humans can be attacked by humans, and people unfortunately will always choose to do so. I think that particularly whenever communities of interest gather, there is always a kind of shadow community of people that don't want to see those communities gather, that don't want to see joy perpetuate, and so someone comes in and tries to disrupt it. That's the real challenge.

I will tell you this example of what happened recently, okay? So knock on wood, like I've been on the road talking about Angelic Intelligence. Yes. Quite a bit. And, you know, like Angelic Intelligence was not even a thought in people's head. And, you know, I wanted to— it was not just a— it was not a wordplay. It was like, you know, I was trying to consciously put that word out there because I wanted to promote intelligence that had good behavior. Yes. And hence, hence the word Angelic Intelligence. And we gained close to— we got up to 2.4 billion views. Yeah. Okay. And, uh, close to about like 4 or 5 million like subscribers like all over. Um, and then I get— we get this weird email right? And it said like, you know, you publish some content which is copyright violated, right? Like it's a DMCA, right? Like you need to respond to it immediately. And oh, by the way, like, you know, we want you to like log in into this account. Yeah, like it looked like just like pure Meta file, like, you know, with Meta underneath. Yeah, everything looks real, looks super real, right?

And I'm like, hold on, like, you know, what content did we put? And like, you know, let's go basically to the account page and figure out, like, you know, did we really violate something or not? Like, and, um, and, and very soon we found out that all of this was a hoax. Yeah, sure. And, um, and it was like a way for someone to come hack my account, take like ownership of my account and all of the, the credentials, and also the following I had, change that.

Yeah.

And sell it to someone else.

Yeah, yeah.

So someone was trying to take, like, you know, like my— like all of the effort that we put in, like, you know, getting all these billions of, like, use— like, you know, views and, and what was happening, uh, someone was just like following that and then just wanted to grab it.

And the, the thing— and the, the scary thing, but also the thing that we need to be really mindful of— is that at the scale now that we can do these things agentically That can just happen to everybody and anybody, whether you've got 10,000 followers or 100,000, you know?

And so— How do you prevent yourself from all this nonsense?

Well, I think that there's a number of ways that we have to kind of get across this. I don't know if you want to kind of go there now, but we can kind of go through it.

I want to stick on the LEGO story a little bit. I want to get the LEGO story behind us and then we'll go to the CI/CD.

Well, so I mean, we had to tackle that then, we have to tackle it now, which is that every line of defense is kind of helping keep things safe as best as you can, but they're all built by humans. And so therefore there are humans out there that will both try and take advantage of it and can undo the defense that you've put in place. So you're constantly trying to kind of keep up in front of bad actors. So that's what I was spending my time doing to begin with, Working at Lego, building social networks for kids, and then kind of we reached like kind of point where like mission complete on that. I have a short attention span and I began to think, okay, well, what are some of the other things that are happening in the world that we should play around with? And that was where I began to return to this interest in audio in particular. And at the time, Amazon Alexa, which— sorry for those that are listening and I've just triggered like 7 devices in your house— But Alexa was kind of just entering into the European market and gaining rapidly in popularity through the Echo devices.

And I thought, maybe there's something here for us to do with kids. And so we started building, in conjunction with the team at Amazon, a number of early experiments to build what were called Alexa skills. They're still called that today, which are essentially little apps that run inside of your Amazon Echo, inside of Alexa. And we built a couple for kids, things like a daily fact of the day about LEGO, and then we built something for LEGO Duplo, which is the bigger bricks that kids, the smallest kids play with, that would allow kids to play alongside an audio story that they would talk to on Alexa and then build with the bricks. And that unlocks some new play patterns that we hadn't seen before, which was that actually you could have this narrative storyline going that would help guide the play of children. I thought this is really fascinating, and so we then went from thinking about it for kids to thinking about it for adults, including some of the early sets where you would have like an audio podcast or story that you can listen to whilst building, which we found adults were actually also really inclined to do, particularly those that wanted like digital distraction to kind of be pushed to one side and they would focus in.

And so we started building some of this stuff within the team that I was working in there called the Emerging Platforms team, which was an innovation unit with inside of the LEGO team. And that's kind of what I then spent the next couple of years doing, was building some of the first AI, again, primitive AI, right, using natural language processing and text-to-speech and speech-to-text experiments. And we began to see that that was really, really powerful. And that was what ultimately led me to go out on my own and found my first company, Vixen Labs, which was focused on helping brands of all kinds try and leverage that technology. And that's been a big part of the story since.

So people, people like actually, um, not appreciate as much this whole movement around NLP. Can you demystify it for people? Like, what is it all about? Because yeah, converting voice to text, yes, right, and the other way around, and the other way around, sounds very simple, but it's like super freaking complicated, to be honest with you.

It's really hard. It's really hard. It looks simple now because ChatGPT can just like do it, right? Yeah, it looks like, oh, and when it doesn't work, people are like, oh, that's rubbish. But it's like, do you know how hard that is for that to work? Yeah, in every language, in every, you know, every part of the world.

And then the who's and the hams and yeah, pronunciation, and it's so complicated. So unpack it for us a little bit.

Like, so I mean, natural language processing, it's as it sounds, is the processing of natural language, right? The speech that we all speak. Up until Up until kind of very recently in the scale of computing and human history, you know, you could kind of get a computer to understand a word if it heard an audio file, but it would have to literally decode that audio file and look at like the process of the waveform and work out where punctuation was, and it could like take a command. And that's like one like miracle in and of itself that it could do that, right? That we could convert kind of decibels and that we could print, you know, speech patterns and synthesis. And turn that into code, and that code could be interpreted into a word that, you know, is made up a bunch of Unicode characters. Natural language processing, though, is like a whole nother achievement, which is not just to say that these are the words, but these words work in conjunction with one another in the form of sentences, in the form of parts of sentences, and, you know, at the scale of paragraphs and now, you know, entire-length books.

That an AI model through machine learning, through recursive self-improvement, could begin to work out, "Ah, these words follow these words." And so natural language processing is what we've kind of got to now where models can hear sound, interpret that sound in long forms, and then turn it into meaning, and then infer that meaning and take an action. And for the longest time we would call that an utterance, right? What you would say to a model. And then we would get some kind of intent that comes from that model. An intent being, well, what did these words mean? And that might trigger on your Alexa, it might, you know, set a reminder or a timer or give you the weather. For Siri, it might do something if you're lucky, you know. And for Google, you know, it kind of gave rise to the Actions framework and many other things. All of that needs to exist and work well before you get anywhere near a large language model, which is what we know today, which is a whole nother scale of prediction going on, on top. Because if you can't break down just what a word is and what a word refers to, then you don't get very far.

You certainly don't get AI as we currently know it. And so that's, that first phase was what we were working with, which was natural language processing. And it's still fundamental to any of these devices that you use today. Like, no matter how powerful the language model is on the other end, like, you, there is a certain amount of processing going on on the front end still for most of your devices to understand text and speech and vice versa.

A lot of companies have somehow, like Amazon as an example, figured how to crack this. And there are quite a few companies which have actually gone through being able to do this. But why has Apple, of all the companies, struggled with this? I don't even understand. Siri is like bullshit.

Even today, right? I mean, it's getting a lot better. Let's give them some credit, but it's getting a lot better. I mean, Siri was originally a company that was bought in. It was an external technology that was bought into Apple. Most people don't think of that. They think that Siri was something that Apple created. They didn't. It was created by a team of people, external Siri Labs, originally Siri Labs became Siri Labs, started by Adam Shire, a friend of mine, and a few other tech guys. That were then bought by Apple and then integrated into the early version of the iPhone. I think it was the iPhone 3GS, I think, was where Siri first shows up around about 2012. So way before the intelligence layer that we have today.

Yeah, because like, you know, I like, well, Steve Jobs was a big fan of it.

Absolutely. And at the time it was actually pretty good for what it could do. I think when you then hold it up today in comparison to what some of the other language models can do is probably where the challenge is. And the fundamental reason why is because Apple takes a very different view view of user privacy than many of the other companies do, right? Google is listening to everything you give it, like, just hands down it is. And that's, you know, that's part of the user agreement. Apple takes a much more privacy-focused approach, which means that they're not processing all your personal data. Inherently, that means it's not as good at doing things with your personal data because it's not looking at your personal data, right? There is, like, the problem. Now, I think we're seeing that shift change, With the latest versions of Apple Silicon, particularly the A19 chips that are running in the new iPhones, the amount of on-device processing that you're able to do, and by that I mean you're able to look at the text and look at the speech and the data that lives on the device and it never leaves the secure enclave of the chip and have to go to the cloud to process, means that you can do way more on the device, right?

You don't have to go to the cloud. Edge AI. Precisely. And so that local edge processing might mean that Siri and all the other tools that work that way get a lot smarter without sacrificing privacy and security, right? You have to accept that everything you say into Claude, into ChatGPT, into Perplexity is being processed in the cloud. It's not happening on your local device. Now, it doesn't mean it's unsafe. It doesn't mean you can't do things to protect yourself, but it's ultimately being processed in the cloud.

Is processing in the cloud faster or is processing on edge faster? I'm sure like edge is faster than the cloud. Edge is faster in the cloud for Siri.

So does for small processors.

So for, would iPhone, like, in, you know, or like Apple have significantly a better advantage in the future than right now?

Some would argue yes, if they can continue to get the, you know, the processing power, uh, onto, yeah, with 2-nanometer architecture using A19 chips and whatever the A20 or whatever comes next. Um, and certainly if you look at the processing power of something like the M5 Pro and the M5 Pro Max chips that have just come out into the new MacBooks, Maybe not on the iPhone, but certainly being able to do like local model run. Like, I mean, you can now, I've seen examples of people running Minimax and Qwen 3.5. These are kind of like large language models that are open source that you can run locally, running pretty darn well on—

But would you trust Qwen? Because like the data would go to China, I'm sure.

Well, not if you're running it locally. So this is the thing. If you're downloading that model open source from Hugging Face or wherever else and self-hosting it through Ollama or through OpenRoot or wherever else you're putting it onto your local device, device. That's the whole point. You're processing it locally. I'm running an M1 Mac Mini at home, an old one, 6 years, 7 years old device. Even it can run some small local models like the open source versions of the Gemma mini models from Google and some of the small QWEM models. Even that, it runs slowly, but like it can run it, you know. And so if I want to do simple things like transcribe a video, the open source version of Whisper, which is, you know, the small language model from OpenAI that does text-to-speech, It can run for an hour or two and it will process an hour's worth of audio on its own. And for that type of task, I'm not sitting babysitting it, so I don't really care. Just running in the background. So bigger chips, they're coming. If Taiwan stays open, free, and, you know, kind of has enough power, we will probably get there.

But Apple and everybody else are going to be in the same fight for silicon and carbon. Like, you know, that is the, that's the race that we're about to enter into.

So now coming back to Lego.

Yeah.

So you started building almost like basically like a playground where you could like interact with like speech and like start building like, you know, like games or like, you know, themes of like stories and so on and so forth. And so what happened after that?

Well, so we found that that was really, you know, kind of the early days of being successful.

Would that change education?

Well, that was one of the plays that we were thinking about, was education, particularly for when you're still forming language with young children. A lot of the science was showing us that, you know, kind of the more words that you say out loud, have read to you, and that you read, is the thing that really kind of shapes formation of language, particularly for young kids. So educational was one of the big plays there.

Would LEGO ever enter into education business?

LEGO Education exists as a business, so it's one, it serves the education market, that LEGO is used massively in the education space, and so has a lot of—

But not in the traditional way, meaning like, you know, like how do I learn math and all that?

Well, the LEGO Education and the LEGO Foundation, which oversees that work today, I think I'm right in saying, has a lot of direct engagement with educators. On STEM? Yes, not just on STEM, but in all fields. You know, you can use characters for story play, you can use it for language development, you can use it for being creative in all different sorts of fields. But STEM is obviously a large part of it as well, particularly in things like programming and robotics. And so, yes, because I've seen the opposite.

I've seen my son play with like, you know, all these printed, like, material, which looks like a Lego, but, like, you know, you can say, like, you know, this plus this is equal to what.

Yes, exactly. Right? Yeah, and so the fact that it has those studs, it has that combination effects, and you're playing with sizes and scales is massive in terms of the educational potential of it, for sure. Yeah, it's, I mean, it's still one of the tools that I come back to many years after leaving, where I regularly bring it back into the workplace, as well as also with my own kids. To help kind of build skills of all different kinds, as well as having fun building from the instructions.

So like, now take me forward, like what's next after that in LEGO?

In the LEGO team? So, well, so I spent some time working on these kind of new projects, and really that was the thing that kind of catalyzed me to actually kind of get out into the marketplace and start thinking about AI in a different way. Alexa in particular, here in the UK, we saw a massive takeoff of the Alexa devices really high adoption. Our research that we went on to do at Vixen Labs showed that very quickly, by kind of the 2018-2019 mark, it was something like a third of all homes had access to one of the devices. Very soon during COVID became more like half of all homes in the UK had a voice or smart speaker. And so that was the direction of travel from there, was to really kind of dig into what did it mean to have this kind of always listening computer somewhere in your house that could actually do things for you. I mean, it's the earliest form of what we might think of as agentic AI. And so that's kind of what we really dug into next. So my co-founder Jen Heap and I started a company to basically try and help brands of all kinds leverage that moment that we were kind of looking into.

And so we very quickly began to do that with a whole variety of brands from the BBC uh, Diageo, um, the drinks manufacturer, and then, uh, started doing things in the entertainment space with Sony Music. Um, and so we quickly began to see that there was lots of different places where voice technology could unlock a different way of interacting.

Like, what would Diageo and like voice have to do anyway?

Well, so one of the first applications that we built was something with our friends at a company called Say It Now, who do audio advertising, um, for Talisker whiskey. Um, if you're a whiskey drinker, it's a deeply peaty whiskey from the Isle of Skye in Scotland. And one of the unique things about that is that they've had a very, very long heritage of master distillers making the whisky in Scotland. But most people never get to go to Talisker. You never get to go to Skye. You never get to go and do this tasting experience that Talisker have when you go to the factory. And so we thought, what if we could bring that into people's homes? And so there are about 4 different variations of Talisker whisky at the time. And so we created an Alexa experience where we went and recorded with the master distiller in Skye, that all of the tasting notes, the way to taste whisky, the way to do it properly. And then we built these kits that you could buy from Diageo. They would send you the 4 little vials of whisky to try, and you would talk to your Alexa device, and you would hear the Talisker distiller walk you through how to taste the whisky properly at home.

And again, it was this physical and digital audio experience that that people could do in their homes, and we've gone on to do a number of other things in that domain since with them.

But would that increase, like, the sales because, like, you know, someone was guiding you through that?

We definitely saw it, particularly around things like Father's Day weekend and, you know, kind of, like, sales opportunities. We promoted it on pack, it was in stores, and yeah, it had an effect. I mean, it was an early innovation test, so these things, you know, kind of scale over time, but what we saw is that people had a deeper connection to the whisky that they were drinking and ultimately a higher affiliation with the brand. And again, it comes back to this thing we were talking about before about IP, right? It's that you have to be creating IP, which is more than just the product these days. Experiences is what kind of people are looking for. And so that was probably an early phase of doing that.

Okay, that's a great example. So basically, you were trying to play around with— so you were trying to bring the voice of the winery or the distiller or whatever it is, and then basically presenting it to the —so that's an authentic voice. How do you make sure it's an authentic voice?

Well, one of the big things is working directly with the people, with the knowledge. And so we recorded original audio with the master distiller at Talisker. And that's true of a lot of these experiences, right? You know, in a world where you can create any voice now with AI and make it sound pretty real, knowing it's the voice, I think, makes a big difference. How do you validate that? Well, at the time, with someone like Taliska, Master of Scylla, no one knows who that voice is, right? So, but we were in the pre-generative AI age, right? Yeah, now, like, it's all like— Now it's much harder. Now it's much harder.

Taliska with a T, Taliska with a D.

That's right. Yeah, yeah, exactly. It's much more tricky now.

And then it's like a remix.

Yeah, I'm hoping that— Well, I mean, this comes back to what I was saying before about, you know, this is where that cloud of AI is making things difficult. If your brand is not recognizable above that cloud, you are going to have challenges, right? People knowing, you know, people could just take a 15-second clip from this podcast and recreate either of our voices very quickly. And sure, they might need to do some verification stuff, but it's not hard to get around. Sadly, it's much easier than most people think. Um, and so, you know, we are going to face this challenge of authenticity, particularly when it comes to voices. Even more so when it comes to likeness, you know, or visual likeness of people. And so yeah, that's on the horizon of being a real challenge.

I'm really scared about basically like, you know, I'm not scared as much about like deepfake as much as I'm, you know, scared about fraud. Because yes, you know, every time like, you know, like you call a bank, it says like authenticate your voice. Yes, right. You know, like say this number or whatever it is, like who are you? And that What could be a digital imprint of you, not permitted by you, but someone can play on your behalf, hack into your account, siphon everything that you have, and the next thing you know is you have zero.

Yes. I mean, thankfully, true kind of like cybersecurity-level voiceprint matching is still very, very sophisticated to the effect that like your voice pattern is so hard to mimic to an absolute 100% accuracy that— But 11Labs is able to like, you know, It's able to get pretty close, yes, but in the moment, on the fly, generating a random code with a voiceprint, we're still not at that stage where we can break that level of biometric security. But will it come? I think if you listen to, for example, my friend Gregory Richardson at BlackBerry, he would say yes, we will probably face that cliff coming soon enough. So, you know, that is something we have to prepare for, that almost all forms of biosecurity are going to be up for challenge.

Did you hear about the story about this lady who actually fell in love with like, uh, one of these actors. And she thought like, you know, like this guy is the real guy, and she ended up giving him like $800,000, like, or pounds, yeah, for savings.

I mean, that's—

if that's— who is that actor? I don't— I forget the name of the actor.

I'm not sure. But I mean, these stories are gonna come thick and fast in the next few years.

She thought that this guy was like real, by Yeah, absolutely.

I mean, does it surprise you really? Like, you've spent a lot of time talking to ChatGPT. I don't talk to ChatGPT.

No? I don't talk.

Which one do you talk to?

I don't talk to any of these things. I'm sure you do. I just like type.

In your darker times. No, I don't talk. Well, even if you type to them, but you spend a lot of time typing to them, right? Like, they feel real.

They don't feel real. They don't sound me. Like, the way I write, the way I talk, the way I—

Oh no, they don't sound like you, but if you talk to them as them, Yeah, obviously, like, you know, they're psycho fancy anyway, you know, they're like, please, I shit on you. Yeah, and that's very appealing to people. Like, if someone rings you and says, like, I want to do all these things for you, it's very appealing. This is what we're going to have to wrestle with. Deal with, yeah. Because, you know, we talked a lot about social media. If anything characterized the social media age, it was the monetization of attention, right? That was the problem. And an extraction philosophy. Absolutely. But the extraction philosophy of AI isn't attention, it's affection, right? These things want to please you. And is it a good thing? No, it's not a good thing. I don't want people falling in love with these chatbots. But no one is preparing people for the fact that that's going to happen at scale. It's going to happen at scale.

Yeah, and very fast, and like it's going to be dangerous.

And it already is, as always, right? The future is here, it's just unevenly distributed.

So talk to me about like the generative AI world. Like, you know, like, like, for everyone who wants to understand, like, what really happened in 2016 or '17, why are we here in 2022, like, you know, with ChatGPT coming, and then what's happening right now, like, that whole transition phase from what, like, Geoffrey Hinton started doing to basically, like, yeah, what happened in the lab to what came out of the lab to basically the whole transition now into the world that we are living in. Yeah. And I think, like, a lot of people think that's AI, but that's not AI. AI. It's generative AI. Yes. And like, now we are talking about agent AI. So like, I want like people to understand what is generative AI. Yeah. What is transformer technology? How do you understand it? And so on and so forth.

Like, I would love to— well, so let's, let's think about the kind of the current moment that we're in and then maybe kind of work back through to it, right? We are at this weird tipping point where you're going to hear a lot this year in 2026 and beyond about agent agents, right? AI agents, agentic AI. That seems to be the phrase that is characterizing the moment that we're in.

How's that different from RPA?

So it's different from RPA, or I would say— What is RPA? Well, okay, let's kind of break these things down. I would— let's use a broader phrase actually, which has come up a lot, which is conversational AI, right? This was the phase that we have been in. I've often described conversational AI as you talk to an AI and then you go and do something, right? Now that's where we were with Siri and with Alexa and the early versions of ChatGPT. Where we are with agentic AI is that you talk to a thing and it goes and does the thing for you, right?

So you're giving up the agency for you to do things.

Precisely. Yeah. Okay. We are soon going to be in the space of organizational AI, right? And that is when you talk to an AI and it goes and talks to a bunch of other AIs and they decide what to do. It goes from the one to the many. And that's where we're heading to. But at the moment, we are in the agentic form, which is where we have begun to give up our agency, to use that word, to these AIs to do the work for us. Okay, how did we get there, right? We got there off the back of large language models. Large language models got there off the back of machine learning in the broader sense. And without trying to unpack all of the science, of which I'm not the expert to do that, but let's just use the kind of modern parlance for it. It's like a large language model, are you ready for this, is a very, very large model trained on a lot of language. Thank you very much. There you go, right? Like, it is the thing that is happening in your head all the time for speech to come out, right?

That you have been training a large language model in your head and in mine since, you know, since birth, which is that you have seen and heard and, you know, written various versions of words. And over time, teachers and parents and and the person at the school gate and the shopkeeper have told you that's the way these words go together. And your brain has made those synaptic connections over time across millions of neurons, trillions of parameters, and worked out that when I say Jack and Jill walk up the hill, that hill is the right word at the end of that sentence, right? Because you've seen it loads and loads of times. Broadly speaking, these large language models have been trained like brains in a vat, which is that a bunch of neurons shown loads of language over time and through process of repetition and recursive self-improvement, which essentially means seeing a thing, being told yes or no, did this work, is this not, and then going back and learning it again, which is the same thing that you've done to be able to learn to walk, to drive a car, or to speak. That is how they have began to grow a knowledge of which word comes next in the sentence.

That gives you the large language model. That's the token. That's the token. The token being a fraction of a word. Why not a word? Well, because words are made up of bits, right? A whole word is more than a sum of its parts, right? Yeah, kind of. And you don't have to have a very big word for it to be the sum of its parts. The example that's often given that GPT-4 had struggled with the word strawberry, right? The strawberry problem, which is to ask it how many Rs are there in strawberry? Well, because there's one R in straw and two in berry, it would often not see that those are two different words because those are two different tokens. And it would say there are two. Well, of course there are three Rs in strawberry if you weren't kind of paying attention. So tokens are these kind of atomic parts of words, and that's what the models have been trained on actually, which is why they sometimes make stuff up, which is why they sometimes get things wrong. The excitement of what happened with Geoffrey Hinton from Google, the transformer, everything that goes into there is essentially to say that the transformer is the technology that gives the rise of this model to be able to take one form of something, tokens, statistically predict what is the next thing that should come in that sequence of numbers, and then output you something else and transform it from one thing to another.

And I'm massively dumbing down the science, but the point being is that, like, without the transformer, we do not get a model like GPT, right, from OpenAI, ultimately using the technology that came from Google. And we don't get any of the large language models that we have today. That transformer is the thing that has made the most significant difference.

Do people understand transformer technology? Can you decipher what's going on? What the heck is happening in transformers?

To an extent. The transformer is both the most amazing invention and also the hugest black box, right? It's like somehow an individual transformation, this statistical prediction of which number comes next, is a kind of logarithmic, you know, kind of algorithm working. The real weird dark magical mysticism thing that's kind of going on, which is why all these guys seem to suddenly have an interest in faith and science again, is because at scale, which of the transformers decided to do which thing, we can't really see because the emergent behavior. Yeah, the emergent qualities of these models, the fact that they seem to exhibit personality, the fact that they can use—

So what is an emergent quality?

Well, an emergent quality would be, you could say personality is an emergent quality. The fact that these models, when given instructions of how to behave, are able to somehow take on that behavior and exert it in all sorts of weird ways. When you talk to Claude and it kind of puts in little quips that you weren't expecting, or when the other day my OpenClaw agent, which we can talk about, you know, decided to just send the email to my assistant and sign off from Jarvis, which is the name it has, rather than sending it from me, even though I've given it strict instructions never to do that, but somehow it's still decided to just, like, do it anyway. These are emergent properties that we would broadly call personality, right? Where does your personality emerge from? Is it the neurons in your brain? Is there something soulful going on there? Is it something of the divine? Was it created in you? Was it in your parents' DNA, like, where did it come from? I don't know. Maybe it's all of those things, and that seems to be the case with these models as well. Like, where did it come from?

Maybe it's just a product of the training. Maybe it's only a reflection of the prompt that you gave it. Maybe it's the system prompt, the, the thing that Anthropic wrote to give it to Claude. We don't quite know, but something is going on in the math somewhere that is allowing these qualities to emerge. So personality is one of them, but it's not just personality. It's also skills and capability. The fact these models are able to to do hard science, to do mathematics, to compute. None of them were trained to do that. They are just emergent qualities that when given a prompt, when asked, hey, can you do this? The scaling laws, which is this idea that basically with more and more compute, more parameters, more neurons in the network, and more power, the model gets bigger and ultimately can do more things. Does it? Well, it seems to, but we don't know what things it's able to do next, and that's why the scaling laws are so interesting. Is that the more we scale the models, the more they seem to be able to do, but none of us quite know.

Do they get better or do they get worse?

Well, they sometimes get better at certain things and worse at others, which again is, you know, I often call it, yeah, I think it's referred to as the Homer Simpson principle, right? Which is, you know, Homer says to Marge, "Oh, you remember that time that I took a winemaking course and I stopped learning how to drive?" It's like certain things kind of get pushed out, you know, because as they expand, the distance from one neuron to the other gets bigger, I begin to stop being able to do certain things because I focus too much time on others. It's kind of like I'm not very good at handwriting anymore. I definitely know how to write by hand, but I'm not very good at it because I don't do it very often anymore. And that seems to be the emergent quality of as I learn to type or speak or code, I get worse at other stuff. And the models seem to behave the same way, which is odd, considering that they are silicon, not human. Interesting. So now, like, so the fact that, like, you know, I don't need to teach you this stuff, you know this stuff anyway.

So you're indulging me highly here in terms of making me the one explain the math, but I know you know this stuff.

Yeah. So, so help me understand this. So, like, when people say we, uh, you know, we understand Transformer technology and, like, you know, we are now able to apply it in an enterprise context. Sure. Right. And we can actually go solve a lot of problems. Yeah. Like, what are they talking about? Like, are they—

I think most of them don't know what they're talking about. Let's be honest. Let's begin with that.

Like, unpack that for me a little bit.

Yeah. I mean, let's just look at it kind of like through a realistic lens, right? I think that AI is going to be the most transformative thing to happen to the modern workplace that we have ever seen. More than the computer, the internet, the typewriter, or, you know, kind of the ledger, right? Why is that?

Is it because you're giving up the agency?

Because you are codifying intelligence, right? And intelligence is the thing that—

But you haven't codified the intelligence. Well, it is emergent.

It is emergent, but it's emergent in us as well.

So if it's emergent, we don't see all the emergent qualities ourselves. Like, you know, so basically, yeah, it's trying to take like 5 million emergent qualities, fuse it together, and find something creative.

Yes, it is.

So it's sort of like very interesting if it does that, but the problem is it begins to optimize the wrong things too.

Well, the reason it also optimizes the wrong things is because we're optimizing it for the wrong things, right? So none of these things at the moment, whilst we say that they're agentic, they're only agentic if they're prompted to be agentic, right? None of them have yet determined their own future and just started off they go. It all starts still with us deciding this is what I want you to do. Do, AI. This is— I'm sending you on this path. It's kind of like, you know, we can have all the self-driving cars in the world, but unless anyone chooses to get in them and put a destination in, they're going nowhere, right? They're not just going to, like, take themselves on, like, pleasure drives down Route 1, you know, and try to just go and— I don't know, maybe there'll be a cyber Tesla truck, you know, just like—

but once you do it once, like, no, once you do it once, like, would it begin to do it automatically in the future?

Possibly, but again, it still originates from the intent of the user. So intent controls it still now. At least at this stage. At this stage. Like I say, there's no Tesla Cybertruck sat on the beach, you know, kind of in Palm Springs just enjoying the sunshine because it decided to, right?

Someone got it. It's not like a Waymo car running around.

No, someone, and the Waymo car is only running around because someone's watching it do it, right? Although there was that incident where like a bunch of them like ganged up on people like in the back lot of like some dry cleaners and just hung out for a while.

Yes, that's why I was asking you this.

Which was a bit strange. So these things are wild, right? Like, it happens. But so what's happening in the workplace is going to be, you know, the significant shift for a number of reasons. One, because, you know, this is the leveling of the playing field of skills, first of all, right? I've had Excel on my desktop for 3 decades. Do you know how to do a pivot table?

Oh, I love to do pivot tables.

I don't know how to do a pivot table.

Oh, like, that was my fucking life.

Okay. I've been pretending for 3 decades. To know how to do a pivot table.

You know where I grew up, like, you know, most of the times, like, you know, like, the organizations I actually work for are traditional, like, large organizations, right? And believe it or not, like, you know, people believe in Excel more than they believe in, like, applications. Absolutely. Like, you know, once you give them applications, they say, like, no, no, no, I don't want to use that. I know I really want to use Excel. So you're forced to use the freaking Excel, right? And, um, and basically like, you know, what, like most of the decision capabilities people understood because there was explainability and traversal.

Yeah, well, let's play this out. Do you know how to use Excel because you inherently emergently knew how to use Excel? No, I had to train myself, right? So you were trained. Okay, no one trained ChatGPT to use Excel, but it can use it. But it can. And this is the leveling of the playing field about skills, which is that software, right, is going to zero. Like, AI knows how to use software better than you do, right? And so we've seen this in coding already. Like, vibe coding, right, is the phrase of the moment. Collins Word of the Year in the dictionary last year. Because you, as Andrej Karpathy said, coined the phrase, one of the leaders at Anthropic x OpenAI, tells us that, like, you give into the vibe of the model and the emergent properties of the model and see what comes out. And that's how you get coding done to the extent where Claude Code is now writing like 90% of the code inside of Anthropic. I think that that is coming for all other fields of work. I think vibe working is the future of the way that we get most stuff done.

If at the moment, if your job is using software, right? And so whether that software is Excel or PowerPoint or Word, whether it's the Google Suite, whether it's Salesforce or HubSpot, whatever it is, I think we are going to see in the next 3 years the biggest transformation of the software industry we've ever seen. Which is that you as a user will no longer operate software. You will talk to an AI that operates the software for you. And if the software doesn't exist, it will build the software for you, either for an audience of one, or it will manipulate a mass market software that already exists. And we're seeing already, right, Einstein inside Salesforce, HubSpot agents, it's happening. And when that happens, there is a flattening and a democratization of people's access to knowledge work that has not happened up until this There's also a massive compression of speed, right? Sam Altman, love him or loathe him, has this phrase which I quite like, which he says that by 2030, he'll expect agents to do what used to take humans a month done in an hour, right? Think about something that you've been working on for a month and imagine it done in an hour, right?

What does that do to your to-do list? What does that do to the 3-year planning cycle at Procter Gamble, right? What does it do to just like the average person thinking about starting up a business that was going to take them 6 months and they get it done in a week, right? We're seeing more company registrations in the US in the past 6 months than we've seen in the past 6 years, right? So AI is going to fundamentally like, you know, compress. Scott Galloway talks about these things, right, as being like a time machine, right? Like all technology is a time machine. Netflix is a time machine, you know, kind of LinkedIn is a time machine. Like it's a a time machine to professional relationships. Netflix is a time machine to entertainment. AI is a time machine for everything. It just is a straight time machine because it compresses all of the processing time of getting things done. You know, like I worry for like David Allen. I don't think he's ever going to have to write Getting Things Done version 17 ever again, right? Because like things will just get done. AI knows how to do it.

And so that remakes the economy. And then if we remake the economy— And you trust AI? There's no way of answering that because what do you mean by AI? Trust which AI? That's a more important question. Trust whose AI? Mine or yours? Do I trust Anthropic more than OpenAI? Do you trust the output of AI? I trust the output of some of the AI from some of the AI companies. But organization don't work on some, works on everything. Exactly. And maybe that's just a— So do you think later on——

It's not a governance problem. I think like, so see, like, I probably, that is the fundamental problem because people think it's the governance problem, right? So what I— Governance, not governments.

No, no, I said governance. Yeah, I'm just checking. Yeah. Yes, yes, yes, yes, yes.

Because governance is still basically like applying patchwork.

Sure. Right? At the moment, certainly.

Yeah. In the future too. Like that's the role of governance. Like basically like when bad things happen, I'm gonna come back and police you. You. Yeah, but like bad things have already happened in the society. Yeah, the problem with like, you know, AI going bad once means that it has already gone bad and it becomes an emergent quality that you cannot restrain anymore in the future. And every time a new model is actually generated, like, you know, like whatever you patched up for doesn't actually work for the new model.

No, I mean, I think that's, that's true of human nature as well. You know, we don't respond to things unless there are crises. Right? Any one of us could have said in the early dawn of Facebook, look forward 15 years, if this thing grows the way it's going to grow, what problems might emerge, right? And people did say that. And there were people that whistleblowed and there were people that said, you know, Cambridge Analytica or whatever scandal is going to emerge. And no one did anything about it. And now we're sat here with Jonathan Haidt, you know, talking about the anxious generation and smartphone-free schools and we're having to respond to a crisis that was completely predictable I think all of the problems of AI that we see today are completely predictable. And we will obviously be too slow to deal with it because we don't respond unless there's a crisis. Like, that's just the way that companies do it, the way that governments do it when it comes to governance, and us in our personal lives as well, right? None of us, you know, kind of like started out using our smartphone with like self-regulation in place.

Like, it's only when you kind of like look at your screen time alert for the 7th time today saying that you've been on the phone for 6 hours that you kind of maybe think about putting it down. Like, none of us do that because, you know, we're broken human beings and we have no self-control. Like, that's the way it goes. So governance is always going to be a problem for sure. It's always going to be a patchwork. Education is something that can be the antidote. I think that's the thing that we need to—

People don't even understand AI, by the way, for crying out loud.

Precisely. So that's why education Absolutely.

So like, you know, see, I— we've all been, you know, we've all been trained, uh, in MBA schools, uh, to think about a decision pattern. Yeah. Uh, which is like strategy, tactics, and operations. Yeah. But intelligence is not delivered that way in an organization. Intelligence is basically transactional, contextual model. Yeah.

And I'll tell you, no enterprise I've ever worked in works in strategy, tactics, and operations. Operations, because that assumes that that's a linear process, right? Yes. When in reality, operations is always happening. Yes. Tactics are always responding, and strategy sometimes catches up.

Exactly. So now, if you, if you take the transactional world, the transactional world doesn't need that, that sort of, like, you know, compute that we're talking about. The model, the contextual have a problem. The only way you can solve the model in the contextual problem, which includes human, is to mimic the human behavior.

Which is what it's doing.

Well, I don't think it is because the models are really trained to take world's garbage. Yes. Assemble the garbage and then say that garbage is equal to basically compassion, that garbage is equal to like empathy. Like, is it really empathy and compassion? And like, how I make a decision? Like, is it how I debate a decision? We have not largely simulated that whole human aspect of a decision-making process in AI. So I, I, when I, when I hear about like, you know, all the words that like people talk about, like this is agentic AI and like organizational AI, I think like the framework that they're using is the wrong framework. They're massively over-exaggerating the use cases, and they, they somehow feel that AI is going to come solve these problems, but they're going to be like very disappointed because it doesn't invoke trust in the system.

100%. I mean, I often think though is that like, as I say, I didn't make this phrase up, I always forget who did, the future is already here, it's unevenly distributed. Yes, absolutely. And I often think like, let's just say we paused all AI development now, but we made sure that everyone in the economy had access and full knowledge of how to use the tools we already have today, we would see the most radical transformation of productivity that you could ever imagine. Like, the world would look completely different in terms of how we got stuff done just using what we have today. Like, you don't need artificial general intelligence or superintelligence or the singularity for things to look radically different. Because I come back to the point of, like, most people don't know how to do a pivot table. Most people have got, like, a superpower in their pocket of productivity in the form of one of these things. And what do they use it for? Scrolling on TikTok and sending memes to each other. Yes. Right? So if that's the reality, and then think about the power of what we know these models are capable of, and that they're basically at the compute cost of $20 a month in the hands of everybody on the planet.

Imagine what the world could look like if just everybody had democratized access to it. Now the challenge is we don't have anywhere near the hardware infrastructure to support everyone actually using it all of the time. Which is a challenge we're going to have to overcome. And it's probably the, like, the geothermal kind of limiting factor for superintelligence emerging anyway. It's a thermodynamic issue. It's not anything to do with intelligence. But even if that was all fixed, this is why I think we need to get ready, because some of this stuff is going to go from the 5% of the society into the 12% of the society into the 37%, right? And as those waves pick up. The economy is going to speed up. There is going to be compression and compaction. And what happens to jobs? Well, and that is the question that I think anyone that tells you that knows exactly what's going to happen is making it up. I can give you my best guess. What's your guess? My guess is the next 10 years is going to be incredibly difficult for the knowledge economy. Because I spend a lot of time working with educators, a lot of time talking to people who are educating our young people and also in secondary in tertiary education.

And what I see is that we've got a bunch of young people who are entering into particularly— well, no, first of all, they're trying to enter into the university economy, into degrees that are trying to prepare them for the workforce that doesn't exist in 10 years' time. So by the time that they get pushed through that factory, they're coming out with the wrong skills to begin with. Then inside of the universities in particular, they are wrestling with AI everywhere already. Students are using it to write term papers, because secretly loads of professors are using it to score term papers. Essentially AI is just scoring other AI and then we're giving people degrees. Like, that's kind of happening in a mass scale. And at the same time, we know that there's new research being done, there's new drugs being discovered, there's new algorithms being unlocked, there's new mathematics and physics being done inside of our universities when AI is applied, right? And then we've got business schools preparing people for businesses that will not exist anymore. Right? So the next 10 years for anyone that's currently in any phase of that pipeline is going to look, I think, very, very difficult.

They're under that cloud that I was talking about before. For the people who are above the cloud, the ones who are already in industry, who are already picking up the tools, people who are probably the elder millennials or the elder Gen Z, the ones who are going to enter into the management practice of the next 10 years, I think they've got an awful lot to be helped, you know, to be excited about, right? Far less busy work, far more opportunity for creativity, far more time for play, opportunity to spend time doing the things that they wish they were doing rather than the admin and the busywork that we've created. And yes, there's going to be a whole class of jobs that disappear, but ones I'm not particularly upset about. Like, we've had, what, TikTok video editors for, what, 7 years? In the scope of all human history, if that goes away, I'm really not that upset about it, right? Who cares? Was it a great job to begin with anyway? Probably not. But it was someone's job, and that's the point, is that every job that gets lost, that matters to that one person. Does it matter at the scale of to the economy?

Probably not. But to that one person, the thing that goes away is their agency, or it's their sense of self or purpose.

But would it not actually push unemployment to a place where it's uncontrollable? Would it not then drive inflation up?

I've seen economists at either end of the scale argue it both ways, and I've been encouraged by both answers, which is that actually we see that people get pushed into entrepreneurialism. We see people being pushed into the jobs that at the moment they say they are not willing to do because they're uncomfortable but actually are absolutely necessary. Our teachers, our carers, people working in the blue-collar workforce. At the moment we are importing people nationally here in the UK and in the US to do those jobs from other parts of the world because most people born in those countries are not willing to do those jobs. So the idea that those jobs do not exist is just fanciful, right? We could have more people doing those things. But people want to do those jobs. Some people do want to do those things.

Like, you know, so like this is a classic argument that I hear here, like, as well. Like, like, let's take orange pickers. Yeah, in like Florida, right? Like, who works those jobs?

Immigrants. Yes.

Like, are those jobs not available to American citizens? They are, of course. They don't want to do that.

No. And they may be forced into the fact that they will probably have to do that.

So you, you would, you would imagine that—

so what makes— as opposed to being a cashier worker at Walmart True, but you would imagine that basically the very jobs that they didn't want to do in the first time are the jobs that they would go after, are the jobs that they may only have the option of, or we have some form of universal basic income.

I think about the opposite, like I think the opposite is going to happen because I've seen this like movie play out. I think all those guys are going to sit in the house and they're going to crib about like how like AI took away their jobs and they're going to get disgruntled and they're going to polarize the society. Sure. Sure, like, that's a possibility too.

That's also a possibility.

And like, you know, then we get into a dystopian world. Yeah.

And then, and then that, that is the most— that's the most crucial point, is that which do we want? Because now is the time to start deciding.

Well, like, you know, like, do we not have like proof of all of these things? Let's take like example of like car industry in, in Detroit. Yeah, right. Um, you know, it was so encumbered with like all of the pension and all of the, the, the bureaucracy that went into the system, they found themselves basically out of the job. Yeah. And then like it created like a whole class of people who just were bitter about everything that was happening around them. Yeah. And then it, you know, it kind of like came back to haunt us like in nationalism and populism that like basically we never subscribed into as a society. And so now we find ourselves in a much more polarized society like we We are much more afraid of our shadow, everyone around us. And I feel like we are just like throwing more gasoline into this problem and letting more people sit there. And all I know, like, when people have a lot of time, like a lot of people say, like, you know, with all the time I have in my world, I can go do something very spectacular. I can go sit on Mount Everest, which I could never do.

Now I'm able to go do that. There are very few people who are going to take upon that. Yeah, a lot of people are going to sit in the house. They don't know what to You know what they're going to do? They're going to do fraud. They're going to do deepfakes.

And this is the argument for the accelerationists.

They're going to do like digital arrest. Yeah. You know, they're going to come up with all the fancy ways of fucking the society. Yeah.

And the argument of the accelerationists would be, well, no, we are going to raise the overall standard of living. Universal basic income.

But who decides universal basic minimum? Exactly.

That's who we have to— that's the choice.

And so, like, so my, my quality of life was my quality of life. What I decided my quality of life— who are you to tell me, like, my basic minimum pay is X? Yeah, you just stripped my job and now you're going to tell me, like, you're going to pay me X? Yeah, sorry, I don't want it. Yeah. And that's what you do with that, like, then you go and sit in the house and then you claim, like, unemployment checks.

And so this is the challenge that we have, is that. At the same time, the democratization of knowledge and the cost of content creation going to zero may give way to entrepreneurialism in a way that we've never seen before, right? More niche problems served by more individual people with smaller companies doing more things, right? We might have a kind of artisanal economy where many more things can get built, right? Rather than we all relying upon the one big blue Salesforce option of CRMs, there is 100,000 different CRM providers built for every individual minor industry that has ever existed. Maybe that's the way forward. My observation is that most of the world doesn't work that way, that most people are not entrepreneurial, that most people's risk tolerance is not that high unless forced to do so. And so maybe the forcing factor of blue-collar work being the only other alternative and white-collar work gradually disappearing, or certainly only becoming something the elite has access to, might force a wave of entrepreneurs and freelancing creativity in the artisanal economy to emerge. It great. At least for a portion of people it might. For everyone else, we've got to start thinking about new models.

We've got to start thinking about new models of what it means to work, what it means to apprentice into these practices. Because we still, I think, for many, many things, I think at a human level, we still want to have some humans in the loop over certain things. Like what? I think our health is one of them. Health. I think the trust of a doctor across the bed from you giving you a diagnosis is one of the things that is still very hard for us to imagine that we would like AIs to be doing for us.

So a doctor's job is like okay, you mean?

A doctor enabled by AI, absolutely. But I think that anything that drives to the human, anything that drives to be built on empathy, whether that's in our caring practices, whether that's in our healthcare positions, whether it's in our arts and creativity, which we have still seen, is that even if we have convincing AI models now that can produce, you know, feature-length documentaries, and yet people still want to go to the theater, and increasingly so. We're seeing it in our spiritual practices that, you know, kind of you've got more access to, you know, kind of spiritual texts and opinions on the world than ever before, and you know what's happening in Western democracies? Young men are going back to church. Why is that?

They're becoming God-fearing, why?

Because I think there's something deep in the human soul that is broken and they recognize that they need something, they need one another, they need community.

Broken people going to the like to fix their souls, or like, you know, like you're broken in economy but you want to go—

I think both. And because when the economy breaks, it breaks our sense of purpose. And when our sense of purpose is broken, then we go looking for value elsewhere, maybe to the most fundamental place which says, who am I as a human and what am I worth? And I think this is the biggest crisis we have to deal with, is the crisis of meaning, right? In a world where most of us have built most of our identities around what we do for a job, what if I walk into a networking thing and say, who are you? Most people answer with what do they do, not who they are. And I love this show because you go into who people actually are, not just what they do, right? But most people answer with what they do. And if what we're saying is that this stuff may take away what you do as being a primary directive of who you are— Your identity is gone. Then identity is a massive purpose and crisis and meaning. These things are innately bound up with one another. And I think work is a good thing. And I think I come back to, you know, Prof G, yeah, Scott Galloway, his thing of like, there's nothing more dangerous to society than a bunch of young men with nothing to do.

Like, that is a really—

which is what I'm deeply concerned about. Like, if 30%— we should be concerned. Yeah, if 30% of the— like, you know, this prediction is that 30% of the white— or 50% of the white-collar jobs, like, you know, entry-level jobs, per Dario, is going to be gone. Yeah, right. So now you're going to create like— to begin with, to begin with. So there's going to be such big unemployment problems that we are creating for ourselves, and these guys are going to —like, you know, so bitter.

Could be. If we don't start telling a better story. And I think that's the thing that I'm trying to get.

So what are they going to go do? Like, what do you want them to do?

So what I think we need to do is that we have always in markets seen that when kind of the government intervenes to correct a market, basically it causes problems. Yeah, absolutely. And you can look at what's happened here in the UK, you know, the intervention of the government into basically telling everybody off the back of the Blairite years, telling everybody you should go 'Go to university,' and the way to go to university is to give you a massive student loan, lumber you with a load of student debt, and then go off and get a degree, right? Was a market correction by the government intervening into the education sector, which ultimately has not really paid off, as we've seen. You know, most of the people who took those degrees are not becoming, you know, kind of research butterfly scientists, or, you know, kind of working out, you know, kind of the molecular structure of hummus, or whatever it is. You know, like, they took those degrees, but they're not doing those things. And so that correction has just lumbered, you know, it's just put a load of debt basically into the ecosystem. We need to think of new models where we put that, you know, kind of the society back in structure and create new answers inside of the fiscal systems that we already have.

So as I said before, yeah, we've solved this in the blue-collar workforce for millennia, right? Jesus was a carpenter and he was probably a carpenter's apprentice to begin with. With, right? For, you know, you watch one, you help with one, then you do one. Like, that's the model of apprenticeship, and we've done it in the workforce, in the blue-collar sector, for the longest of times. We haven't done that very well in the white-collar sector. The white-collar sector is— what we've done is give a bottom slice of what I call the stack of work to the youngest people, which is the administrative, the repetitive, the boring. And at the same time, whilst we give those people things that were was economically valuable to be done at the time. We exposed them to the other parts of the work stack, mostly through osmosis, by being around people, watching them in the world, and learning what it means to be a good human, right? And then gradually you get given more responsibility. With AI, that work stack goes from being bottom to top to being horizontal, which is that you can now interact with all parts of the work of a job from the get-go in safe environment because you can do it in a simulation, right?

You can talk to an AI about management and leadership and strategy at the same time as also directing an AI to do the bottom of the stack work. And that's what we need to work out is that what are our apprenticeship models going to be? And we need to put the onus on industry, on business leaders, and on civil society to step in and say, okay, young people coming into the workforce, whilst we can't give you economically justifiable work to do because it's not economically sensible for you to do it, and AI would be more sensible for you to do it. We still need you to apprentice into this work, which means that we need to give you simulated and real environments to learn to be a good human operating in the world. Because as the— and then they do work— because they actually— so one is that we are going to have to put it back on the economies of these different industries to subsidize those young workers. I don't think we can do it at the scale of the whole economy, and I think universal basic income is a, is a very bad thing when it comes to meaning and purpose.

But giving people apprenticeships where they are paid by the companies that they will deliver future value to, to mature into, so that they can become the human in the loop in the long term that manages and orchestrates AIs to do a lot of the grunt work, then I think we're giving them meaning and purpose again. Then we're also giving them human interaction, we're putting them into society, we're putting them back into the clusters of the industries that they might do.

How many, uh, like you mentioned something like, you know, remarkably interesting when you just walked in. You said like what used to take like some 6 people's job, yeah, in your startup previously would now be like you plus like half a person, like half an AI. Yeah, right. So you're compressing 6 jobs to 1, like it could be 10 jobs to 1. So where is the— where are the jobs?

Well, I think it's multiple. One is that some jobs for now are going to disappear. Until new industries begin to emerge, new needs and new opportunities.

Like, what would that look like?

Well, I don't think I can answer that because we haven't got to superintelligence yet, which might make up a bunch of stuff. I mean, there's going to be a lot of jobs in space, there's going to be a lot of jobs in data centers, there's going to be a lot of jobs in lots of parts of the economy that don't exist today. I don't know.

Or building data centers.

Or building them. Or, you know. Like putting roads. Yeah. There's also massive parts of the world, and we forget this in the West, of the West, and I know you come from a diverse background, have lived in lots of places, that don't have access to any of this stuff, right? There's a lot of the world that lives below the poverty line. There's a few billion people that, like, the idea— I mean, I often get up about this when people say we don't want AI in the classroom teaching people. I'm like, there are a lot of parts of the world where a classroom would be pretty nice, let alone a teacher, right? Like, it's a very privileged white Western way of looking at the problem. Most of the world do not have access to running water. Most of the world do not have access— you know, I sit on the board at Christian Aid, you know, we spend a lot of time trying to help develop nations that don't have access to this. So maybe the economy isn't so much of a problem, it's just that it's not in the right places, you know. So I think that for, you know, some of the world's largest markets of poor are going to massively benefit from AI, and there'll be a lot of jobs to be done in other parts of the world where we don't currently have those societies.

So I think there are— it's not a net kind of net zero kind of thing. Like, AI is going to take a lot of jobs from places where we really don't want AI jobs to be taken right now. Boo-hoo. There's a lot of places where jobs are going to get created which, you know, have never seen prosperity before, and I'm pretty up for that, you know. I think that the great leveling of the society isn't just going to happen in Silicon Valley. This is a global thing. Again, if we can keep up with the demand, which is the limiting factor. So I'm not so sad about it.

You're not sad about it?

No, I'm hopeful. I think we have to be hopeful because otherwise what's the point, you know? I think we have to be hopeful, but we have to strive for a better narrative. We have to say that actually, you know, if you think about AI being the sum or average of everything, right, by definition 50% of people in the world are above average, right, and 50% of people are below the average. Well, if AI is the average, then to everyone below the average, AI is an absolute savior. To everyone above the average, you look down on it and go, oh, you can't do what I can do. Well, well done for you. But for everyone that's sat below that line, they can't wait to have an AI teacher, an AI mental health helper, an AI pastor, an AI doctor to come alongside them, because right now they're below the average and they would really love it was there. And so maybe we can be hopeful about that. At the same time, as—

If it's pointed in the right direction. Precisely.

And that's the moment we live in now. I think the thing that I wake up every day—

How many companies, like, I know like a lot of companies are deploying AI. How many companies are really like going after reskilling of the labor?

Very, very few. Very, very few.

Why do you think that is the case?

Well, so what we've seen from some of the surveys that we've run over the past couple of years is that when you look at the amount of companies that have adopted the tools, it's somewhere in the 50 to 60 percentile, particularly when talking about kind kind of your typical enterprise, Fortune 500, Fortune 1,000, the amount of them that have actually trained people on how to use those tools is sub-30%. And that, when I say training, that might mean that they gave people a license to a tool and they gave them an instructional video. But it's the equivalent of like saying to the same group of people, hey, like, I'm gonna— you want to get fit, right? It's January 1st, you come to me and you say, JP, I want to get fit, and I say, here's a pass to the gym, off you go. Maybe you do an induction. By February 14th, you are no longer in the gym getting fit, by average, right? Only about 95% of people, you know, kind of make it part— yeah, well, 95% of people don't make it past February 14th, right, with the gym membership bought on January 1st.

But if I induct you into the gym, if I give you a tour around the gym, if I teach you how to lift a weight correctly, maybe how to do a deadlift correctly, safely, so you don't hurt yourself, crucially, which is the important point. And then maybe if I put you in a community, in a weekly class, and get you a personal trainer, funnily enough, you might get fit. I don't know when, I don't know how long it's going to take, but if you just keep doing those things, sooner or later you're going to get fit. The enterprise approach at the moment to rolling out AI is buy people a gym pass on January 1st, and if you're lucky, you might be in one of the 30% that gave you a tutorial and gave you a taster class. Boss, but that's about it for most companies. Yeah, that's the first problem.

The second problem is what happens to all the people who are going outside the company?

Well, the shadow AI is the secondary point, right? Which is that when you ask people, do you use AI, in 75% of the cases they are using it, which we definitely know that there are people using AI that has not been sanctioned by the company. The same survey we did found that only about 37% of companies that have given AI out in a corporate setting have written an AI policy, which tells you everything you need to know, right? It's that we are not governing because we can't keep up. The governance can't keep up with the tools. And it can't keep up with the adoption because, you know, this is not like smartphones or even like voice technology or smart speakers. There's no hardware barrier. There's no learning barrier because you just talk to it. And it's basically free at the cost of nothing, right? So it's almost everywhere immediately, which is why we're struggling so hard to keep up with with it.

Got it. So tell me, like, what are your thoughts, like, for the world? What are your lessons? What are your thoughts? What are your hopes and dreams?

I come back to this story in the early part of the Old Testament on a regular basis, which is at the end of Deuteronomy, which is the fourth book of the Pentateuch, the first four books of the Bible, and it's the story of Moses, that book ends with Moses dying, right? And then the very next chapter is the first chapter of the book of Joseph and it picks— sorry, Joshua— and it picks up and it says, 'Moses was dead.' And I'm like, 'Okay, that's a weird verse to be quite fascinated by.' But the reason it's fascinating interesting is because Joshua, who is Moses's apprentice, gets told by God, "Hey, like, your job now is to pick up the job that Moses didn't finish, which is to lead the people of Israel into the Promised Land." A thing that Joshua has basically been watching Moses do for 40 years, wandering around in the desert, basically getting nowhere, a journey that if they had actually walked directly in a straight line would have taken them 11 days. They spent 40 years wandering in the desert, right? And so Joshua kind of picks up this mantle and he's been his apprentice, right, like he's probably in his 40s, you know, Moses probably is in his 90s, so probably about a similar age to I am now, and he's basically told, you've got to lead all these people of Israel into the Promised Land.

And God says to him, be bold and very courageous, be bold and very courageous, says it 3 times. And I think that that is the moment that we are in, is that we are about to have the biggest transitional generational transfer of wealth, right, a mantle if you like, or a charge of wealth from from the boomers into the Gen Xs. The Gen Xs are leaving the workforce because they're aging out of it and the millennials are the ones picking it up and the Gen Z are coming in disaffected and not really knowing where they're going. They're a bit like the Israelites who've been wandering in the desert for 40 years and we are the Joshua generation. If you're a leader listening to this, you are a part of the Joshua generation. You are being told, hey, Moses is dead, right, the guys that built this thing are not going to be around to deal with the consequences, but you've got to pick it up. You've got to be bold and very, very courageous at this moment. And that's where I think we are. I think we are in a Joshua 1 kind of moment that, like, you are looking down the barrel of the biggest change that has, you know, arguably, no matter your belief, the shape and the story of what's happened in Israel has shaped all of Western civilization from that moment onwards, right?

And I think we're at one of those epoch-defining kind of moments again that we are being told to be bold and very courageous and not to shy away from this thing. Like, you can choose to and you'll get what you're dealt with, right? Like, if you choose not to engage in this moment, you're very welcome to the future, but you had no shaping or no hand in it. Like, if you want to be part of the future, you have to lean in and be part of shaping it. And that's something that we all have to take responsibility for.. And I mean literally all of us, because we've seen what happens when we don't, when we leave it up to very few people to decide what's going to happen next. Doesn't tend to go so well. And so we need to massively educate, you know, the workforce, we need to educate the populace at large about the risks of what's about to happen and the opportunities that it presents.

Well, the opportunities is not known because like it's in the nascent stage.

It is, yeah.

But like, including the argument, like, you know, I would like to do an apprenticeship, like, you know, if, if, like, let's say, like, I go from point A to point B, like, you know, learning things, like, the point B is gone before I know, like, I can start there. Yeah, that's the, that's the rate of change.

Yeah, the half-life of skills is very, very short right now, for sure. But I don't think that gives us, like, an individual level and an opportunity to excuse ourselves. For sure, for sure.

We've got to learn as humans all the time.

Exactly. And so that And that's why I'm hopeful is that I think the self-directing inquisitive nature of the human spirit encourages me that wherever you are in the world, that people are always trying to find new ways of doing things. And this presents more opportunity to do things in new ways than we've ever seen before. And so I think we'll work it out. And so that's why I'm more hopeful maybe than I am nihilistic about it.

I also think the same thing that you think, to be honest, JP. I think like we are going to go to a very dystopian world before we realize that we have gotten to the dystopian world and we cannot see any further below. And all we have to see is up. Yeah. And then we start climbing up the hill. Yeah. And that hill and the climb is going to be extraordinarily difficult for a lot of people. But the climb has to be done.

And it's worthwhile doing.

It's worthwhile doing. Yeah, I see. But along the way, we should just make sure where there are not a lot of people who are disenfranchised or feeling neglected or feeling like, you know, left out. Yeah. Because that, like, the people that we leave behind and the obligation not to leave behind— that's right— is even more paramount in that world than it is, or it was in the past.

Exactly. And that, that's why we need this to be democratized. It's why it's something we need everyone to have equal access to. I mean, we've spent a long time in the past decade arguing about DEI, D-I-D-I-D-I. I would like us to redefine that around digital equity and inclusion. I think that's actually one of the biggest challenges we face is that this doesn't become something that only the Silicon Valley elite get to determine what happens next. That we need equity and inclusion everywhere because, as I said, come back to the idea like, okay, maybe most people don't have access to the best human that they could have access to in whatever aspect of their life they need a person to be there. In lieu of that, a great, well-aligned, well-designed AI is a good solution, but it's not an excuse for there not being a good human, right? So as we get more and more of these things to be a great human replacement, it doesn't excuse the human race for showing up, right? I don't think that's where we get to. We have to be bold and very courageous, right? We have to pick it up.

Man, I could sit with you for the next 5 hours and keep interviewing, and I think we'll do another episode of these things because I think we need to dig into like, you know, what it means to have ethical AI and other things, which is like, I think, like your chops actually, like, which I would love to get into. But what a fascinating conversation. I really loved it. I really, really, really loved it. I know I cannot like— I interviewed like so many great people today. Like, you know, I interviewed Raima, like, you know, like, you know, she's an activist from Iran. Shirin Ebadi, Nobel laureate. Like I was talking to Sir Walter, like, you know, he's like, you know, genetic scientist, like, you know, the guy who actually like, you know, proposed, you know, the Genome Project, the Human Genome Project. Right. So a lot of fascinating. And, you know, I'm trying to pick and choose like, you know, like where all of these were landing today. Everyone was like spectacular. Yeah. Including this one. So like, it's like almost like 10 hours after I have like, you know, recorded like all day long, and I still feel excited about like the conversations and what it means to the society.

So thank you for actually this lively conversation. I really loved it. I wish you all the best. Thank you. I want everyone to go like read your book. Yeah, I want everyone to start like reaching out to you because you're a fascinating guy, and I think you're going to change the world.

I hope so, in my small way, wherever I can. But thank you for the conversation. Absolutely. Yeah, tomorrow's today, right? I mean, that's what we're, that's what we're dealing with.

Absolutely. Thank you so much.

Appreciate you. Absolutely.

Wow, that was very good.