Generative AI - A Solution That's Finding Its Way Into Every Possible Problem

Richard Kramer:

Welcome back to Bubble Trouble, conversations between two humans with real voices and jokes From Will Page that are so bad that no AI could possibly have written them. Today we're represented by not deep fakes of ourselves, but actually myself, independent analyst, Richard Kramer, and the economist and author, Will Page.

And this is what we do, lay out some inconvenient truths about how business and financial markets really work. We've been training our respective brains at the dinner table topic du jour. Artificial intelligence for three episodes now, and we've generated views about banking analyst notes, academia, mixtape culture, and now we land one of the most important voices there is, a founder who's built the first generative AI company on the block, Jukedeck, and moved on to ByteDance, and now finds himself on the front line of the AI trench warfare over images. We're honored to have on our podcast Ed Newton Rex, one of few who can say he's been there and done that and is now working for Stability, the parent company of Stable Diffusion, which many people have probably heard of.

We'll be back in a moment with more from Ed as Will gets him to tell us a little bit about where he's been and how he's got here.

Will Page:

Ed, we've been wanting to get you on the podcast for a little while now. If my memory's certainly right, last time we all gathered, the three of us, was when Donald Trump was separating TikTok and you're wondering which half you're going to land on. Nice to see that things have calmed down a little since then, but let's give you the microphone to introduce yourself to our audience and especially how our audience can find and follow your work.

Ed Newton Rex:

Yeah, I mean it's great to be on the podcast. Thanks so much for having me. I've been wanting to come on for a while. So my name's Ed Newton Rex. I founded the first generative music startup way back in 2010. It was called Jukedeck and then join ByteDance, which I think most of your listeners will know. And I'm now a VP of audio at Stability where we're basically doing the same thing that Jukedeck was 13 years ago, but better now because tech has come on. And when I can, I have a side hustle as a classical composer, which I try to keep up.

Will Page:

Okay, so I've got two icebreakers for you. The first one is a question I've asked all previous guests on this particular topic. We asked Professor Chris Speed this question, we asked Jessica Powell's question, and it's just as we get into the hype cycle that AI right now, I just want to ask whether you view AI as a problem in search of a solution or a solution in search of a problem. That is, the tech use case getting ahead of the reality use case. I'm interested to see how you unpack that question but also how you answer it.

Ed Newton Rex:

Yeah, I mean I spent five months last year talking to people about a different technology, about blockchain, chatting to people I knew who were working hard on blockchain, investors, startup founders, that sort of thing, and honestly the conclusion I came to at the end of that process was that blockchain is a solution in search of a problem. And I'm now pretty open on my views about that tech, which is fascinating, but I just don't really see the use cases, as I know is now an increasingly popular opinions. I can hardly claim to be the per person who's thought that, but in my mind, AI is the reverse. Theoretically, it's hard to think of a problem that AI won't help address, but I think what's interesting about it is that we don't have to think theoretically, we can just look at how it's being applied practically today.

As someone in product management, I've already used regularly used large language models to write job ads, to write product requirement documents, to brainstorm product names, like these models have already saved me countless hours of work. And you hear stories of people who are buying another monitor at their computer terminal just to have ChatGPT open all day to help them out.

As a musician, I've used image generation AI to create album artwork and getting something that's way better than the stock images I would've used before, because the stuff I'm putting out on Spotify, I'm not spending a huge amount of time on, to be honest. And AI is incredible for that. As a composer, I recently used AI to write the text that I've set to music and it's just been recorded by a great group I know called Watches Eight. And so there AI's given me something to work with in my creative process and that's just me personally. AI is used everywhere already today. Coming from TikTok, TikTok has shown that an AI recommendation system is just a way better way of finding compelling content than using a network of your friends. And that's just one example, right? So yeah, I think this is very much as a solution that is gradually finding its way into every possible problem.

Richard Kramer:

Wow.

Will Page:

That second icebreaker for me, before Richard steps into the future with you, is I want to step back. I think you're a brilliant public speaker. I'd encourage our audience to check out your TED lectures, but one thing that you do when you address audiences is you kind of give us a sense of how to feel a bit more assured, how to take cool heads when approaching this topic. And we're fond of saying, to quote Mark Twain, "History doesn't necessarily repeat itself, but it sure as hell rhymes." And I've seen you speak, but for our audience's benefit, before we get into this topic of AI, what's the best way to take a cool head before you unpack this subject? What's the lessons from history you'd like to see to our audience?

Ed Newton Rex:

I think it's very hard to separate the kind of height and the chatter. I think one of the best ways to think about AI is not as something that has just sort of snuck up on us in the last year, but is something that has been being built up to and built on for really quite a long time now.

The first AI music systems were built in the 1950s in the same decade that the term artificial intelligence was coined. People have been creating rule-based systems for writing music and different systems over the years have sprung up. And this applies not just in music but in art as well. You've had image generation algorithms. A guy called Harold Cohen created this art-creating algorithm that he created artworks with over decades and you have different artists, you have different computer scientists over the last 60, 70 years building these systems.

So this is not something new. And I think we have to put AI in its historical context. What is new is its capability in the last couple of years, but we can maybe go back to 2017 for the start of this sort of new ramp up in capabilities but has relatively quickly been getting much, much better.

I think one of the things to do is just to look to the people who have been speaking about this stuff for a while. I think for a long time people did not think these kinds of use cases would be possible. I remember having talks with investors for my startup back in sort of 2012 where I was saying, "Hey, I think creative AI is going to be a thing and this is kind of how it's going to work." And people honestly thought it was insane. I mean just that there were a lot of people had no time for this. So I think that looking to the many people who have been in this industry for a while is a very helpful thing to do just to put it in the context of where we've got to.

Will Page:

Bold, underline, exclamation mark.

Richard Kramer:

I'd like to step in, Ed. I had a coffee yesterday with the innovation editor of the FT, who you probably know, John Thornhill. And we were talking about something he wrote about today in the FT about winners and losers of this AI arms race. And while they had a very scary article on the front page of the FT magazine a few weeks ago about killer AI and killer robots and so forth, all these dire warnings, at the same time we were reflecting on how 50 years ago in the UK, there used to be banks of typists who would type out bank statements before they were posted to you. You literally had thousands or tens of thousands of people in typing pools to type out your bank statement.

So what is causing so much fear and angst when what we're looking at is just an extension of everything we went through in the industrial revolution and everything we went through in the information technology revolution? Is it that so few people understand the linear algebra that underpins the machine learning that underpins AI? Is it that it feels somehow out of their personal control? What do you think is really causing this fear and angst that is so widely part of the discussion about AI?

Ed Newton Rex:

Yeah, it's a really good question. I think maybe what it is not something technical, but it's just the kind of feeling of one day thinking that something is absolutely impossible and the next day waking up to see that it's been achieved. And that I think is what we're seeing in generative AI. A year ago, two years ago, if you'd sort of said to someone, "Hey, you'll be able to just type any line of text into a box and you'll be able to create a photorealistic image of that text," people would just not have believed you.

But now, it's just normal. The impossible has become normal. I think there's a fear around that. Because in the short term, these systems are essentially going to be replace mostly mundane tasks. It's the mundane tasks in my job that I turn to large language models for when you need to summarize or expand something for different readers, when you need to do some quick brainstorming, whatever it is, or things that are maybe more than mundane but that are at least universal and maybe aren't your area of expertise, right?

So music, a good example for me is like mastering. I can't master tracks and I turn to AI often to master a track for me. So I think there's definitely a sort of... There's a short term where we don't... Which I think is very separate from these risks that people are thinking of. I do though think I personally believe we should be real about the risks. I think killer robots might be extreme, but it seems pretty clear that we're on a rapid path to artificial general intelligence. And who knows when it'll arrive? But we are presumably on the way there. And when we get there, really, we have no idea what will happen. Anyone who tells you otherwise, I think, either was either lying or misguided. There was that famous letter recently calling for a pause on the development of larger models than GPT-4, which I actually think is a pretty good idea.

I signed that letter not to stop all AI development, there's a huge amount that can be worked on, but just to say, "Well, hang on. We're not really sure where this is going to go. Let's maybe just make sure that our safety teams, our alignment work, has time to catch up." And I think basically we need that, right? We need companies in the space to have great safety teams. We need open debate. The AI world is pretty polarized right now, right? You've got kind of techno-utopians versus AI doomers and it's... The debate's good, but there's a lot of pretty blinkered people out there, I think. I think we need more of an open mindset about what some of the risks might be and what some of the benefits might be.

Richard Kramer:

Yeah, and one of the things we've talked about before on this podcast, we've talked about what goes into the calculations of inflation from the Office of National Statistics. And when you look at what's captured on a lot of these algorithms that are underpinning AI models is statistics. But we don't have an office of national algorithms or AI, so we don't have a way to unpack the sources in a lot of those AI models and ascertain their validity. Do you think it's just a time and comfort factor for the benign uses that we've already seen for those of us who use AI can have a real positive impact?

Ed Newton Rex:

I think time and comfort are both very important here. I think that it's a sort of short time window over which really, at least in the public consciousness, all of this has happened that I think-

Richard Kramer:

Exploding.

Ed Newton Rex:

Yeah, exactly. And I think that's spooks people. And honestly I think people are right to be spooked in that respect. Whenever any technology comes on very rapidly, we should bear in mind that probably we haven't had time yet to put the systems in place to mitigate against risks there. And we need to... And this is why I signed that letter and other people did as well. Sometimes we just need time to take a step back. Now I believe that the benefits of these systems are huge and are going to be huge. And we're already seeing many of them, but we do just need to... Any new technology like this that's as powerful as this changes the world and it will change the world in some good ways and some bad ways. And we need to make sure that we're steering things towards the good.

Will Page:

Ed. I want to bring it back towards the economics for a second. Something that yourself and Richard were talking to is alluding to a term that we hear called productivity gains. And it's a lopsided expression because it's purely gains. We know there's costs and benefits, but if you think about productivity for a second, that's when you can get more out of the same or the same out of less or more out of less. That's how a classic CFO would look at a productivity gain. And I just wanted to turn the productivity question back to you. Richard's example of bank clerks.

The best expression I ever got for what capitalism actually means came from Alan Blinder, Federal Reserve governor in the United States. And he said to me, "Capitalism is when you employ somebody else to cut your grass because you can do something more productive with your time." The point being, you might enjoy gardening, you might really enjoy cutting your grass, but when you employ a gardener to cut your grass it's because you can justify that by doing something more productive with your time. Talk me through the sort of productivity gains we're going to see as we approach AGI and this whole big picture of AI where we're going like, is it all upside productivity gains or is there going to be winners and losers in the economic fallout as well?

Ed Newton Rex:

Yeah, I think we're already seeing huge productivity gains in a number of fields. I think a really good example here is on the image side, which is where kind of Stable Diffusion came in last year. It is now very easy and quick to come up with things like storyboards just to simply... Or first stage character designs for a video game or kind of ideas piece of marketing material or artwork. It is now unbelievably quick to come up with these things and that is hours and across industries that is going to be hundreds of thousands, millions of hours of time taken off the work people have to do. And again, I come back to my own work which is with someone in product and technology tends to be very focused around things like documents and these sorts of things.

And so we look to large language models and we find that these things are already shaving hours and hours off what you have to do. And that is freeing up time to think more about your product roadmap or whatever it is or all these other sort of important but boring things that we have to do in this world.

So the productivity gains are clearly real. You're absolutely right, I think, to ask will there be winners and losers? And of course there will be. I think with any new technology there are winners and losers. I tend to think that this is going to change a number of industries in a huge number of ways. And the people who, as with any technological shift, the people who come out on top, it won't necessarily be a kind of certain type of person or certain type of job. It'll be, I think, the people who spot this early and figure out how they can make the most of it. Like right now. There's a huge opportunity for people I think who are willing to use this tech and learn how to use it. Because it's still not the easiest thing in the world to use, you've got to get used to it. So I think, like anything else, you'll see people who are really driven to make use of this will find ways to make use of it.

Will Page:

Let me just put you on the spot a little. That was a very eloquent analysis of how this thing might play out. Let me just put a scenario in front of you, you tell me yay or nay, okay? So let's just say the UK workforce today, roughly speaking, 33 million people are working in this country. Now, I'm just take a guess. Let's say, hypothetically, 3 million of them are working in call centers. Do you think by 2025, we'll have 3 million unemployed?

Ed Newton Rex:

I very much doubt it as a direct result from that. Will jobs have changed? Almost certainly. Speech synthesis plus LLMs is already at the stage or actually probably a lot of that work could be done by AI systems. I would [inaudible 00:17:07] but that doesn't mean I think it would be a relatively kind of naive future gazer who would think that means 3 million people are going to be unemployed, right? These things, in sort of driving productivity and in opening up new industries almost certainly, tend to historically create more jobs than they destroy. Now does that mean they definitely will? Absolutely not, and I don't think future gazing is particularly helpful here either, because who on earth knows honestly? But I'd be very surprised if the net result of LLMs plus speech synthesis leads to 3 million unemployed.

Richard?

Richard Kramer:

Yeah, I want to go back to one of the notions that you sort of touched on briefly, but makes me think of Walter Benjamin and Art in the Age of Mechanical Reproduction, which obviously as a digital composer, you're benefiting from. And it used to be that a hundred years ago, all performances were live performances. We didn't have this crazy thing called recorded music if we could imagine those times.

But what's going to happen to the underlying creativity? I know I can ask Stability or Stable Diffusion, "Give me an image of a frog wearing a pink tutu dunking a basketball," and presto. I can get that image. But what about the underlying work of refining and crafting and creating that nutty image, making it just right, getting the angles right? How do you ensure that the people who are doing that underlying creative work that have the integrity and the vision behind it aren't supplanted by the technology? And I think that's what scares so many people in the creative industries about AI.

Ed Newton Rex:

For me, I think it's a tale of two types of creators really. I think the first is sort of artists and I don't think things are going to change very much for this sort of be that someone, be that your kind of small shop jewelry designer or your Ed Sheeran. Whoever it is, your kind of not only very talented person, but person who, sort of creator, who people appreciate because of who they are as well, because of their story, because that they have come to love that person. And that kind of creator I think will... Well, if there's any change, I think it'll only be positive, right? Because if anything in an age of mass reproduction of on an unprecedented scale and automation, I think that we're likely to come to value this kind of creator even more. Being human will really, I think, set you apart, right?

So I think that we do, in that kind of instance, care much more about the person behind the creativity than the creative output itself. For every Ed Sheeran, there are a hundred songwriters who are probably just as good as him but who aren't as famous. There is a lot of talent out there and so it's not all about the music or the jewelry or whatever it is. So I think these kinds of people will not see their creativity changed or threatened in any way really.

I do think that for the long tail, the people working more functionally, people working on commission to fulfill a specific spec, it's harder to tell what happens there, right? Will the work shift more to working with generative AI systems? Maybe it will. Maybe these kinds of people will have new workflows, maybe they'll... Absolutely. Will some of these people end up in totally different careers? Maybe, I honestly have no idea. And I think that kind of trying to predict this stuff is very difficult. But I think that ends up actually being almost a user experience question. If you look at currently there's all this talk of sort of prompt engineering, which is a new skill that people are espousing, but really-

Will Page:

Promptography.

Ed Newton Rex:

Promptography, yeah. But to me, this is basically... This kind of skilled prompting, which I wouldn't call engineering. I think that engineering, to me, is something that is picked up over is highly skilled and is picked up over years and years of work. And prompt engineering can't be right and this is a relatively unpopular view in this industry, but it can't be in that it's only even existed for a year. Right? So this is obviously not something that takes the same level of skill as many of the under other engineering disciplines, but it's only required right now because these systems really aren't fully mature yet. And so I think you get down to a UX question where ultimately we haven't really liked this tech sprung up so fast that we haven't really figured out how people are going to work with this in future. We don't know what the interfaces will be.

We don't know whether they're going to have to prompt it or actually whether... This is much more likely, I think. Essentially these systems will learn to basically answer pretty well. You won't have to say anything special like, "I want..." At the moment, you sometimes have to put an entire paragraph to describe the exact kind of picture you want. But that basically just means the system hasn't really understood what you want, right? It's a UX problem. I think it'll be solved.

But that kind of demonstrates to me how far we are still from really understanding what even interacting with these systems is going to look like in even five years time. So I think it is super hard to predict where people's efforts will be best placed and where jobs will be created and where they'll be destroyed.

Will Page:

Taking it to the break here, you've given us a great foundation for Richard to go down a rabbit hole in part two. But before we hit the break, just a very quick you have any question. If you went to the Sony Photography of the Year competition this year and you pay 20 pounds to get in and see these works, and then you learn after the event the winner had used AI to create that photograph, would you have asked for your money back or would you have bought his book?

Ed Newton Rex:

Yeah, I mean I think I would've asked for my money back. I love that they admitted it. I think one of the key principles of creative AI that I think is really important is that we should be transparent about when we've used AI just as much about artistic integrity as anything else. But yeah, that photo shouldn't have been entered into that competition, right? Prompting can be tricky and prompting might be a skill in itself, but it's not. Let's be clear, it's not a skill on par with the actual creative endeavor of art, of photography, of whatever it is. Learning these skills often takes a lifetime.

And so yeah, think that we should get clear these kinds of competitions are going to continue to exist and they should be competitions for human creativity and we should have other competitions for AI creativity. Absolutely. There should be, and I think probably already are, prompting competitions. Absolutely, let's-

Will Page:

As we discussed before, we can have a Eurovision Song Contest and we can have a Eurovision AI song contest and where indeed,

Richard Kramer:

Share on our favorite computers.

Ed Newton Rex:

Indeed. And we have. Yeah. A few years ago we set up the AI song contest. And the AI song contest is great and it gets bigger every year. And the point is to collaborate with an AI system. It's not just kind of giving a prize to the best fully AI-composed song. Idea is let's collaborate with AI systems, let's do interesting stuff. Let's make interesting music that used AI in novel ways but-

Will Page:

Be upfront about what you do.

Ed Newton Rex:

But that should always be a separate competition right from your vision.

Will Page:

And my got it. Thank you so much. Ed, you've really made me feel a bit more assured about knowing that we don't know where we're going with this. That's great. We'll take a break now. We'll come back in part two and go down a rabbit hole on this incredibly fast-paced topic. Thank you so much.

Richard Kramer:

Welcome back to the second part with Ed Newton Rex talking about AI. And we want to go down the rabbit hole and understand the technology a little bit better. Everyone on the planet who has any sentiment being or artificial sentiments, whatever you call it, has heard about open AI and ChatGPT. I think just after that Stable Diffusion dolly and other image generators have captured a lot of attention.

And one of our previous podcasts, we talked about our favorite situationist guru, Guy Debord, and the Society of the Spectacle, and we're clearly in an image-addicted age.

Can you talk us through a little bit what Stability and Stable Diffusion does with images? And what have you seen people using it for? How does it foster and foment human creativity as opposed to narrow it down to those stock images we all got so bored of seeing in corporate presentations?

Ed Newton Rex:

Yeah, well, Stability came out with Stable Diffusion, I think it was last August. And it was I think one of maybe sort of three kind of huge high impact AI image generation systems to come out last summer that really kind of kicked off, I think, this wave of people being fascinated with AI. And put simply, what you can do with these systems, with Stable Diffusion, is you can type in any sentence, you let any kind of description of an image you want to see, and Stable Diffusion will incredibly rapidly and usually very accurately create that for you. So you can sort of say... You can ask for any style you want, you can ask for any objects you want. In some of our recent releases, you can actually ask for text as well. So you can say, "I'd like a neon hotel sign with this hotel name or a kind of pub sign with this name," or whatever it is.

And it will just create these images for you. It really is just magic. Seeing these systems in action for the first time, as most of your listeners probably have done, is an incredible moment. Stability is what Stability works on other media as well. So we've just released Stable Animation, so you can now create animations as well. We released Stable LM, which is our large language model along the lines of things like ChatGPT, I run the audio team where we're working on music generation, so we're sort of very much multimedia. Yeah, but look in image sides, it really has been used for just an astonishing array of things. The thing about these kinds of systems is that they are so foundational that it's almost impossible to predict how they'll be used.

I've used image generation for album art work and for illustrating blog posts. These things are quite predictable and they're the first things you kind of think of. People have used Stable Diffusion to create video game artwork to create textures in video games. Bands have created music videos using these animation systems. I think that's something we're going to see much more of.

I think some of my favorite use cases have been just entirely out there and you never would've thought of. So there's one, there's a group who created a bunch of synthetic images to add into a data set to train an image recognition system for things like reading X-rays, right? So you can actually generate images, add these to a data set and train better image recognition systems, which is kind of like crazy iteration there of AI kind of improving itself. Or and then I think possibly my favorite was a paper that was released a few months ago where people had managed to essentially scan brainwaves and do a pretty decent job of creating images that showed what people were thinking. When they saw an image, they could scan the brainwaves and then recreate that image. So you had these-

Richard Kramer:

I presume that was 80 or 90% porn, basically.

Ed Newton Rex:

I think maybe they filtered that out of the paper, but I'm sure in the original research, yeah.

Will Page:

I will study. I was not part of the study, just to be clear.

Richard Kramer:

And before we use Stable Diffusion to generate an image of Will Page in an England rugby jersey, which would be sort of qualify in the deepfakes category, which we'll talk about in a moment, it does make me think about the way... A word you used before is thought of in the English language, which is engineering.

Now in English, engineering has an engine as an industrial age metaphor. If you go to France, it's ingénieur. There's invention. There's the idea that the engineering task is to add your ingenuity to extend something. And I think what you're giving us a good example of is that cusp between providing people tools but not describing or proscribing what they're going to do with them. And I'm sure we'll talk in a moment about some of the malevolent uses of these, but I guess you've just given us some of the more positive or more sort of heartwarming ideas of how these tools can be used.

Ed Newton Rex:

Yeah, I think so. Look, one of the things that really sets Stability apart in this space is the fact that we're an open-source first company. And the fundamental belief-

Will Page:

That's important.

Ed Newton Rex:

... there behind being an open source company is essentially that we as a group of however many people are unlikely to be able to take these things to their full potential on our own. And sure enough, what we see, once Stable Diffusion was launched almost immediately there was this wave of iterations on it and improvements where you can start to... People coming up with much better ways of controlling what's in the image that's being generated. People coming up with much better ways of generating images full stop, all built on the top of this system. There were so many, with access to the internet, with increased access to compute and to being able to train models yourself as an individual, we are at the stage where hackers around the world, individual tinkerers, sitting at their laptop can make mind-blowing advances in one night. And that's kind of the benefit you get from putting this stuff out there.

Will Page:

Richard, just as you were talking about using an AI-generated image of me in an English [inaudible 00:30:44] top, I was using ChatGTP to ask how to pursue a defamation case, so you've got AI down on each other.

Richard Kramer:

There you go.

Will Page:

I love it when Richard goes down a rabbit hole in part two because you always hear something that inspires a line of thought that you want to pursue further. And you mentioned animation. And what that made me think of is we're talking in 2023. I remember noticing toddlers, kids 18 months to two years old, learn how to skip adverts on YouTube 10 years ago. In 10 years time they're going to be graduating from university. If you're seeing Stability working with animation in an open source way, what do you think that's going to do to children growing up with this technology and what they can do with their creative selves? I'm just thinking about kids are using iPhones to create movies and edit movies right now. That's unimaginable 10 years ago, but it's happening now. Talk me through this, the animation use case, particularly with the view to children growing up with this technology as opposed to adults like ourselves scratching their heads saying, "What the hell do we do with this technology?"

Ed Newton Rex:

Yeah, I think we're in the early stages of animation. But if you look at where we were with image generation even a year ago, you get an idea of how quickly all these fields are likely to advance. Right now, you can essentially do things like upload a video of yourself and have it change its style. You can change it so that the video of you playing football is changed into a Play-Doh man playing football, right? Which is fun, which is creative, and there are already decent apps springing up that are fun and I think will be fun for children to use and fun for anyone to use. But I think they're kind of low hanging fruit compared to where we're going to get, where you can absolutely imagine in the not too distant future being able to frankly create a sort of Hollywood standard, or in it's all intents and purposes, Hollywood standard film very easily perhaps on your phone, perhaps as an individual, perhaps by entering purely text, perhaps by giving the system some dialogue.

Again, so much of this, I want to caveat this with it's so hard to predict, and I think... I'm always careful about my predictions here, because 10 years ago when I was talking about generative AI, I very confidently used to say, "I thought music was going to be the first art form to be mastered by generative AI," because I considered it the most mathematical. And it made a lot of sense to me that it would be mastered first. And what's happened is that images honestly have been the art form that have been mastered first. And I never-

Will Page:

Interesting.

Ed Newton Rex:

... would've predicted that. Music's not far behind, let's be clear, but the timing was wrong. And I think it just shows that this stuff is incredibly hard to predict, but it is... I cannot imagine a world in 10 years time where you were not able to create feature length films on your phone, right? I just can't imagine a world where that's not possible.

Richard Kramer:

But again, when you think about the disruption and having had raised twin sons and therefore watched every single Marvel movie to the end, because there's always that last bit at the end, and watch the scroll through of literally thousands of names of digital FX people working on that film, what will happen to those folks? Because it was in those original Avengers movies that I took my sons and their friends to see on their 10th birthday, we would watch that end scroll by and you knew that there were dozens and dozens of companies working on small aspects of those digital FX and animation. And will that all highly technical work that goes from storyboarding through the final full-length feature get replaced by AI. That seems a stretch, but at the same time, that's going to be a huge dislocation for the hundreds of thousands of people that are employed in that field.

Ed Newton Rex:

Yeah, I mean possibly, I think definitely some of that, it'd be very surprising if some of that didn't successfully get done by AI and there'll probably be cost savings there. And I'm sure that some of these effects will be as sort of is already the case, right? These companies are all using computer systems already, right? This is not happening in a vacuum of technology. But yeah, I'd be very surprised if things don't change a bit. But at the same time, once it's easier to make a film from scratch, will some of these people end up actually making their own films? It's new things, AI will just make all of these things possible. Will it let those people go and actually be filmmakers themselves? Is that what some of them want to do? I have no idea, but I think it's at least possible, right?

And so do you move to a world... You'll still probably have your Marvel movies, right? Maybe unfortunately. I actually have quite a soft spot for them, I'm not going to lie. You're still going to have these kind of blockbusters, but will you move to a world where actually more people are making films that are seen by their close circle? Is it easier to actually get into filmmaking and to rise up through the ranks? Does it just become easier to make content in general and does that give more people a chance? Something that I've been thinking about a lot for the last 10 years is the democratizing potential of this technology. And it's real, right? You are able to, for instance, all of these quote, unquote, "AI songs" coming out where you've got [inaudible 00:35:56] is being replicated.

Will Page:

You've got Liam Gallica sounding better.

Ed Newton Rex:

[inaudible 00:36:00] suddenly being able to know that there are potentially license issues there, sure. But suddenly being able to use an artist's voice on a track and just sitting in your bedroom, it's just a sort of the next step change in terms of what you can do from your bedroom without having to be someone who lives in LA and has the right contacts. And I think that this... I find that element of AI incredibly exciting, the ability to let more people get involved in the creative process.

Now there's a balance, right? You don't want to replace creativity and that would be a disaster, but you do want more people to be able to get into the game. And I think walking that tight rope is what I think that most people I know who are working in this space are trying to do.

Will Page:

So bringing that point home, Eddie, about what it means for children who are growing up this technology, I once was told that children won't look at screens they can't touch, now they don't look at screens they can't interact with, perhaps in 10 years time, they won't look at screens that they didn't create the content on in the first place. It just goes full circle. Maybe before we go to smoke signals, I think this is a great way to wrap up where the conversations got to.

Your ability to make us have cool heads when we approach this topic is admirable. And I hope more and more people get to listen to this podcast and follow your work. But if we go back to the GitHub co-pilot story, in a matter of months, let's say four or five months, we went from Armageddon, that death is now, coders are redundant, copyright infringement lawsuits to, "Actually, I can code four times more productively thanks to this technology than ever before," so from threat to opportunity and the journey was literally less than six months. We went from staring into the abyss to, "This is pretty cool." Do you think we're going to see that story be mirrored across society of, "Ah! I'm looking into the abyss! Well, actually I can use this to make tomorrow better than today"?

Ed Newton Rex:

I think we will almost certainly see that mirrored in many ways. I think music's a good example where the AI systems that we are building already... They're capable of doing just really interesting stuff, right? One of the things that we like to use our systems for internally and with testers is kind of creating samples. You can use AI to create new sounds that you would never have created otherwise. And I think this is really empowering and it helps your creative process.

Another thing I've always really loved around using creative AI is using it as an inspirational tool. When I ran my startup, we had a hack day where we used our generative AI system to just build some random product. And I spent the day building an iPhone app where you would just swipe. It was kind of like a Tinder for musical ideas.

You would just swipe left and it would kind of generate chord sequences or little melodies. And that was meant to solve a problem for me, which is where whenever I'm writing a song or a piece of music, unfortunately, maybe I'm just not good enough at it. But I sit down at the piano and my hands just fall into the same positions and I end up playing the same kind of thing, right? And so I think that all these kinds of things will be possible with these sorts of systems and there are huge benefits to be had and we are yet to see how it all plays out. But I think that people are going to do... When we release a music system, when video generation becomes as big as image generation is, these are going to be used for absolutely fascinating things and it's going to empower a lot of people, I think.

Richard Kramer:

Do you think there's going to be an agreed upon digital watermarking where we know something's not a deepfake because it's very easy to imagine, again, my Will Wage example that the ability of bad actors to come up with plausible deepfakes is endless.

Ed Newton Rex:

Yeah.

Richard Kramer:

Do you think we'll get an agreed upon global standard for digitally watermarking things to know, "Yeah, that was the real person in that image," and we don't have to worry about that our likeness can be used in a way that we wouldn't want?

Ed Newton Rex:

Yeah, I think we might well. We at Stability, we just joined the Content Authenticity Initiative that's being driven by Adobe. Universal Music also joined at the same time, I think I saw. And this is trying to do exactly that. It's saying, "Let's watermark AI-generated content," like this is... There's no reason not to, ultimately, unless you basically have sort of nefarious intentions.

Richard Kramer:

Yeah.

Ed Newton Rex:

So yeah, I think that we could well gather... I think more and more companies are thinking about this. I absolutely think that it is something we should be doing. I can't think of really a good reason to want to hoodwink people. So yeah, I think that's definitely possible. People are already working on this and some of the biggest companies in the world are already signing up to these systems.

Will Page:

And just to reiterate the point that grass is not necessarily green on the other side, there's a lot of problems with the current model of who authored that work, who is the mother of that child that we know about. Metadata isn't perfect. AI might make it a little bit more perfect. Richard, time to get smoking.

Richard Kramer:

Yeah, we typically ask our guests for a couple of smoke signals, what we call the sort of uh-oh moments when you hear someone pontificating about a subject you know well, but you just know it's completely wrong. So now what sort of terminology or metaphors just make you cringe? And if you had to warn our listeners about a couple of things in the discourse about AI, what would you say, well, when you hear that, you'd just want to shake your head in disbelief?

Ed Newton Rex:

Yeah, there is one at the moment that really gets my back up, which is people talking about AI songs. So I'm sure you've probably listened to every single kind of AI Drake song and AI Oasis song and everything that's kind of come out in the last month or so. I certainly have. Some of them are really good. But there is a lot of confusion even from people in the music industry and sometimes in the AI industry on what these things actually are, I think because of the kind of just how vague the term AI song is. I think there's a lot of people out there who have heard these songs and think that AI is at the stage where it can generate a Drake song from scratch, when in fact what's happening is you've got a human writing a new song, a human recording a vocal, and then using an AI system to essentially transform that vocal into the sound of Drake singing it, right?

And so there's actually nothing kind of creative happening in that process from the AI's point of view. This is essentially just voice to voice. And I think it's funny. It's led to a lot of computer... It's led to a lot more fear than I think there should be. People are just suddenly incredibly worried that AI music is going to be generated and is going to storm streaming services and the whole music industry is going to fall apart and it's just not the case.

Richard Kramer:

Is there another one that springs to mind in all of the FT Magazine, AI is going to wipe out humanity discussion that you've seen or the Geoffrey Hinton resigning from Google and saying, "We're a lot closer to AGI than we think and sentience," anything in there that you just feel is overwrought or makes you shake your head in disbelief?

Ed Newton Rex:

Yeah, I think it's sort of on the other side of it, to be honest. I see a lot of sort of half of the people in the AI world, the kind of techno-utopians, often point to the fact that technological advances have essentially almost always been a good thing in the past, have always created jobs, have always led to growth and that sort of thing. And they point to this to totally dispel any notion that there should be any risk around AI, around artificial intelligence.

Richard Kramer:

Or regulations. Yeah.

Ed Newton Rex:

Yeah. And honestly, I don't think that we can just say, "Well, things have always gone well before, so they'll go, well this time." Because there will inevitably, it'll be some technology, it may not be AI, but some technology will come along at some point and it will not follow that pattern, right? And the people saying, "It's always been fine before," will be proved wrong. And so I'm really of the belief that we shouldn't pretend that there are no risks in AI. We should make sure that we are really open about the risks. And yeah, I do get pretty frustrated when I hear people in the AI world just kind of ignoring the risks, because I think it's... If the people in the AI industry who understand the tech aren't thinking about the risks, then who's going to, honestly?

Richard Kramer:

Wow. That's a great note to close on.

Will Page:

Absolutely.

It's been... I knew, going into this podcast, I was going to feel a little bit more assured about where we're going with AI than before we started. And you've certainly delivered that. You give me balance, you give me hope. The one thing that's at the end of the Jessica Powell podcast I kept on thinking about bespoke suits versus on-the-peg suits and how this could help, not when you say automate, let's just call bespoke everything. The trade-off between buying an off-the-peg suit, it's cheaper, faster, and a bespoke suit takes longer, cost more. How AI is kind of narrowing that gap...

Richard Kramer:

But the bespoke suit fits better.

Will Page:

For most people, but for six feet four running athletes like yourself, you're going along usual shape on yourself to be proud of too. But listen. I think here, the lingering thought in my head is Jevons paradox.

I don't quite know how to square it, but I'll just bounce it off you, Ed, which is Jevons famously put it as when we became more coal-efficient, we didn't use less coal, we ended up using more of it. So if fuel becomes more efficient for transport, we don't use less fuel. We live further away from busy cities, travel longer in and use more fuel. When refrigeration became more efficient, we didn't buy smaller refrigerators, we bought bigger ones. And I'm just thinking through the unintended consequences of the perverse effects of all of this. When we allow for more automation, do we all become more automated or does a pendulum swing back the other way? My head's juggling with that and I'm sure in future episodes, we'll get closer to solving it. But Ed, you've inspired a lot of thinking in the past 40 minutes and I want to thank you so much for giving up your precious time. I wish you all the best in that rollercoaster ride over at Stability. Stability by name, but not by nature. Thank you, Ed Newton Rex, for coming on Bubble Trouble.

Ed Newton Rex:

Been awesome to be here. Thank you.

Will Page:

If you are new to Bubble Trouble, we hope you'll follow the show wherever you listen to podcasts. And please share it on your socials. Bubble Trouble is produced by Eric Nuzum, Jesse Baker and Julia Nat at Magnificent Noise. You can learn more@bubbletroublepodcast.com. Until next time, from my co-host Richard Kramer, I'm Will Page.