Transcript: Can AI be truly creative?

This is an audio transcript of the Working It podcast episode: ‘Can AI really do creative work?’

Isabel Berwick
Hello, and welcome to Working It from the Financial Times. I’m Isabel Berwick. 

[MUSIC PLAYING]

This week, I’m looking at one of the most contentious questions around AI: how it will affect creative work? As AI’s capabilities have grown, so have concerns over its impact on creative jobs. But there’s plenty of reason for optimism about how AI might help people in those jobs be more efficient and produce more imaginative work faster.

My first guest is someone who uses AI for exactly those reasons. Dan Sherratt is VP of Creative and Innovation at Poppins, a digital creative agency. I spoke to Dan last week and a lot of what he said surprised me. He explained how generative AI models like Midjourney could replace some of the easier, low-value tasks that his company does and how that could make creative work like his faster and more profitable. He also told me how work made without AI could become unusual and artisanal, like a bespoke suit in a world where AI-assisted work is more off the peg. I started my conversation with Dan by asking him to tell me a little more about his work.

Dan Sherratt
My job is really to help our clients use technology — the technology of tomorrow, technology of the future — understand it, apply it in the right way. What’s the problem we’re solving with it? Is it useful? Does it actually result in happy customers? Happy clients?

Isabel Berwick
So talking about the technology of tomorrow, could you talk us through a process you do in the course of your work and how AI has changed the way you do it?

Dan Sherratt
The biggest change we’ve seen is using it with pitching. Pitching obviously is unpaid. When you try and ask a client to imagine something completely wild, it’s a real struggle.

For example, if we were asking a client to imagine an augmented reality experience in which a giant 50ft piñata is hanging from London Bridge, and a client could tap on it and it would release sweets, and each of those tweets had a code on it that meant that your customers could discount their products. Asking them to imagine that is previously it was just impossible. And when I was first starting out, it would be a case of OK, go on Google image search, find a picture of a piñata, then find a picture of Tower Bridge, then cut out the piñata, then put that over the top of Tower Bridge. And now I can ask gen AI to do that. And I can have my team do that. And I think that that’s where it really helps.

It isn’t a case of replacing anything that we’re doing, but it speeds up the process and especially in unpaid pitch and proposal work, it levels the playing field because we’re not a big agency. I mean, big agencies have the resources to pitch and throw resources at it. So it really has levelled the playing field for us.

Isabel Berwick
That’s such an interesting point. I hadn’t thought of that. Because that’s a huge time- and money-saving, isn’t it? 

Dan Sherratt
Yeah. And producing decks and that sort of thing is seen as like the sort of grunt work in agencies, right? And obviously when I first started out, a really good example is you need a picture of what teamwork looked like. And in 10 out of 10 stock photography websites in the ’90s, you would go “teamwork” and it would be a really awkward stock photo like photograph of teams high-fiving or a big handshake over a desk.

And now you can just be far more creative with that. You can ask gen AI to imagine a gritty black-and-white high-contrast photograph of a Hackney football team in the rain, and that’s teamwork. And that’s so much better than this awkward stock photography high-five.

Isabel Berwick
Yeah, I think I’ve used all of those (Dan laughs). So can you tell us which AI tools you use and what are they good at and what they’re less good at?

Dan Sherratt
So primarily we use Midjourney. It’s the most artistic of the generative AI image production tools. It definitely understands creative prompts far better. I think a lot of the models that are trained with OpenAI, and I mean with the way that Apple are using it as well, it feels a lot more literal and what you get is almost clumsy. Whereas it feels like Midjourney, you are talking to an artist in a way. That sounds weird, but it’s a lot more understanding of creative prompts.

And we do use it almost exclusively in the pitch and proposal process. We wouldn’t seek to use it for the actual execution work apart from things like background imagery and things like that, where we’re not, where we would never have hired an independent artist or illustrator, anyway. But again, that’s not to say that that won’t change in the future.

So at the moment, it’s really, really good for conceptual work, and it’s not quite as good as you’d expect it to be yet at producing finished files. But it’s only a matter of time.

Isabel Berwick
Can you realistically see yourself ever not using AI? You know, is that genie ever going back in the bottle?

Dan Sherratt
Realistically, we can’t not use it because everyone else is using it. The big agencies are using it. It’s out there. That horse is well and truly bolted. And so not to use it, I think you either have to commit to openly trying to sell your work as not having AI and using that as a plus, and being like this is owned, made only by humans. No AI involved in this, like a label or sticker.

Isabel Berwick
That’ll be in five years’ time, that would be like an artisanal.

Dan Sherratt
I mean, look at the way that the . . . we could be looking at a future where that’s the case, where higher value is applied to things that were made with harder work. Because that happens in fashion, right? You buy a custom pair of shoes from an Italian shoemaker who’s got a family business of hundreds of years old versus a mass-produced sweatshop sneaker. So creatively, that could be where we’re heading.

Isabel Berwick
I mean, that kind of comes on to my next question and partly, partially answers it. Are there things that AI won’t be able to do? Are there some human things that will always be protected?

Dan Sherratt
I mean, right now it’s not great at innovation or playful innovation, really. It’s not brilliant at copywriting. And like that human wit, it struggles with snappy one-liners. So that’s not to say it won’t figure that out, right? I think that we can only utilise it to the best of its current ability.

When we were playing with Dall-E, which is another one of the tools, when it was first launched, it was so clumsy and awkward and it couldn’t figure out hands, but they acknowledged that and they fixed that. It was a matter of months. And so it would be foolish of me to say that there are things that it can’t do and never will be able to do.

Isabel Berwick
And has it changed the way you think about creativity?

Dan Sherratt
No. I always valued . . . going back to my previous example, if my task was to interpret a better example of teamwork that wasn’t a high-fiving group of individuals versus like, OK, what does teamwork actually mean? And that would be my job as a creative to find that, be that through stock photography or illustration. That was still a creative job. That was a job that not everybody could do.

And so that level of creativity still applies to what is essentially what is called prompt engineering. So you still have to have the right commands and understanding. And I don’t think that will ever change. Because if to any of these models you just used the word, “show me an image of teamwork”, you are going to get the same level of high-fiving people, big handshakes. But there will always be a creative level of engineering of what you’re asking AI to do.

Isabel Berwick
Yeah. And I’m interested to think about how, you know, you’ve given an example where AI has enhanced the value of your work because, you know, you’re not spending time on things you’re not being paid for. How could it depress? Are you already seeing AI depressing some agency or creative work?

Dan Sherratt
There will always be people, agencies, creatives that use these tools to devalue work. I would hope that there would always be a value for quality and craftsmanship, and I don’t think that will always come from purely AI-generated work. I think handcrafted elements are going to see much more value applied to them. And I think there is a level of craft that we apply to our work, be that a logo design, some smiling mind interaction design, or a really impressive augmented reality brand launch.

Those are ideas that come from us, and it’s essentially still the people that you’re investing in. We would use AI to moodboard and to communicate ideas, but handcrafting things, getting the right smoothing on a on a logo or choosing the right letterforms for a mark or the way that a loading animation might work when onboarding someone on to a product. Those things I don’t think AI can do.

Isabel Berwick
So, it’s interesting. So the hierarchies of value already exist.

Dan Sherratt
Yeah.

Isabel Berwick
And it’s just people who are outside. So AI’s been sort of lumped in as this democratising or devaluing technology, but actually those things were already there in a lot of industries.

Dan Sherratt
Yeah, definitely. These values exist in today’s society and we just hadn’t yet applied them to digital work.

Isabel Berwick
But do you think there are jobs that could be replaced in the creative industries?

Dan Sherratt
Certainly, yes. (Laughter) I think, again, it would be foolish of me or idealistic of me to suggest otherwise. I think that a struggling start-up — say a local bakery or coffee shop that needs an identity, they need a website, they need some marketing materials and they need a social media account — certainly all of these things can be done by AI. It’s almost like there’s just a level of project value that probably just gets removed, and everyone has to work a little bit harder on the mid-level and top-level projects.

And maybe in the future, as AI develops, that bar gets raised again and the mid-level projects become completed by AI. And that’s when we get to the really artisanal. We’ve hired this individual who is a master craftsman, and when that company launches, they can say our logo, our brand, our identity was designed by this famed craftsman, which again exists in today’s world. I mean, the fashion industry is booming. At the partnerships, you look at Pharrell at Louis Vuitton; like, it’s still an individual that is a creative director on something, and they have invested in him to sell products. I don’t think it’s any different for us necessarily and I imagine behind the doors, LVMH, they’re using AI for ideation the same way that we are.

Isabel Berwick
Yeah. You’re obviously immersed in it. But when you go out to clients, what’s the general kind of awareness in the world of AI? What’s a client’s expectations? What do they understand?

Dan Sherratt
It’s a mix. Sometimes they will ask us, like, how are you using it? How are you using other products? We still use it as a magic trick, really. Because we love to impress people with pageantry and storytelling and magic. And being able to reveal something really beautiful and impressive and then add on “this was created by AI”, I believe to be an impressive feat. It shouldn’t be.

And we’re lucky again to be living through it, because in the future you’ll probably get eyeroll. “Of course it was”. At the moment it’s, “Wow, that’s impressive”. And I think that people do value speed, because ultimately, speed equals value. And we as an agency use it in the same way, like we can level up our work, we can impress our clients at a smaller cost.

[MUSIC PLAYING]

Isabel Berwick
Thank you so much. This has been amazing and prompted a lot of thoughts on my part.

Dan Sherratt
No, thanks. It’s been my pleasure.

Isabel Berwick
Dan’s impressively relaxed for someone whose industry is changing so quickly, but speaking to him, I could understand why. In design, as in many other areas, generative AI is being used to speed up low-value grunt work, allowing creative people to focus their energy on more complicated and more rewarding tasks.

Dan talked about AI tools in that interview, but AI could become rather more than just a tool. Some people believe it can be or already is genuinely creative in its own right. Here’s one of them.

Marcus du Sautoy
My name’s Marcus du Sautoy. I’m a professor of mathematics and the Simonyi Professor for the Public Understanding of Science at the University of Oxford. And I’m also the author of The Creativity Code.

Isabel Berwick
The Creativity Code is all about what creativity actually is and whether machines could learn to replicate it. One argument Marcus makes is that collaborating with AI could actually help people work in a less machine-like way. I asked him to explain that.

Marcus du Sautoy
I think that creative people very often get stuck in a rut. They find something that they . . . it’s successful for them and then they just keep on repeating that behaviour. And so I see people actually becoming more mechanistic in their creativity because they just repeat things that have worked in the past.

And the role I’m seeing of AI in creativity in humans is to take the data that a creative might be making — maybe it’s a musician — and then by analysing that data, it can often offer up new ways of using that data that the creative person hasn’t thought of before.

And so I’ve seen this, for example, in a jazz improviser, Bernard Lubat, who trained up an AI in Paris. And when he improvised with the AI in a concert, he said, I recognise everything; in some sense that is my sound, but it’s doing things I’ve never ever thought of doing before. And so in that case, here’s somebody who got stuck in a particular way of playing and the AI sort of liberated him to show him other things that he could do with his sound world.

And in my own collaborations with AI, what I found is I don’t particularly like what the AI is offering, and I will reject it, but it will make me think, why don’t I like this? And actually there is something in there that I think I can take further that I’ve never even thought of doing before.

Isabel Berwick
Is this in mathematics, or in all sorts of things that you do?

Marcus du Sautoy
Well, in all sorts of things, actually. I do a lot of work with creative artists, which is . . . really informed the book I write. For example, I do a lot of theatre work, and I had a problem with a piece that I’m making at the moment and I couldn’t see quite how to solve getting something off the floor visible to the audience. And I asked ChatGPT for some ideas of how to do this and it came back with some really innovative ideas that presumably other people had done in the past. But it was really beautiful at synthesising these ideas. None of them were quite what I wanted, but it gave me the ideas for how to solve the problem.

Isabel Berwick
So is the future of creativity in the humanity of the prompt, or will we have solo human creativity in future without AI assistance?

Marcus du Sautoy
I think we will always have solo human creativity, but I think there’s a role to play for those who find this an interesting new collaborator. There are really new possibilities opening up here for the new collaborations and perhaps a recognition that there were many people involved in all creative work.

You know, there’s a real recognition, I think, that we have to have that, just as AI is learning from the data that we’re giving it — the novels that we’ve written, the paintings that we’ve painted, the music we’ve composed — actually, every human creative has gone through a similar process. Picasso learned how to paint in the style of the masters of the past before he broke that style.

And I think, you know, that’s what we’re seeing. AI is doing for a fairly similar course to the human and learning from the art of the past. And the interesting thing is creating algorithms that don’t just do pastiche. And that’s what I think most people think, oh, surely an AI can only reproduce style. It can make something in the style of Van Gogh or Bach. But actually there are algorithms out there which are genuinely pushing into the new.

Isabel Berwick
I mean, is it almost like a kind of modern alchemy. What is this extra bit of originality?

Marcus du Sautoy
It’s not modern alchemy, it’s actually modern mathematics. (Isabel laughs) So we have this new algorithm called a Gan — Generative Adversarial Network. And it’s actually two algorithms sort of working in competition against each other to make something new. So one learns the stars of the past and the second algorithm is actually given the task to break style, to make something new. But then the original algorithm either says, “No, you haven’t broken enough” or “No, that’s far too wild”.

So this is an algorithm which I’ve seen producing a new visual artistic style, new musical styles. And you may say, well, I don’t like them, but that’s actually a good sign for creativity. If it actually puts some people off and others go wow, that’s . . . yeah, I wanna investigate that more.

Isabel Berwick
Yeah. How long is it before AI starts to work creatively on its own without supervision?

Marcus du Sautoy
I think that that might be a measure of when AI becomes conscious, because the agency and the intention to want to share something of what’s happening inside the algorithm is probably what will prompt it to start creating when we haven’t even pressed the go button.

[MUSIC PLAYING]

Isabel Berwick
Marcus, thank you so much.

Marcus du Sautoy
Thank you.

Isabel Berwick
The idea that an artificial intelligence might be able to do good creative work would have been hard to wrap your head around even a few years ago. But as Dan and Marcus said, AI is very much in the loop already, helping people in the creative industries work faster and work better. For now, it mostly seems to be helping with early-stage tasks — brainstorming or sketching out ideas — rather than producing a finished piece of work. But there are hints that these pieces of software can actually make original art and might one day do so with minimal human supervision. If the last few years have taught us anything, it’s that you shouldn’t bet against AI.

[MUSIC PLAYING]

This episode of Working It was produced by Mischa Frankl-Duval and mixed by Simon Panayi. Manuela Saragosa was the executive producer and Cheryl Brumley is the FT’s global head of audio. Thanks for listening.

[MUSIC PLAYING]