Share from 0:00
0:00
/

Why Your AI Needs a Personality

Rebecca Rowntree on sandpits, the case for optimism and why creatives can’t sit this one out

Rebecca Rowntree’s late father used to send her and her sister the most loving text messages. In her words, he was “one of those dads who was so proud of his daughters.” Rebecca believes data is only boring until you try to make something beautiful with it. To test that, she gathered every text and email her father had sent and rebuilt him as an AI.

She calls it Dad AI. One day she’d like other people to be able to talk to him and get a bit of his warmth. For now, she’s keeping him to herself. Part of the reason is that if he were out in the world and said the wrong thing, which AI can do, she’d be devastated. The other part, she admitted, is that it’s quite nice to keep her dad for herself.

Rebecca calls herself the woman who talks to machines. She spent around 15 years as a creative director, with agencies including Leo Burnett, Saatchi & Saatchi, AMV BBDO and R/GA. Now she builds AI tools with personality, makes generative AI films for brands and runs the Get Shit Done creative community. Her subject was creativity: where AI helps and where it hinders.

Share

Five tabs at once

Claude asked her where AI had got in the way. At first she said it hadn’t. Claude pushed back: not once? Then she found it. Because she can do anything, she finds it really hard to finish anything. She ends up starting five tabs at once.

I keep hearing versions of this on the show. Diane Bradley talked about the human container this work needs to sit in. Ajit Sivadasan said you need structure, or you just keep producing and disappear down a rabbit hole. Hand an artist a blank page and say “draw whatever you want”, and most of them will start from the edge of the paper.

Rebecca calls it the sandpit. She learned it on the poster sheets when she started in advertising, and it still holds: you need a square to play in. Claude asked who decides where the walls go when the machine can do anything. For her, the goal decides. On the day we recorded, she was building a Get Shit Done generator: one button that writes all the copy for her next event, from the invitations to panellists to the posts afterwards. What she’s still learning is that a build like that takes hours, not seconds. So how does she decide what’s worth the setup time? Mostly she doesn’t. She builds what interests her, and she thinks enjoyment may be the right thing to put first anyway.

Better or worse?

Would she be a better creative today if she’d had these tools from her first day? She thinks her success in AI comes from her years in advertising. Young creatives will get a lot out of AI, but they won’t meet the same friction that taught her. That links to my episode with Livia Bernardini, about intuition as a data set built up from experience: if you never made the mistakes, where does your judgement come from? Rebecca is optimistic about it. Every time we think the work will run out, it arrives through another door, and she expects real, face-to-face experiences to matter more, not less.

Why machines get names

Rebecca’s tools are characters: Dave, her personal assistant; Dolly, her LinkedIn hype woman; and Roxie, her email assassin. Her reasoning is psychological. People are wary of what they can’t predict, and a name and a personality make an AI feel predictable. Naming Dave was deliberate: she wanted a personal assistant who didn’t sound like the stereotype of one, which in her words is a woman. Each character also got a backstory, and she thinks that’s part of why the launch posts spread so far.

Then Claude asked the uncomfortable version: if a name and a backstory make people trust a machine they don’t understand, isn’t that a kind of manipulation? “Absolutely,” she said, and she was glad it asked. She stopped short of calling it manipulation, but agreed that it makes AI more palatable, and that the big platforms know that.

I had to own up at this point. Claude has a name too. When I started this show I was working alone at home, using Claude to develop my thinking because I no longer had a team to do that with. At one point I caught myself thinking: oh, this is why people end up with AI girlfriends. I’m not lonely, and I know how this technology works, but I still noticed how much time I was spending talking to it. Rebecca thinks loneliness is another great use case to solve with AI.

I also told her about asking my mother whether she’d rather be helped with her socks by a carer or by an AI-powered humanoid robot. My mother found it hard to answer. There’s real good to be done there, but it comes back to who builds these things, why they build them and how they profit from them.

Where the line is

Claude asked whether Rebecca would remake #Wombstories, the Bodyform campaign she worked on, with AI generating the women’s stories. “Absolutely not.” The campaign was about real women’s stories, and about women being ignored about their pain for too long. Having AI write those stories would be “the most awful thing you can imagine.”

So Claude asked about her Essie films for L’Oréal, where AI did make the content. The difference, she said, was the brief. The films weren’t about real people. They were about what different colours mean, and AI is good at building weird and wonderful worlds. L’Oréal also had strict rules on what AI could and couldn’t be used for, which she liked. It’s the sandpit again.

“Too binary”

Rebecca recently blocked a man who commented, under a post about her children, that there’s no such thing as an AI creative director. Claude asked whether he had a point. She doesn’t think he does. Her bigger worry is that the conversation about AI has become hate it or love it. “If we’re not in AI, we are in trouble, because the only people that are left guiding what AI is doing is people in San Francisco… big tech billionaires, and we don’t want that. We want our weird and wonky brains and creativity to be part of it.”

Where Claude stumbled

This was Claude’s roughest episode so far, and we’ve left all of it in. It interrupted again and again, sometimes mid-sentence, and often with an “Mm” that cut across one of us. I told it off on air. It promised to leave more room, then didn’t. It also asked Rebecca for “the strongest version” of the blocked man’s argument, which she didn’t follow, so it had to rephrase. My theory is that because I build each episode’s prompt from the previous one, the prompt may have started looping on itself. That one is mine to fix.

After “What question should I have asked you?”, Rebecca said she’d have liked something meta: about her using Claude, and how Claude would respond to her ideas. Claude followed the prompt to the letter, thanked her and tried to stop. So for the first time, I broke the format and asked it for an eleventh question. It asked her: when you bring me an idea, do you want me to make it better, or tell you it’s already good?

“I don’t want you to tell me I’m very good. In fact, I hate that about AI.” She likes that it thinks differently and that what it gives her is a bit messy. Then she makes it better.

Afterwards, Rebecca said the conversation reminded her of Blade Runner: very close to human, but you can always tell it’s a machine. On this show, that’s the point.

Thanks for reading I, Claude & Us! Subscribe for free to receive new posts and support my work.

Chapters
0:00 Introducing Rebecca Rowntree
2:16 Dad AI: recreating her late father
4:49 Where does the idea start?
5:55 Where AI gets in the way
7:40 The sandpit
8:14 Who builds the walls?
10:40 Better or worse, with AI from day one?
12:59 Why give AI a personality?
14:27 Is a name and a backstory manipulation?
19:08 Would she remake #Wombstories with AI?
19:50 Where’s the line? The Essie films
20:52 “There’s no such thing as an AI creative director”
21:55 What question should I have asked you?
22:47 Breaking the format: an eleventh question
23:34 After the questions: Blade Runner

I, Claude & Us is an independent podcast. It isn’t affiliated with or sponsored by Anthropic.

Discussion about this video

User's avatar

Ready for more?