New Scientist
AI can expand creative thinking, but its benefits appear to peak when it is used in moderation.
AI’s Goldilocks zone
Author: David Robson
“Mischa’s parents are hosting their daughter’s engagement party at their country mansion. As she makes a toast, her wine glass slips from her hand to reveal an object embedded in the stem – a tiny memory card. The discovery unravels a decades-old corruption scandal involving her own family, forcing her to choose between justice and loyalty.”
I have just asked ChatGPT to create some concepts for a movie, based on the prompt “Her face went still as her eyes widened, watching the glass shatter on the floor”, and this is one of its better suggestions. My aim is to develop a short outline of a plot within 30 minutes.
No, I don’t have any serious aspirations to break into Hollywood. But I wanted to test out the findings of a recent paper by Hsuan-Che Brad Huang during his PhD at the University of British Columbia in Canada. Huang wanted to see whether AI can enhance our own creativity – and the answer depends on how much we use it. Too little or too much and our thinking suffers. In the “Goldilocks zone”, however, it can inspire new ideas, while leaving us with a satisfying sense of ownership over the work produced.
Huang’s hypothesis was based on the fact that human creativity is often limited by our experience, so it can be useful to hear other points of view. AI could also be a tool for getting a different perspective.
I am worried about what the employment of AI might do to my long-term development
The technology has its limits, of course. Large language models, which are in essence statistical tools, often present a kind of “average response” to questions, without the idiosyncrasies of the human brain, and they often absorb biases from their training data. Worse still, they can undermine our own feelings of competence, producing a kind of apathy. So Huang suspected that moderation was key – and he tested the idea in a series of experiments.
The first involved a classic test of creativity, in which around 150 participants were asked to help a student named Mike come up with a business, given just $10 of seed funding. They were encouraged to first generate multiple ideas, before emailing Mike the best proposal.
Some participants were encouraged to use ChatGPT just once, others were recommended to use around four to six prompts and the rest were told to use it at least nine times. Their final ideas were then assessed by experts on novelty, utility and business value.
As predicted, participants in the middle group had higher average scores. Huang found the same results in a second experiment with a further 319 participants.
Finally, he surveyed professional fashion designers, visual artists, authors, animators, technologists and influencers. The participants rated statements such as “I use artificial intelligence to carry out most of my job functions” on a scale of 1 (strongly disagree) to 7 (strongly agree). Based on their bosses’ assessments, he found that those who reported moderate use of AI (around 4 or 5 on the scale) had the most creative ideas, like his experiments had suggested.
During my little experiment, I could easily see the advantage. I first asked the chatbot to think up 10 different ideas in diverse genres, before picking my favourite. I built on that idea and then asked the AI to offer a continuation of the plot.
The AI helped broaden the range of events that I was considering, but I had to use my human mind to keep the story coherent and to create a narrative arc. This almost certainly helped me come up with a more intricate plot than I could have on my own in the time that I had. But I still have reservations.
It isn’t just the ethical issues I would have in trying to pass off an AI collaboration as my own work, and the risk of plagiarising others’ work through the algorithm’s suggestions, but what I think is lost when we outsource our thinking. I am worried about what the employment of AI might do to my long-term development.
In a recent study, currently under peer review, Grace Liu at Carnegie Mellon University in Pennsylvania and her colleagues have shown that people quickly become reliant on AI and less persistent in the absence of the technology. I doubt this is a permanent change, but still – I hate the thought that I am somehow setting myself up for lazy thinking in the future. Writing, for me, should be a continual learning curve.
This isn’t to judge anyone who does choose to create with AI. I am sure many people can do so without stunting their intellect. But if genius really does involve more perspiration than inspiration, as the famous saying claims, I’d much prefer to sweat it out myself.David’s week
What I’m reading
I’m just settling into a proof copy of psychologist Vanessa Bohns’s book Should I Say Something?, which is out in October. Bohns is a wonderful writer and an expert on social interactions, so I can’t wait to get stuck in
What I’m watching
I’m loving Chris Packham’s new series Evolution on BBC Two, and rewatching Schitt’s Creek for light relief
What I’m working on
I’m now deep in the writing of my next book. More soon!
Credits: TCA, LLC.