Does AI Dream of Original Ideas
Campaign Monitor's first talk about AI, given at the company's Dev Camp in June 2025. It asks whether AI can have original ideas, looks at why AI output drifts towards the average, and makes the case for the lived experience, intention and weirdness that only people bring, including in front-end engineering.
Beautiful but empty
The talk opens with an AI-generated cyberpunk sunset set beside real photos of Sydney. The AI image is polished, but it flattens a city that mixes eras and styles into a single look.
Averaging towards safe
Generative models produce content quickly by averaging existing human work, which pulls their output towards what's common and safe. The talk connects this to a wider flattening of taste, such as near-identical Airbnb interiors.
A double decay
Models decay in two ways that compound: the world moves on while their training stays frozen, and future models increasingly train on AI-generated content, so they become both outdated and less human.
What AI can't bring
Art moves people because someone turned lived experience into something you can see or hear, as in the work of Frida Kahlo, Yayoi Kusama or Bo Burnham's Inside. AI struggles with intentional absurdity, cultural nuance and detail, and makes different mistakes from people. Some artists use it as an instrument rather than a replacement.
Constraints and creativity in code
Constraints often spark more creativity than total freedom, and deliberately making the familiar strange is a creative technique in its own right. Front-end and accessibility work is creative and empathetic, and AI's training data lags behind a field that moves this fast.
Protecting time to think
Faster tools can raise what clients and employers expect, squeezing out time for deep creative work. The talk ends by urging people to lean into meaning, intention, emotion and purpose, and to take it as permission to let their unique and weird run free.
Do androids dream of original ideas? No. But you do.
An internal talk. The post on AI Dreams is a slightly modified transcript.
Read the full transcript on AI Dreams