ChatGPT Struggles to Innovate Beyond Mimicry for Engineering Design

Deep generative AI models like ChatGPT excel at mimicking existing content but struggle to innovate and generate novel engineering designs.

The models' focus on statistical similarity limits their ability to meet specific engineering requirements and performance metrics.

When tasked with designing bicycle frames, standard models generated realistic but unimproved designs.

Models designed for engineering objectives produced higher performing though sometimes physically invalid designs.

Refocusing objectives beyond similarity, like performance and constraints, is key for generative AI to aid engineering innovation.