How safe are design engineers from AI?

In a study from MIT, engineers have highlighted the inherent limitations of deep generative models (DGMs) in driving true innovation in engineering design. The research, spearheaded by Lyle Regenwetter, a doctoral candidate in mechanical engineering, delves into the paradox that while DGMs are excellent at emulating existing datasets, the essence of engineering often lies in devising completely new solutions that do not yet exist.

Regenwetter and his team at MIT argue that the prevailing use of DGMs, which are calibrated for statistical similarity, may fall short when the goal is to achieve unprecedented engineering breakthroughs. "The objective of these models is to mimic a dataset," Regenwetter explains, "But as engineers and designers, we often don't want to create a design that's already out there."

This sentiment was echoed by Faez Ahmed, an assistant professor of mechanical engineering at MIT, who added, "The performance of a lot of these models is explicitly tied to how statistically similar a generated sample is to what the model has already seen. But in design, being different could be important if you want to innovate."

Their study illustrates these concepts through a case study on bicycle frame design, showing that while DGMs can generate numerous new designs, they tend to replicate existing frames without improving on structural performance or meeting new engineering criteria. Conversely, when the research team directed the same models to prioritise engineering objectives over statistical mimicry, the results were significantly more creative and performance-oriented.

The implications of this research are profound, as it suggests a recalibration of AI tools towards engineering-focused objectives could transform them into potent partners in the innovation process, offering a sophisticated 'co-pilot' for engineers.

Yet, the conversation around AI's role in innovation opens a broader debate on the definition and nature of creativity.

So can AI truly innovate?

Innovation, as defined by the Department of Industry, Innovation and Science, involves the development of more effective processes, products, and ideas, typically by challenging conventional views and drawing inspiration from a variety of sources. This is a conceptually forward-looking process, and it hinges on an ability to envisage future possibilities without being solely anchored to the past— a capability that AI does not innately possess.

AI, in its current form, is adept at analysing and learning from data to optimise existing solutions, such as with Large Language Models (LLMs). Nevertheless, it lacks the capacity for the forward vision that characterises human innovation. While AI may one day excel at completing complex tasks such as surgery or contributing to new literary works, the essence of innovation — which often involves a leap into the unknown — remains uniquely human. AI may optimise the path to a known destination, but spotting the next horizon is a journey it has yet to undertake. But who's to say it won't undertake such a journey in the future?