Generative AI not sophisticated enough to make scientific breakthroughs – yet!

Generative AI lacks the creativity to achieve breakthrough scientific discoveries from scratch in comparison to humans, new research from emlyon business school and Keeley School of Business reveals.

In a recent study, conducted by Amy Wenxuan Ding, Professor of Artificial Intelligence and Business Analytics at emlyon business school, and Shibo Li, Professor at Keeley School of Business, the researchers tested ChatGPT-4’s ability to function as a scientist in the molecular genetics field.

The researchers found that because current GenAI systems rely on graphs and statistics, which aren’t good at understanding cause-and-effect or the unknown, they can't yet move beyond preexisting knowledge to produce original hypotheses.

As a result of this, while ChatGPT-4 could suggest ideas and plan experiments, its discoveries were small and regardless of the originality of the findings, ChatGPT-4 had the illusion of making a completely successful discovery with overconfidence.

The researchers say that the reason for this is because GenAI models lack curiosity and imagination, something abundant in humans.

The research showed that when facing both an unknown hypothesis space and an unknown experiment space, human subjects can break through the constraints of existing knowledge by using curiosity and imagination – GenAI, however, cannot.

GenAI behaves in nearly the opposite way to humans, only seeking to overcome the cognitive limitations of individual humans as opposed to escaping the boundaries of the known hypothesis.

“If GenAI can make a (Nobel-worthy) scientific discovery similar to that of humans, it may help to change the paradigm of scientific discovery by altering the R&D process, accelerating scientific productivity, and rapidly expanding the human knowledge base,” says Professor Amy Wenxuan Ding. ”But for now, this is simply not the case, and it is useful as an assistant rather than a lead researcher.”

The findings suggest, for GenAI to make truly original scientific discoveries, scientists and researchers should examine how to improve GenAI so it can embrace curiosity and imagination.

This way, GenAI can move from being an assistant to an innovator, say the researchers.

For the time being, however, the researchers say that GenAI can be a huge benefit to improving efficiencies in research, suggesting experiments and ideas, as well as taking on time-consuming data collection tasks. But it can’t make a breakthrough discovery itself, just yet.

The study was published in the journal Scientific Reports.

//ENDS//

To receive the full research paper or speak with the researchers, please contact Ella at BlueSky Education – ella@bluesky-pr.com