Study: Scientists make more discoveries with help from AI

A new study by a PhD student at MIT indicates that AI tools can help scientists make more discoveries, according to Nature.

In the study by Aidan Toner-Rodgers, an unspecified laboratory in the US with 1,018 researchers used an unspecified custom machine-learning tool. The work teams that were randomly assigned to use the AI ​​tool then discovered 44% more materials and created 39% more patent applications than those that did not use the tool.

At the same time, the technology had different effects on how productivity was distributed. The bottom third of researchers saw little benefit, while output doubled for top researchers.

AI automated 57% of the “idea generation” tasks and reassigned scientists to the new task of evaluating model-produced material candidates. Top-performing researchers then used their expertise to prioritize promising AI proposals, while other researchers wound up wasting significant resources testing false positives.

The survey also showed that progress has a price: 82% of researchers reported decreased satisfaction with their work due to decreased creativity and underutilization of their skills. The study has not yet been peer-reviewed by other researchers.