Laboratories with Ia promise a faster and faster catalyst investigation with human supervision

The urgent need for a transition to sustainable energy sources requires a significant acceleration of traditional research and development cycles. Autonomous laboratories (SDL), driven by artificial intelligence (AI), could play a fundamental role in this transformation.

In role in perspective In the newspaper Nature catalysisResearchers from the Fritz Institute theory Department have discussed the role played by humans in the future of such autonomous laboratories for catalysis research.

An autonomous laboratory integrates with laboratory and robotics automation. IA planning experiments, which are executed in increasingly automated modules (robotized). In practice, this process occurs in active learning loops, where the last loop data is used to refine an automatic learning model. The AI ​​then uses this model to plan later experiments in the following loop. In this way, only these synthesis, characterizations and tests that are more informative on the basis of all the previous data are carried out. Simultaneously, automation improves performance, reproducibility and security, promulgating significant acceleration compared to traditional development processes led by humans.

In the first implementations of this concept to discover improved catalysts, the approach often lies in replacing human tasks with synthesis robots. Researchers Christoph Scheurer and Karsten Reuter emphasize that the slowest step of such types of catalysis research is typically the explicit test of the materials. Given the growing importance of sustainability, the degradation behavior of materials in the reactor must be monitored for a long time. Therefore, performance improvements are more likely to be achieved by developing new test procedures specifically designed for SDL, instead of simply automating existing procedures.

Especially when the performance remains limited, the role of AI in the planning of the experiment is crucial. The fewer loops that need to be executed, the better. Also here, humans will continue to play a vital role in the predictable future. While the current AIS can determine optimal experiments within a given general framework, they still cannot question this frame or even redefine scientific questions themselves. At the moment, these creative tasks remain the domain of humans, which requires a function of human control within the loops.

The authors advocate for the “human in the loop” principle and analyze their implications for the development of AI in SDLs. No less important, the AIS must be able to respond in a flexible, robust and accessible way to the human modifications of loop structures, a methodological challenge that currently addresses ongoing research in the Department of Theory.