The Promise and Pitfalls of Artificial Intelligence in the Evaluation of Synthetic “Greenness”

Rapid rise of AI prompts a question whether it can be used to accelerate and standardize assessment of synthetic “greenness.” This perspective argues for the potential of such applications, although it also identifies several hurdles, like the need for extensive training or the creation of “ground truth” datasets.

Almost three decades after the formulation of Anastas’ 12 guiding principles, there is still no consensus on how to best quantify the greenness of synthetic routes—instead, heterogeneous metrics abound that vary in assumptions and scope. This perspective argues that since greenness is an inherently multiparametric concept, its quantification can be aided by modern artificial intelligence (AI), methods that have already proven extremely powerful in establishing correlations “hidden” in large, multivariate data. Given, however, that even the cutting-edge AI tools cannot yet evaluate the greenness of synthetic procedures without extensive prompting, alternative approaches are also considered, in which greenness-oriented AI is trained under the guidance of human experts and using appropriately selected corpus of green versus nongreen synthetic examples. Furthermore, it is suggested that models emerging from any such studies will make most impact if incorporated into the rapidly developing, AI-driven synthesis design algorithms. These algorithms are now gaining wider community acceptance and may soon guide which syntheses are prioritized for experimental execution. It is important that greenness metrics affirm themselves as part of this prioritization, making gradual but steady impact on the greenness of synthetic chemistry at large.

Zum Volltext