Humans are no longer the only ones capable of making scientific discoveries. Kepler's third law of planetary motion has been re-discovered centuries after it was first described – but this time, an artificial intelligence system is taking the credit.

Dubbed AI-Descartes, this "AI scientist" was developed by a team of researchers from IBM Research, Samsung AI, and the University of Maryland, Baltimore County (UMBC).

"I think scientists have so many different problems to solve. And if we solve them faster with AI, they just open up brand new questions for us to go after next," Tyler Josephson told Quirks & Quarks host Bob McDonald. 

Artist's concept of our solar system showing a sense of scale and distance.

The planets and dwarf planet Pluto are shown in their correct order of distance from the sun, their correct relative sizes and their correct relative orbital distances. The sizes of the bodies are greatly exaggerated relative to the orbital distances.

The faint rings of Jupiter, Uranus and Neptune are not shown. Eris, Haumea and Makemake do not appear in the illustration owing to their highly tilted orbits. The dwarf planet Ceres is not shown separately; it resides in the asteroid belt between Mars and Jupiter.

The production of synthetic zeolites for use as catalysts and adsorbents is a multi-billion-dollar industry. Zeolites are materials having molecule-scale pores that enable catalytic applications in petroleum refining, automotive emissions treatment, and biomass processing. Zeolites frequently catalyze chemical reactions in water, as is the case when producing renewable fuels and chemicals from biomass. The interactions of water molecules with the zeolite catalysts are incredibly complex, but it is important to understand how these systems behave so that zeolites can be optimized for application-based performance. Unfortunately, existing computational modeling tools struggle to both accurately and efficiently predict how complex chemical mixtures interact with zeolite catalytic active sites. This project will develop simulation techniques that will enable orders-of-magnitude improvements over current approaches, thereby generating fundamental knowledge of how water interacts with zeolite catalyst active sites. Symbolic regression, a machine learning tool for identifying equations that represent a given dataset, will also be explored for describing fundamental interactions between acids and bases. Community-based learning and outreach activities are planned to engage high school students in symbolic regression for math, science, and machine learning.

This project will advance Monte Carlo simulation methods to enable efficient sampling of water adsorption in acidic zeolites. Specific aims include 1) introducing Monte Carlo moves for directly sampling protonation states of water clusters in zeolites, 2) demonstrating simulations of competitive adsorption of water and alkanes using literature force fields, and 3) generating interatomic potentials for acid-base interactions by using symbolic regression to learn equations from quantum chemistry calculations. The anticipated outcomes will lay the groundwork for simulations of competitive chemisorption in porous materials with active sites, as well as discover new equations for simply characterizing potential energy surfaces of acid-base interactions.