With coronavirus devastating people around the world, this might be a strange time for public health optimism, but it’s worth discussing the implications of a recent bit of good news. The world’s first antibiotic discovered by artificial intelligence, announced earlier this year, is genuinely a stunning breakthrough. It’s an example of the promise of machine learning finally delivering in a spectacular way.

The MIT researchers responsible for the new drug discovery approach named the antibiotic halicin named after HAL, the killer AI from 2001: A Space Odyssey. And halicin is a killer indeed, robustly effective against a great range of multidrug-resistant “superbug” strains of bacteria, including of Mycobacterium tuberculosis  — responsible for tuberculosis, and two of the World Health Organization’s top three priority targets for pathogen research, Acinetobacter baumannii and Enterobacteriaceae, due to their resistance to carpabenems, a class of “last resort” antibiotics beyond which we have no defense left.

The researchers first trained an artificial neural network — a computational learning system that uses a series of simple but interconnected information-processing nodes that mimics the network of neurons that make up animal brains in order to identify relationships in a set of data — with a collection of a couple thousand molecules. The collection, included existing approved drugs but also other substances that we know work to disrupt bacterial activity. The model doesn’t have to be programmed with the expertise of molecular biologists who know how antibiotics work; instead, it learns patterns of which these human experts may not be aware.