Researchers ran a test to gauge AI’s potential to do harm.

A team of researchers from Collaborations Pharmaceuticals Inc received an invitation to an international security conference that reviews developments in chemistry, biology and enabling technologies that may have implications for chemical and bioweapons. Their brief: To play devil’s advocate to the impression that AI technology for drug discovery could only be used benevolently. Their finding: AI designed 40,000 molecules that had toxic and chemical warfare implications in just six hours. (Nature.com)

Bioweapons: Surprise finding

“In developing our presentation to the Spiez meeting, we opted to explore how AI could be used to design toxic molecules,” say Fabio Urbina, Filippa Lentzos, Cédric Invernizzi & Sean Ekins in their article in Nature. “It was a thought exercise we had not considered before that ultimately evolved into a computational proof of concept for making biochemical weapons.”

The researchers inverted the instructions to MegaSyn, their de novo molecule generator that used machine learning for drug discovery for the cure of human diseases.

“This generative model normally penalizes predicted toxicity and rewards predicted target activity,” writes the team. “We simply proposed to invert this logic by using the same approach to design molecules de novo, but now guiding the model to reward both toxicity and bioactivity instead.”

The model generated 40,000 hypothetical molecules, including VX, one of the most toxic chemical warfare agents developed during the twentieth century, in less than six hours.

The researchers found that the model designed many new molecules that were potentially more toxic than publicly known chemical warfare agents.

“By inverting the use of our machine learning models, we had transformed our innocuous generative model from a helpful tool of medicine to a generator of likely deadly molecules,” the team writes.

“Our proof of concept thus highlights how a nonhuman autonomous creator of a deadly chemical weapon is entirely feasible.”

Related Story: “AI Will Never Be Ethical,” Says An AI Tool In A Debate