In today’s society, we use artificial intelligence in order to do good. But what if we reversed the intentions? What if we seek to do evil? This is the question Collaborations Pharmaceuticals researchers asked themselves when conducting experiments using an AI that was built to search for useful drugs aimed at curing diseases.
Looking for toxicity instead of eliminating it
The AI was modified, therefore, instead of looking for drugs to cure, to find lethal chemicals for humans and, surprisingly, the machine learning algorithm found no more and no less than 40,000 options in just six hours, according to the scientists in their article published in the journal Nature Machine Intelligence.
“The idea had never occurred to us before. We were vaguely aware of safety concerns around working with pathogens or toxic chemicals, but that didn’t relate to us; we mostly operate in a virtual environment. Our work is based on creating machine learning models for therapeutic and toxic targets to better assist in the design of new molecules for drug discovery,” the authors wrote.
A colossal array of lethal biochemical weaponry
The model created more than 40,000 lethal molecules in just six hours of operation, including several closely related to the nerve toxin VX , an inhibitor of what is known as acetylcholinesterase. The colorless or pale yellow substance VX is extremely toxic ; so much so that it is considered a weapon of mass destruction by the United Nations in its Resolution 687. The VX nerve agent was used to poison Kim Jong-nam, the half-brother of the North Korean leader in the Kuala Lumpur airport. February 2017. Kim Jong-nam was the true successor to the North Korean communist regime. This compound can cause spasms even in small doses; a larger dose can cause seizures and cause a person to stop breathing as the muscles in the lungs become paralyzed.
According to experts, the most terrifying thing about the experiment was how easy it was to use an AI to design chemical weapons similar to the most dangerous ones today.
How does the AI create toxic compounds?
“We do this by giving it a little scoring function, which gives it a high score if the molecules it generates go towards something we want. Instead of giving a low score to toxic molecules, we give a high score to toxic molecules”, explains Fabio Urbina, leader of the study.
The current experiment is a clear indication of why we need to monitor AI models more closely and really think about the consequences of our work.
This experiment was motivated by a series of convergence conferences organized by the Swiss Federal Institute for NBC (Nuclear, Biological, and Chemical) Protection and the Swiss government to identify developments in chemistry, biology, and enabling technologies that may have implications for chemistry and science. Biology.
The authors received an invitation to contribute a presentation on how AI technologies for drug discovery could potentially be misused.
Referencia: Urbina, F., Lentzos, F., Invernizzi, C. et al. Dual use of artificial-intelligence-powered drug discovery. Nature Machine Intelligence volume 4, 189–191 (2022). DOI https://doi.org/10.1038/s42256-022-00465-9
