AI Suggested 40,000 Possible Chemical Weapons Within Six Hours
The time it took for AI drug-developing AIs to create 40,000 potentially deadly molecules was less than six hours. To show how easily AI could be misused at a conference on biological arms control, researchers put AI that is normally used for searching for useful drugs in a “bad actor” mode.
Researchers had to tweak their methods to find, rather than eliminate, toxic substances. The AI came up with tens of thousands of new substances, some of which are similar to VX, the most potent nerve agent ever developed. They were shaken and published their findings in the Nature Machine Intelligence.
We were on The verge shaken by the paper. The Verge talked to Fabio Urbina about the paper’s lead author, in order to determine how concerned we should be. Collaborations Pharmaceuticals, Inc., which focuses on rare disease drug treatment, also employs him as a senior scientist.
This interview was lightly edited to improve clarity and length.
This paper can completely change the way you do your work. Please tell me about your day-to-day job.
My job is to develop and implement machine learning models for drug discovery. Many of the machine learning models we use are designed to predict toxicity. It doesn’t matter what type of drug you are trying to create, it’s important to ensure that they’re safe. It’s not a good idea to have a drug that reduces blood pressure but hits one of the important heart channels.
So, then, what was the spark that led to this study on biochemical weaponry? What was the spark?
The Convergence Conference was invited to us by the Swiss Federal Institute for Nuclear, Biological and Chemical Protection, Spiez Laboratory. This conference is intended to inform the public about new developments in tools that could have implications for Chemical/Biological Weapons Convention.
Next >>
