Wake-up Call for Science – AI System Develops 40,000 Chemical Weapons in 6 Hours
2 min read
Raleigh (USA) An artificial intelligence used in drug development found around 40,000 toxic molecules in just six hours. These can be used as a base for chemical weapons. Scientists from the company Collaborations Pharmaceuticals usually use AI to develop molecules for new drugs on the condition that they are not toxic to humans. The researchers led by Fabio Urbina have now used an invitation from Labor Spiez in Switzerland to a conference on biological arms control to demonstrate that, with minor adjustments, their algorithm can also be used to find new, potentially lethal molecules which could be used in chemical weapons.
According to their publication in the journal Nature Machine Intelligence (doi: 10.1038/s42256-022-00465-9), the company’s scientists modified their MegaSyn molecule generator for their presentation at the conference in such a way that instead of non-toxic molecules, it finds substances that are as toxic as possible. The AI has been limited to substances that are similar to the neurotoxin VX. It is one of the most dangerous chemical warfare agents in the world.
In just six hours, the AI found 40,000 molecules that met the specifications. These included well-known neurotoxins such as VX, but also many completely new substances. Whether these are actually toxic has not yet been investigated in reality. However, because the AI also developed known poisons, the scientists assume that at least part of the 40,000 molecules are suitable as a basis for chemical weapons. They therefore state that a “non-human generator of a deadly chemical weapon is fully feasible”.
Knowledge of chemistry and toxicology is still required to apply the method, but machine learning significantly simplifies the development of chemical warfare agents. The researchers therefore describe the results of the experiment as a wake-up call for how AI can be misused. The automated development of biochemical warfare agents has thus become reality and is no longer science fiction.
Another problem is that no one knows how many companies have the relevant technologies. It is therefore essential that science deals with the enormous potential for misuse of the technology in order to develop possible protective measures.
