A new study yielded surprising results when their artificial intelligence system found around 40,000 molecules that could be used as chemical weapons.
AI is currently being used in many fields, including finding new molecular structures to make medicines.
However, by abusing and reversing this process, a new study yielded astonishing results, as their artificial intelligence system found around 40,000 molecules that could be used as chemical weapons in just 6 hours. Some of these “candidates” were found to be more toxic than VX, the famous nerve agent that killed many important political figures and is now banned from war on Earth. . .all around the world.
According to Fabio Ubina, a member of Collaborations Pharmaceuticals’ drug development AI research group, when using AI to develop a drug called “MegaSyn,” he and his colleagues tried to change their choices. choose. Search queries from “reduce toxicity” to “increase toxicity”. And running this system for 6 hours gave the above surprising results.
Nonetheless, many molecular candidates similar to existing VX and chemical weapons have been identified. Unlike existing chemical weapons, however, some molecules are considered more toxic than VX.
From these results, Ubina said, “The innocuous tool involved in developing a useful drug has been transformed into a deadly molecular search engine by reversing its purpose.”
“What disappointed us the most was that many of the expected compounds were actually more toxic than VX. The point is that VX is essentially one of the most potent poisons known. You will.” Without a proven prediction, we Certainly don’t want to verify them myself. But predictive models are usually pretty good. So even though there are a lot of guesses that are false positives, we’re concerned that there are some stronger molecules out there. ”
Although Ubina did not disclose the details of the abuse prevention trial, he said that free tools such as Python and free drug datasets were used in the tests. So if you are proficient in technology and science, the content of this experiment is repeatable, low cost and short time.
“I don’t want everything to sound too sensational, but it’s easy for someone to repeat what we’ve done,” he added.
Of course, this requires some expertise. The end result won’t be very useful if someone puts them together without knowing anything about chemistry. The next step remains to synthesize these molecules. Finding a potential new drug or potentially toxic molecule is one thing, the next step in synthesis — creating a new molecule in the real world — is another.
The big problem is that no one really knows if these molecules are actually toxic. There will be some false positives.
However, the team still wants to raise awareness of the problem so that people actually pay attention to it. Mr Ubina warned that candidate molecules for chemical weapons could be easily identified using artificial intelligence and called on scientists around the world to discuss mechanisms to prevent the use of chemical weapons. Use chemical weapons. Use chemical weapons. Use chemical weapons. Abuse of open source AI systems.
“I don’t want to imply that AI will start making toxic molecules and that whole new biological warfare agents will emerge. I don’t want to sound nervous when I say there will be chemical warfare. Learning. AI control. I don’t think so.” I don’t think this will happen anytime soon. However, it is starting to become a potential possibility.
The report has now been published in the journal Nature Machine Intelligence.