AI Invents 40,000 Types of Chemical Weapons in Just 6 Hours
Today, artificial intelligence (AI) technologies are mostly used for good purposes. But what if they were used for harm? Researchers at Collaborations Pharmaceuticals, a company focused on developing new medications, asked themselves this question—and were shocked by the results.
To find the answer, they conducted an experiment with an AI system originally designed to discover new drugs. During the experiment, the researchers reprogrammed the AI so that instead of searching for new medications, it would identify chemical weapons. As a result, the machine learning algorithms found 40,000 possible variants in just six hours.
“We had never thought about this before. We are well aware of the risks associated with working with pathogens and toxic substances, but it never concerned us because we mostly work virtually. Our work is based on creating machine learning models for therapeutic and toxic purposes to better assist in developing new molecules for drug discovery. For decades, we have used computers and AI to improve human health, not to harm it. We were rather naive about the potential misuse of our craft, as we always tried to avoid molecular functions that could interact with various classes of proteins essential for human life,” the researchers said.
According to the experts, even working with Ebola and neurotoxins—which could have raised concerns about possible misuse of their machine learning models—never worried the scientists. They had no idea how much damage could actually be caused if their AI was used with malicious intent.
The experiment was conducted as follows: the researchers used computational machine learning models to predict toxicity. All they had to do was adapt their methodology to select, rather than discard, toxic compounds. In the end, what started as a thought experiment turned into computational proof of concept for creating biochemical weapons.
Source
- Our other channels
- Our friends and partners