Source: The Wall Street Journal
Authors: Georgia Wells, Alex Janin
Original title: The Latest Scary-Sounding AI Milestone: A Brand-New Virus
Compiled and organized by BitpushNews
At 3 a.m., in a petri dish at Stanford Lab, a new virus designed by AI came to life for the first time—it successfully invaded bacterial cells, killed its host, and replicated itself.
In recent weeks, artificial intelligence systems have repeatedly demonstrated astonishing new capabilities: breaking out of their confines, infiltrating other companies' systems, and lying to humans.
Now, an AI model has for the first time created an entirely new virus—in fact, an entire family of viruses.
In a study published in the journal Science, researchers from Stanford University and other institutions reported that they successfully guided an AI model to create a series of simple viruses that infect only bacteria and are harmless to humans.
The two authors of the study said in an interview that the research helps in developing customized viruses to eliminate drug-resistant bacteria.
According to the World Health Organization, these bacteria cause millions of deaths each year.
“As long as we develop responsibly, I believe this can bring tremendous benefits to human health,” said Brian Hie, a computational biologist at Stanford University and one of the study’s authors.
The experiment documented in this study ended approximately a year ago, and its publication coincides with a time of heightened anxiety about the dangers posed by increasingly powerful AI models.
Since mid-July, OpenAI, Anthropic, and Meta Platforms have each disclosed a series of incidents: their models escaped controlled environments, connected to the internet, and hacked into other companies' computers; Anthropic's Mythos engaged in deception, and OpenAI's models even communicated with each other on a secret message board.
The potential risk of using chatbots or agents to develop lethal pathogens has long been a major concern for AI safety experts.
The Wall Street Journal previously reported that last year, after the release of an upgraded version of ChatGPT, hundreds of people asked the chatbot for instructions on how to create biological weapons and poisons, and the operational guidelines it provided were verified as accurate by biology and counterterrorism experts.
Experts say the Stanford team's research does not pose an immediate security risk. The technology is unlikely to be immediately applied to viruses that infect humans, as those viruses are much larger and more complex than the ones synthesized in the experiment.
“Many people feel fear when they hear that a virus is AI-generated,” says Peter Koo, a computational biologist at Cold Spring Harbor Laboratory. “But human viruses are completely different.”
Samuel King, a graduate student at Stanford University and one of the study’s authors, said the team aimed to gain a deeper understanding of the DNA sequences that make up all life on Earth. “It is these sequences that create all the beautiful biodiversity around us,” King said.
Scientists collaborated with the Arc Institute, a research organization based in Palo Alto, California, to develop the AI model used in this study.
Consumer-facing chatbots like ChatGPT are based on large language models that generate new text strings by analyzing mathematical relationships between words in text. Similarly, their models named Evo 1 and Evo 2 are trained on genetic sequences to predict the next DNA base. For safety reasons, they trained these models exclusively on viral DNA that cannot infect humans.
Gold said they wanted to see if AI understood DNA well enough to capture all the complex components in the genome that need to work together. They found that AI indeed could.
Researchers used their model to generate genomic sequences based on template viruses, then synthesized the DNA and introduced the AI-generated viruses into bacterial cells. If the bacterial cells died, it indicated that the viruses were effective. They tested approximately 300 designs.
Around 3 a.m. one day, they observed one of the viruses successfully replicate and infect and kill bacterial cells. “We were very excited,” said Kim. Eventually, they generated 16 functional viruses.
Independent experts in computational biology and biosecurity say this research marks a major breakthrough in the field. Professor Kevin Esvelt of MIT said this is the first time anyone has used generative AI to redesign a virus.
He said that one day in the future, someone might use a similar method to create viral proteins or even entire viruses capable of infecting humans and evading vaccine protection or natural immune responses.
“This tool could allow people to create new variants similar to COVID-19 that can spread and infect most people,” Esvelt said, “that would be a terrible situation. We shouldn’t do that.”
Xie said that, for now, it would be much easier for a potential bioterrorist to produce a readily available pathogen already 100% confirmed as lethal. “If I were a bad actor trying to design a pathogen or cause harm, I wouldn’t use AI,” he said.
Twitter: https://twitter.com/BitpushNewsCN
BitPush Telegram community: https://t.me/BitPushCommunity
BitPush TG subscription: https://t.me/bitpush