Researchers at Stanford University have used artificial intelligence to design functional viral genomes from scratch – the first time generative AI has produced viruses that actually work in the lab. It's a genuine scientific milestone, and a double-edged one.
The team used two language models, Evo 1 and Evo 2. Much like AI systems behind text generation, they were fed with data, on which they output content. The outcome following in-depth fine-tuning and filtering was 16 different synthetic viruses, so called bacteriophages, capable of infecting bacteria. These viruses could infect E.coli strains, for example. But the coin is, as always, double-sided.
While the appeal is clear, and using highly tailored viruses to kill damaging bacteria is only one use-case, this capability of superpowered bio engineering could cut both ways. The models were not given any data that could possibly design human-infecting viruses, and the team deliberately excluded viruses that infect humans, focussing only on outputting viruses that target bacteria.
The team deliberately excluded data that would let the AI generate viruses capable of infecting humans. But the paper underlines that a safeguard like this could be circumvented. While the work is "promising for life science applications," the researchers stress it comes with "urgent biosafety and biosecurity" issues – and as AI increasingly finds its way into biology, the regulatory frameworks meant to govern it are lagging behind.
In an interview with the Guardian, Tom Ellis, a professor of synthetic genome engineering at Imperial College London, called the work impressive but stressed how difficult it would be to design more complex genomes. An AI trained on dangerous pathogens could in theory be used to design harmful viruses, he acknowledged, but controlling access to genetic data and restricting the synthesis of risky genomes would help.
Still, he cautioned against overstating the danger, explaining that the threat from a fully AI-designed and written viral or bacterial genome is very overblown – and that taking existing pathogens and modifying them would be easier, and more likely to pose a real threat.
