Cheers and fears in seemingly equal measure are greeting the development of AI-created viruses never before appearing in nature. On the positive side, the 16 new AI viruses are designed to neutralize deadly E.coli infections, raising hopes of a new frontier in the treatment of drug-resistant illnesses. On the negative side, these artificially generated viruses raise concerns that the technology could make it easier to create dangerous pathogens.
As reported in the journal Science, researchers at Stanford University and the ARC Institute used genome language models, Evo 1 and Evo 2, to design complete viral genomes from scratch. Evo 1 and Evo 2 work much like the large language models common to AI but are trained using DNA rather than text. The researchers worked through thousands of possibilities. In the end, 16 genomes were synthesized and introduced to bacteria in laboratory dishes. Fully functional bacteriophages—viruses that infect and kill E.coli bacteria—were produced. The bottom line is that E.coli resistance to natural “phages” was rapidly overcome using a cocktail of AI-designed phages. E.coli might develop resistance to a single AI phage but the cocktail approach makes it less likely.
The hope is that synthetic AI phages can be used to counter antimicrobial resistance that renders antibiotics ineffective over time, reportedly killing more than one million people annually. AI phages could be used to target other harmful bacteria like tuberculosis, MRSA, and pseudomonas aeruginosa, a leading cause of medically-resistant infections acquired in hospitals, for example.
Then there is the risk. “Although this is promising for life sciences applications, it also raises urgent biosafety and biosecurity questions,” wrote Professor Tom Inglesby and Dr. Moritz Hanke of the Center for Health Security at Johns Hopkins University in an accompanying article. “The ability to compose viral genomes using generative AI now exists; the governance to safely steer it does not.”
Basically, this means there is no law against asking a genomic language model to create an influenza genome that is more transmissible or more lethal, said Dr. Hanke to the New York Times. Complicating any risk assessment is the difficulty of evaluating something that has never been seen before. Synthetic DNA purchases are not currently regulated. The National Institute of Health now bars scientists from experiments that would make biological agents more harmful but apparently does not cover AI virus generation unless it involves “a known entity of concern” like smallpox.
Compounding worries are reported instances of AI chatbots explaining to researchers how to modify and release existing pathogens so they would resist known treatments and spread more quickly, according to the New York Times. Existing safeguards have been compared to a flimsy wooden fence.
The Stanford research, led by chemical engineer Dr. Brian Hie, deliberately was not the model of efficiency. The AI models were trained on genetic data from two million phages of which 300 were selected before the final 16 phages were developed. The AI phages tended to be very similar to natural ones and just as resilient.
The researchers deliberately left out the genetic code for viruses harmful to plants, animals and humans to reduce risk. “Such genomes might encode new pathogens that cannot be contained by existing countermeasures,” said Hie and his team. Evo 2 is open-source and free of charge so anyone can build on it. Hie believes existing pathogens pose a greater risk than AI designs.
One thing is for certain: viruses once written by Nature over thousands of years can now be written by an algorithm to create something that matches nothing in any biological database on Earth.
