You might be inclined to greet the news that AI has been used to design new viruses with the sardonic comment: what could possibly go wrong? But while there are reasons to worry about the safety of such a seemingly hubristic enterprise, don’t write off the idea too hastily. For one thing, this could be a valuable way to combat the alarming rise of antibiotic-resistant bacteria.
Here’s what happened. A team based at Stanford University in California has described in Science how they used AI-based biological “foundation models” – which can predict properties of simple organisms from their genomes – to devise genomes of new viruses that attack bacteria, called bacteriophages.
Such viral genomes are small enough that they can be synthesised chemically from scratch, and the researchers made about 300 of them, which they inserted into E. coli bacteria, so that the bugs made the new viruses. Sixteen of these proved able, in combination, to kill off strains of E. coli that are resistant to existing antibiotics.
So-called phage therapies, which use bacteria-targeting viruses to attack bacterial strains that resist standard antibacterial drugs, have been mooted for many decades, and were actively pursued for years behind the Iron Curtain, especially in Georgia.
They have occasionally proved effective against otherwise untreatable infections. But they are hard to commercialise or to make reliable because the treatments often contain an unknown mix of bacteriophages. Making them – and even designing them – from scratch is one way of being sure you know what you’re administering.
Suggested Reading
The grotesque show trial of Anthony Fauci
All the same, it’s right to be worried by the fact that AI models can now be used to create designer viruses. The Stanford team made sure that their viral genomes contained no genetic sequences of viruses known to infect humans or plants. We surely don’t need any new viruses in the world that are pathogenic to humans, and any that are entirely novel and non-natural might evade existing containment measures for research on such viruses.
Many researchers say that there needs to be strict regulation of genetic data relating to pathogenic viruses, so that it can’t fall into the hands of rogue actors seeking to create new bioweapons using these AI techniques.
Alarms like this were first raised in 2002, when a team in the US used chemical methods to create the genome of the polio virus entirely from scratch by stitching together DNA fragments obtained by mail order. The work showed that you don’t need access to such a pathogen to potentially weaponise it – you just need the genetic information. That work received some strong criticism, some considering it a dangerous publicity stunt.
Suggested Reading
Are you over 45 and female? Sorry, you’re not getting a job
We should keep this in proportion, however. As usual with weapons of mass destruction, you can usually do more damage with fast and dirty techniques than with efforts to make a bespoke, high-tech device. A rogue agent could far more easily genetically modify an existing killer, like the anthrax bacterium or indeed a coronavirus, to make it more virulent than they could create an entirely new one.
Still, there’s no room for complacency. Tom Inglesby and Moritz Hanke of the Johns Hopkins University School of Public Health in Baltimore say in a commentary in Science that the governance to “safely steer” the use of generative AI in virus design doesn’t yet exist; currently it relies to some extent on voluntary guidelines formulated by researchers. That had better change, and urgently.
Most popular
What the hell has happened to John Cleese?
The end of Putin is at last on the table – but we must be sceptical
My Muskpuppet moment with ruddy idiot Rupert Lowe
The letter Andy Burnham should frame and hang on the Cabinet Office wall