AI can now design a virus from scratch and no one knows what’s next
Scientists in California have done something new: they got an AI to design a virus from nothing and it worked. They weren’t tweaking an existing one, it was done from scratch the way you’d ask a chatbot to write an essay.
The viruses in question are bacteriophages or bugs that eat bacteria, are harmless to humans, and increasingly useful as antibiotics become less effective. A team at Stanford, led by Dr Brian Hie, fed an AI model thousands of genetic sequences and asked it to invent new phage genomes. Most of what it came up with didn’t work but 16 did and when they mixed together, they killed strains of E coli that had already learned to fight off the naturally occurring version. Published in Science this week, it’s a genuine advance for something called phage therapy, which may end up treating infections that antibiotics can no longer touch.
But this is also, by the researchers’ own admission, a bit of a warning shot.
You wouldn’t think this has much to do with Ireland until you remember what this country actually makes. We’re the third-largest pharmaceutical exporter on earth, home to roughly 25 large-scale biologics facilities and nine of the world’s 10 biggest drug companies. In other words, pharma represents one fifth of our economy. So when scientists start asking who should be allowed to design a genome, and how, that’s not an abstract ethics seminar. It’s a question about an industry that a lot of us either work in or live beside.
The reassuring part is how hard this actually was. Tom Ellis, a genome engineer at Imperial College London, pointed out to The Guardian that these bacteriophages are about as simple as a genome gets because they’re just a few thousand genetic letters that are built on a template scientists already understand inside out. As of yet, nobody has shown that this trick works on anything bigger. And thankfully the team left out any genetic data on viruses that infect humans, animals or plants so we’re not going to get a 28 Days Later situation.
Reboot potential
That safeguard has a hole in it, though, and the researchers say so themselves: a model can be re-trained later on the very data it was never shown the first time round. Thomas Inglesby and Moritz Hanke of Johns Hopkins put it plainly: the ability to design a viral genome with AI now exists, and the rules for doing it safely do not. Uh oh. So maybe 28 Days Later?
Where should the guardrails actually go if we want to control this? Filippa Lentzos of King’s College London thinks the smartest place to watch isn’t the AI at all, it’s actually the point where a genetic design turns into actual DNA in the lab. Screen the orders, know your customers, keep the old-fashioned lab safety rules in place, all that jazz. Yes, it’s a more boring answer than “regulate the robots” but it’s also realistic in terms of how this work is really done.
So, the phages won their fight against the E coli. Good news? Mostly. But I can’t help think of Jeff Goldblum in Jurassic Park telling the room that scientists were so preoccupied with whether they could, they never stopped to ask if they should. Now, nobody in this story is cloning a T-Rex. It’s a virus you can’t see that’s built to kill bacteria and by every account it worked exactly as intended. But that’s precisely the point: the gap Inglesby and Hanke are worried about doesn’t open with a monster on the loose, it opens quietly, in a lab, with a screening process someone decided wasn’t worth the paperwork.
Ireland doesn’t get a say in what Stanford does next but we’ll be sitting pretty close to wherever that gap gets tested.
Subscribers 0
Fans 0
Followers 0
Followers