AI Has Potential to Create Deadly Virus
The idea of synthesizing viruses is not a recent development as researchers have been studying viruses and know how to create vaccines to counter the virus. But now, building on that advancement, scientists at Arc Institute are teaching AI to recognize patterns of DNA structure in nature and creating new viruses.
DNA is made up of nucleotides bonded together in a shape that looks like a twisted ladder, called a double helix. Every DNA has its own unique sequence and pattern. Scientists have been using an AI called Evo and using a strategy to train it like other large language models. The researchers trained Evo on genetic sequences from millions of organisms as well as from viruses, scanning about 9 trillion nucleotides overall. Eventually, Evo was able to recognize the patterns and create singular genes for certain tasks. Seeing the success in this, scientists wanted to test if Evo was able to create an entire genome, the complete set of instructions for DNA, specifically, a virus. Since bacteriophages, viruses that only infect bacteria, have tiny genomes, it was the most suitable to test using the AI.
They had Evo scan Phi X-174, a virus that only infects bacteria E. Coli and prompted it to generate a new version. Then, the scientists selected 285 genomes, they believed would be most successful, from the 700,000 versions Evo generated to test on bacteria. They made DNA molecules following the sequence, and then put them into petri dishes with bacteria. Although many of the viruses failed, one petri dish began to develop clear dots within the bacteria, meaning plaques, where the bacteria died, had formed. The scientists saw that the virus spread and left the bacteria host dead. This virus was just as resilient as Phi X-174 and multiplied even faster than the original.
The success of this brings a lot of potential to benefitting healthcare and biotechnology for more patients. Oliver Crook, a protein chemist at the University of Oxford, believes some AI generated viruses help healthcare because doctors treat patients with genetic disorders using viruses. And so, AI generated viruses can help produce beneficial proteins and remove harmful genetic disorders.
However, many people are also concerned AI could potentially bring harm to humans and other living things. Dr. Moritz Hanke, who works at Johns Hopkins Center for Health Security, is worried that the government and scientific organizations will be slow to stop deadly viruses from being created.[“the Johns Hopkins Center.”]
“You could say, ‘Hey, genomic language model, make me an influenza genome that is modified to be more transmissible or to be more lethal,” Dr. Hanke said.
As a result of these concerns, scientists did not provide Evo info on infecting humans, animals, plants, or fungi so it can’t generate any genomes that can harm humans.
