The scientific breakthrough that scares the world: AI created completely new viruses

The researchers, from Stanford University and the Arc Research Institute in California, made use of an advanced artificial intelligence model known as “Evo”. Similar to the ChatGPT software, which learns to generate text based on reading vast amounts of verbal information on the web, the Evo model has been trained to read and recognize patterns in the language of DNA. In terms of the software, DNA is a kind of book consisting of an alphabet with only four letters, the nucleotides A, C, G and T, which make up the instructions for building proteins and molecules.

To keep the experiment safe and because the model is only able to deal with small genomes at this stage, the researchers focused the artificial intelligence on a specific family of viruses, bacteriophages. These are viruses that only attack bacteria and are unable to infect humans. The model was trained on a well-known virus called “Phi X-174”, which attacks the E. coli bacterium, and on about 15 thousand of its relatives.

After the targeted training, the system produced about 700 thousand potential designs for new viruses. From these, the researchers selected the most promising ones, and built DNA molecules of 285 sequences in the laboratory. When the artificial DNA was injected into E.coli bacteria in petri dishes, scientific magic happened: bright spots began to appear in the cloudy layer of bacteria, a clear sign that viruses are multiplying, bursting out of the cells and leaving behind only the emptied shells of the dead bacteria. In total, 16 of the genomes produced by the software became live and active viruses.

The new viruses proved to be extremely efficient, and in fact, some even multiplied at a faster rate than the original natural virus. One researcher noted that the artificial viruses displayed features that were evolutionarily far removed from the original, implying that the software made genetic jumps that would have taken millions of years to evolve in nature.

Along with the enthusiasm, biosecurity experts are sounding dire warnings. Researchers from Johns Hopkins University noted that despite the medical potential, the technology raises urgent questions regarding international security. “The ability to assemble viral genomes using artificial intelligence now exists; but the control and regulatory mechanisms to navigate it safely – have not yet been created,” they warned.