Scientists use AI to create new viruses: Should humans be worried as artificial intelligence learns DNA patterns?

Scientists have used artificial intelligence (AI) to create brand new viruses. This marks a significant first for the field. It raises hopes for future medical breakthroughs. However, it also raises serious safety concerns.

Creating viruses from scratch isn't entirely new in science. Researchers have long built viral genomes in laboratories. These help study antiviral drugs and vaccines. They also help scientists understand how viruses function.

This new study goes considerably further, though. It was published on 6 August in the Science journal. Researchers came from Stanford University and the Arc Institute.

The team taught AI to recognise natural DNA patterns. It then used that knowledge to design new viruses. Essentially, the AI wrote genetic "recipes" for viruses.

Scientists followed these recipes to build DNA molecules. These were then inserted into bacteria for testing. The bacteria produced viruses never before seen in nature.

These new viruses subsequently infected other bacteria. This proved the artificial viruses were genuinely functional.

Patrick Cai, a scientist at the University of Manchester, told The New York Times that it was an “important milestone”. He was not involved in the original research.

ChatGPT-like AI

The researchers used an AI system called Evo. It works somewhat like the chatbot ChatGPT. Both are types of advanced computer language models.

Instead of learning human language, Evo learned genetic code. It studied millions of genomes across different life forms. This helped it better understand natural evolutionary patterns.

Using a real virus as a template, researchers generated thousands of new genome designs. Nearly 300 were built and tested physically. Of these, sixteen proved to be viable viruses.

These new viruses differed significantly from anything found naturally. Some even borrowed components from distantly related viruses. The team also tested them against resistant bacterial strains.

A mixture of AI-designed viruses quickly overcame bacterial resistance. Natural viruses alone could not manage this. This suggests real promise for future medical therapies.

Should humans be worried?

What is important to note is the fact that these particular viruses cannot harm humans. They resemble a natural virus called Phi X-174. This virus only infects bacteria, not people.

Researchers took safety precautions while training their AI model, Evo. They deliberately excluded data on viruses infecting humans. Similar viruses affecting animals, plants and fungi were also excluded. This means Evo cannot generate viruses threatening human health.

“We just wanted to be extra careful,” said Brian Hie, an author of the new study.

Recently, the National Institutes of Health introduced new safety policies. These aim to prevent high-risk life-science research. However, computer-based research currently faces fewer restrictions, according to the NYT.

According to the National Institutes of Health (NIH), such research isn't generally prohibited. Exceptions apply only to existing biological entities.

Judging AI-generated virus risks remains genuinely difficult, though. Dr Hanke, a fellow at the Johns Hopkins Centre for Health Security, questioned whether there were hidden dangers.

“You could say, ‘Hey, genomic language model, make me an influenza genome that is modified to be more transmissible or to be more lethal,” NYT quoted Dr Hanke as speculating.