AI Models Show Notable Skill In Aiding Development Of Potentially Hazardous Chemical, Biological Processes

AI models are showing substantial improvement at undertaking potentially hazardous biological and chemical processes at a breakneck pace, according to a report by the UK's AI Security Institute.

These models were shown to make it five times easier for a beginner to write workable experimental protocols to recreate a virus from scratch than they would have with the internet, the report said, which tested this in a real wet lab.

Troubleshooting expertise was also observed in these large language models as they were more helpful in surpassing roadblocks in experiments than experts who had a doctorate, the report found in the Frontier AI Trends report said.

The AI models also displayed an improved ability to help with Plasmid design, an important part of genetic engineering.

"What was previously a time-intensive, multi-step process … might now be streamlined from weeks to days," the report said.

The organisation noted that “some of the barriers limiting risky research to trained specialists are eroding.”

AISI Chief Technology Officer Jade Leung informed reporters that “AI development continues to be very, very fast," according to Transformer. The research body discovered that the amount of cyber tasks that AI systems can complete on their own, such as identifying weaknesses in code, has doubled every eight months.

AISI also pointed out that AI models were exhibiting notable self-replicating abilities.

“In controlled environments, Al models are increasingly exhibiting some of the capabilities required to self-replicate across the internet,” AISI said, with success rates on its self-replication evaluations going up from below 5% to over 60% in a span of two years.

The body did however note that these AI models would be unlikely to succeed in real world settings.