Chelsea Frantzeck proposes regulating the convergence of AI and biotechnology based on emerging biological capabilities and risk
Chelsea Frantzeck proposes regulating the convergence of AI and biotechnology based on emerging biological capabilities and risk
On August 17, law graduate student Chelsea R. Frantzeck published an essay on regulating projects at the intersection of AI and biotechnology. She begins with a recent study in which models helped design the genomes of bacteriophages, viruses that infect bacteria.
In a study by Samuel King and colleagues, genomic language models generated complete genome variants of the ΦX174 phage. The researchers selected 302 variants and synthesized and assembled 285 of them. In experiments with E. coli C, 16 proved viable. Once the DNA has been synthesized and tested in bacteria, a digital genome variant can become a phage that infects the bacterium.
“What happens when a new biological capability emerges from a combination of technologies that regulators are accustomed to considering separately?” Frantzeck asks in her essay.
The model proposes a DNA sequence, synthesis makes it physical, and a laboratory test determines whether the assembled virus performs the intended function. Frantzeck includes DNA editing and laboratory work in the same chain. The same capability could help combat antibiotic-resistant bacteria or create a risk if misused. Frantzeck proposes tying oversight to the resulting capability and its expected harm, rather than to the name of any individual tool.
Turning a model’s proposal into a reproducible biological function requires ordering DNA, gaining access to equipment, and conducting laboratory validation. Each stage provides a separate point of control. This chain is examined in an analysis of the biological risk involved when a trained biologist completes the entire workflow.
Following its 1992 update, the United States federal framework for regulating biotechnology directs regulators to consider a product’s properties and the environment in which it will be used, rather than how it was created. Frantzeck proposes extending this risk-based approach. Regulatory categories should keep pace with new combinations of tools that produce biological capabilities. The level of oversight could then be tied to the risk posed by each capability.
“The goal is not to choose between innovation and regulation, but to build a governance system that recognizes emerging risks without sacrificing the potential benefits of biotechnology,” Frantzeck writes.