An expert in the field (Amelie Schreiber) predicts the convergence of AI models for biology to the state where designing drugs, targeted genetic mutations (which then could be enacted through genetic therapy), or viruses for specific purposes becomes really easy in a few years. (Note: apparently, the podcast was recorded before the release of AlphaFold 3.)
Unlike AGI safety, where OpenAI's argument of "iterative deployment" merits more weight[1], in biology I think we can already map the capabilities that will be unlocked. In fact, this podcast does it pretty well, it seems to me. So, it should also be possible to develop effective regulation and controls ahead of the creation of the bio-capabilities.
This raises the questions of the relative speeds of bio-capabilities, regulation, and controls development. And the ethics of pushing bio-capabilities if regulation and controls are lagging behind.
- ^
Even though many people, including me still don't think "iterative development" is justified the way OpenAI does it.