A lot of this article, by Sean Ekins, Filippa Lentzos, Max Brackmann, and Cédric Invernizzi, published by Bulletin of the Atomic Scientists on March 24, makes good sense – except the following two sentences:
Nature took millions of years to design proteins. AI can generate meaningful protein sequences in seconds.
The bigger question to ask here would be: if AI had to design complex life-forms from scratch, would it design protein sequences at all, forget in seconds? More broadly, “nature took millions of years to design proteins”, which knowledge was then used to train AI models, and which then generated new proteomic possibilities. Perhaps I’m reading too much into this but it’s hard not to see in throwaway lines like these the reflection of our seemingly normalised oversight of the invisible trainers and the oft-acquired-without-permission knowledge that go into making technologies like this possible. It’s also a symptom of what we have come to typically prize more: the ability of a machine to do ‘cool’ things rather than our ability to create such machines on the back of not inconsiderable human and intellectual exploitation.