Ezra Klein: belief

Transcript context

…I think maybe a fifth of them think that. Maybe it’s step two or step three, but it’s a pretty early step, right? I think that the question of AI or AGI in the federal government, in anywhere — and this is one reason I’ve not bought this argument about DOGE — is you have to ask, “Well what is this AI or AGI doing? What is its value function? What prompt have you given it? What have you asked it to execute across the government and how?” Alignment, which we have primarily talked about in terms of whether or not the AI, the superintelligence makes us all into paperclips, is a constant question of just near-term systems as well. I think the question of how should we prepare for AGI or for AI in the federal government first has to do with deciding what we would like the AI or the AGI to do. That could be different things to different areas. My sense — talking to a bunch of people in the companies has helped me conceptualize this better — is that the first thing I would do is begin to ask, what do I think the opportunities of AI are, scientifically and in terms of different kinds of discoveries. Let’s take drug development. I am pretty convinced that within the next five to ten years that the set of tools we are developing using AI — not all of that AGI; some of it are narrow systems, like the AlphaFold system — should get us to the place where the systems are developing molecules with a much higher probability of actually making it through trial and doing something beneficial in the human body. I think we are getting better models of different parts of the human body. ARC is trying to create a cell model right now, and I think this is going to work. Then I think you need to actually ask, how do you prepare other parts of the system to absorb what in theory will be a rapid increase in valuable candidates. Yes, so more bureaucrats for the FDA if we need them, even though AI could evaluate those drugs. A lot of government, we could just lay people off.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.