Posts

Comments

Comment by delightfullyherald on RTFB: On the New Proposed CAIP AI Bill · 2024-04-23T19:42:24.368Z · LW · GW

The rulemaking authority procedures are anything but "standard issue boilerplate." They're novel and extremely unusual, like a lot of other things in the draft bill. 

Section 6, for example, creates a sort of one-way ratchet for rulemaking where the agency has basically unlimited authority to make rules or promulgate definitions that make it harder to get a permit, but has to make findings to make it easier. That is not how regulation usually works.

The abbreviated notice period is also really wild. 

I think the draft bill introduces a lot of interesting ideas, and that's valuable, but as actual proposed legislative language I think it's highly unrealistic and would almost certainly do more harm than good if anyone seriously tried to enact it. 

For every "wow, this has never been done before in the history of federal legislation" measure in the bill--and there are at least 50 or so--there's almost certainly going to be a pretty good reason why it hasn't been done before. In my opinion, it's not wise to try and do 50 incredibly daring new things at once in a single piece of legislation, because it creates far too many failure points. It's like following a baking recipe--if you try to make one or two tweaks to the recipe that seem like good ideas, you can then observe the effect on the finished product and draw conclusions from the results you get. If you try to write your own recipe from scratch, and you've never written a recipe before, you're going to end up with a soggy mess and no real lessons will have been learned about any of the individual elements that you tried out.