and the consequence of bias and/or abuse by regulators is much greater for adding things than for taking them out.
I agree for "things" as a general class, but once we've conditioned on a particular thing ("what's the optimal level of chemical Q?") it seems to me that we should have symmetric levels of knowledge about moving the level of that thing up and down when it's possible to move both directions. (Fluoride and lithium groundwater levels already vary significantly between areas--that's how we discovered their effects in the first place--and so saying "let's artificially make our groundwater like their groundwater" doesn't seem that prone to bias or abuse.)
It may be worth explicitly mentioning that if we're introducing something completely novel, when we shrink) the state of the evidence towards the reference class, the level of danger for "completely novel thing" is higher than the level of danger for "abundant common thing." I expect this will be minor, though, and many completely novel things are actually much better studied than abundant common things, because the abundant common things were grandfathered in rather than receiving serious scrutiny. (Here I'm thinking of particular artificial sweeteners, which were proven safe at levels where sugar would be toxic because they had to go to dramatically higher levels of the artificial sweetener to find any toxicity.)
I suppose we should also mention the argument that if we create the ability to add molecules to the water supply, that ability could be corrupted to nefarious ends- but I think that's a fully general argument against any infrastructure development, and should be responded to by investing in security (and secure design) rather than not investing in infrastructure.
but once we've conditioned on a particular thing ("what's the optimal level of chemical Q?") it seems to me that we should have symmetric levels of knowledge about moving the level of that thing up and down when it's possible to move both directions.
Moving the level of that thing down is limited at 0, and thus the effect of bias and abuse is also limited. Moving the level up is not so limited.
Deciding that you'll condition on a particular thing is itself subject to the same bias and abuse that deciding to add something is. Imagine regulat
A post from Gregory Cochran's and Henry Harpending's excellent blog West Hunter.
The commenter Ron Pavellas adds:
The Wasserman Test.