[jessriedel]

“Seriousness of outcomes” is a bad way to approximate the appropriate degree of regulation. We have many examples of very serious industries (e.g. pharmaceuticals, autonomous vehicles) where overly cautious regulation results in many more deaths on net.

Better rules of thumb are “whether externalities are internalized”, “asymmetry in upside vs downside”, “ability to measure outcomes”.

[voisin]

> overly cautious regulation results in many more deaths on net

How do you know this?

[jessriedel]

There's no comprehensive answer that's going to fit in this textbox but: For self-driving cars, it arises from (a) there are 30k American deaths per year from normal cars, (b) the number of deaths from self-driving cars is negligible (a few per year if you include Teslas, and zero per year if you focus on the best technology from Waymo and Cruise), and (c) everything we know about R&D points to significant acceleration possible by testing and iterating more at larger scale. For pharmaceuticals, it's a longer answer but it still starts by comparing the deaths caused by the marginal approved drug (relatively few) to the long-term benefits of trying out more drugs (high). The root cause, ultimately, is that the regulatory agency gets blamed for every death caused by something they allow but do not get blamed for deaths from the absence of things that were not allowed.

[stavros]

How many thousands of regular cars are there for every self-driving car?

[jessriedel]

It's a few hundred thousand regular cars per self-driving car. If you think my argument rests on the current self-driving cars being safer than the current regular cars, you didn't understand it.

[stavros]

You're right, I read it again and I missed the ending the first time.