"You don't do something that might kill or maim people to test whether or not doing the thing does kill or maim people — unless you're Mengele or something."

Have you checked out the self-driving car threads recently?

Nope. That's not what they're testing. They're testing whether or not the thing even works — which necessarily includes "doesn't kill people", sure. But that's not the specific thing being measured; it's a consequence of the system not being ready yet.

Your argument is more like, "We're testing this new chemotherapy agent to see how many people it kills", versus, "We're testing this new chemotherapy agent for safety and efficacy" (phase I and II, respectively). We aren't specifically screening for deadliness; we're trying to determine the therapeutic index of the agent, and get a baseline on its side effects (which, yes, may include death).

It's a subtle distinction, but it's a critical one.

All that aside, it's fundamentally disanalogous, anyway, because the populations in chemo clinical trials are already sick, and have given informed consent. Dude walking down the street getting smoked by a Tesla that thought he was a wall, not so much.

Regular cars come with a substantial threat of death or maiming.

> There's a very good reason we don't run experiments to test, e.g., the hypothesis that a certain chemical might cause birth defects: the consequences of the hypothesis being accurate are morally abhorrent.

Just because we don't know if it causes birth defects doesn't mean every untested chemical certainly does cause birth defects. In fact it's obvious that since we don't know, we don't know! You might save people by telling them that lie but it's still a lie and won't help anyone understand nature.

You're really confusing "We don't know if ignoring the environment will wipe us out" with "Let's ignore the environment and see what happens.". I'm not proposing we conduct that experiment, just pointing out that we don't know what the outcome will be.