Data showing that people maintain functional vigilance is not "consistent with" (even the weakest kind of support) your claim that people lose functional vigilance.
It is consistent, because the paper explains that they may well maintain functional vigilance only because they are using a limited capability system where the drivers expect it to have problems. I mean, just read what the authors say. lt's not that complicated. You are absolutely correct that there will be no real data yet, because no such system capable enough to exhibit an issue yet exists.
anti-Tesla/anti-autopilot jihadists like Missy Cummings at Duke and Michael DeKort
I don't know these people and I don't care about them. My perspective is from a driver who drives to work on a multi-lane surface street LOADED with Model 3s. I think I see about 20 every day on my way to work, so there are hundreds of them around every day. It's reasonable to assume that some of these drivers may eventually decide to get FSD. I want FSD to work really well, and I don't want these people with FSD driving around looking at their phones! They'll crash into me! I want them paying full attention to the road at all times as the car drives itself - because that's what FSD will need to be to be safe, at least for the next several years. It's bad enough already with Autopilot, with a bunch of people driving around with a casual single hand at the bottom of the wheel - but that's much less of a danger to me. My fear is that this is potentially not what we will get...I don't know how having a hand on the wheel while staring at your phone while FSD does its magic is going to be safe at all, and I'm not aware that Tesla can do more than that to check driver awareness. That's what motivates my concern. Admittedly, right now, people are already staring at their phones while they drive, which is really bad, but the key difference is that they are still trying to drive (with mixed results). With FSD people might stop trying.
And the accidents while under AP is never contrasted with all the accidents avoided while under AP -- that is the overall worthwhile safety issue which the jihadists never deal with.
It's hard to measure the accidents avoided. As you say, we really just need the overall data, presented fairly and impartially, clearly indicating all relevant variables (freeway miles, surface street miles, time of day, etc., etc.), and compare comparable vehicles to Teslas. They may well be safer with AP. I have no idea. I haven't see any data that allows me to compare.
But the analysis shouldn't stop there. What is the overall assessment of actual safety results -- how many lives and cars saved compared to lost?
I agree that that is the metric. We just need really good data.
Even when some will misuse L2 and L3 and not pay attention and have accidents, my hypothesis is that many more accidents will be avoided and humans with L2 will be safer than mere humans without L2.
That is the question - what will the balance be? As I said earlier, I think the net safety level is a strong function of the % success rate of the system in an L2+ system (highly capable L2). At some point the systems will be good enough that definitely it will be safer with an L2 system and lives will be saved vs. no L2. I just don't know when that will be, and the path from the current level of safety (which may well be net beneficial, but I don't know) to that extremely capable system (which will be net beneficial) is potentially perilous.
This is simply an emprical question that will eventually be answered. Tesla, and MIT and insurance companies will provide the answers eventually.
I will believe the insurance companies for sure. I have 4 cars currently and my Tesla makes up half of my insurance premium. The Highlander has similar value and is 1/3 the cost to insure. It has primitive L2 features as well. So I suspect the reasons for the high premium in this case is the high cost & great length of Tesla repairs. These rates won't be a reflection of L2 safety for a while.
