The drug connection was the context in which the prof explained “acceptable risk” concept and how it was developed. He then applied the concept to “how do government regulators define acceptable levels of environmental contaminants, such as contaminants in water, air, soil”. LD50 was also part of that lesson.
Air, soil, and water are also environments to which we all are exposed without our direct control or choice.
Ironically, Tesla has a biohazard mode on some models.
I expect that some similar “acceptable risk” will be determined for edge cases and FSD.
rainphakir, you get it.
Exactly as to risk. In modern times we have a deep social political blindness of the costs of various risks felt not directly but only statistically as members of a community. Having worked in urban planning I have come to see how costly in both deaths/injuries and loss of valuable social space has been the uncompensated taking of public roads by deadly private automobiles.
I actually have wild hope. As doctors have told me since I was 14, and emphasized since I survived HIV before useful drugs were available, I am weird. The immunologists I have consulted are convinced that my susceptibility to herpes zoster and my survival of HIV and multiple other nasties are likely linked. I now have hope that i might have coaxed the Shingrix vaccine to work for me by patiently using pre-antibiotic tuberculosis treatment regimens when I have attacks, abstaining from using acyclovir and related drugs as long as possible and forcing my vaccinated immune system system to mount whatever response it can. Over the last two years I have had weaker and weaker attacks that occur with longer and longer intervals. This attack is pathetic to those I had ten years ago,
I started with no faith. I wavered a bit, since my nephew sort of believes. ,
The problem is liability. Accidents will happen. Who owes them?
The other odd problem that industry let us get 2 billion miles and it will work? That it still won’t work.
My nephew’s job in voice dictation and other aspects of SEO for youtube videos through automation was claiming 99% success rate would be within grasp. Now they hope to get to 97%. It is BS.
Our behaviours are more complicated than the computers can keep up with.
You realize how you have to say, “I did not quite hear that”? That is normal. We are faster than the recipient of our comments.
That was a little puzzling to me as well. I didn’t see a really clear description of the accident - but this account suggests that the car had come to a complete stop, and then entered a “pull to the side of the road” mode even though the pedestrian was pinned underneath the car. The idea, I guess, is that a human driver would/should have just remained stopped and gotten out of the car to assess the situation before moving the vehicle.
I mean, I think they do. This is the DMV, after all. They’re aware that hit and runs occur. In fact, the driver who originally hit that pedestrian drove off without stopping.
I think here, the idea is that after the second car came to a complete stop, the AI programming ought to have had something that caused it to figure out whether executing the “move to curb” mode could be done without further injuring the pedestrian. That’s not a crazy thought. A human who had already come to a stop - who wasn’t going to hit and run - might do that. Get out of the car and see how badly the person had been hurt and where they were before they moved the car off to the side of the road.
You see I think that would be a logical thought. Move to the curb sounds very reasonable. At least they stayed at the location unlike the human that actually caused the accident.
Except they didn’t. They ran. But a human might have also moved to the curb. I think I would give Tesla a B+ in this situation.
Not Tesla. Cruise. And even though a human might have also moved to the curb without checking to see where the injured person was, that wouldn’t have been the correct thing to do.
The critical thing isn’t what you - or I - would give as a grade in this situation. What’s revealing is that the DMV gave Cruise an “F.” Partially on the fact that they felt Cruise wasn’t transparent, but also on the substance. They made a determination that the car didn’t safely handle that situation.
That’s why I think Tesla’s (or at least Musk’s? hard to tell where the line is) comments on regulatory approval of FSD are so laughably misguided. They seem to think that all they need to show is that FSD is safer than the average driver overall (including all the drunks and teenagers). I think that’s a misunderstanding of how these regulatory agencies are actually going to approach AI drivers.
No, I meant that metaphorically, continuing your usage. They suspended their license to operate their vehicles autonomously without a safety driver - a “failing grade” in any book. Actually, closer to being expelled.
The point being that even the California DMV may end up being a tough “grader” when it comes to allowing vehicles to operate autonomously. That’s not consistent with Tesla’s (or, again, Musk’s) characterization of Tesla’s plan for regulatory approval of FSD.
No? I am fine admitting when I’m wrong. I really mean that - I don’t think anything that I wrote or linked to about the DMV’s action would have given anyone the impression that they literally give letter grades as part of their response to life-threatening incidents. Did you really think I meant that? I kind of thought it was obvious that I was responding in like manner to your use of the metaphor of giving a letter grade.
Sigh - I didn’t think I was, or that anyone would have read my post that way. I’m sorry to have created that misimpression.
Regardless, the broader point still stands. Regulatory agencies - especially those with visible public safety purview like a state DMV - face enormous incentives to respond to very visible incidents. From a dispassionate, high-level view of public safety it might make perfect sense to allow a system that (for illustration) would prevent two drunk-driving deaths even if it causes one very high-profile pedestrian fatality. But avoided deaths are mostly invisible, while the “new” fatality is very visible - and might be unusual enough to get a lot more media attention than a run-of-the-mill DUI crash.
Again, I think Tesla’s approach to FSD will end up forcing them to do a lot more regulatory heavy lifting than their public statements indicate they are anticipating. Their strategy is to be in a position to “turn on” several hundred thousands of autonomous cars with a single update. It’s hard to imagine any set of circumstances that would result in the CA DMV letting them do that without having a very long and extensive pilot program first.
I think some context is needed here. The implication is that Cruise was banned for this one incident. I don’t think that is true. It is more the case that there has been a history of incidents and this recent one was the final straw.
In its press release, the California DMV cited “an unreasonable risk to public safety” posed by Cruise robotaxis. It is not hard to see how the DMV might arrive at that conclusion. For months, San Francisco police, fire, and transportation officials have drawn attention to a litany of incidents involving robotaxis from Cruise as well as its rival Waymo (a subsidiary of Alphabet, which also owns Google) involving blocked traffic, interrupted emergency responses, and hindered public transportation. In just the past few weeks, Cruise vehicles have collided with a fire truck, gotten stuck in wet concrete, and halted atop a pedestrian who had been struck by a hit-and-run driver.
Just this past August Cruise was ordered to cut its SF fleet in half because of traffic incidents. It looks to me like the regulators gave Cruise a lot of slack and were reluctant to pull the plug.
It is also worth noting that Waymo is not affected by the banning of Cruise. The regulators are acting against Cruise specifically rather than robotaxis in general, as seems to be implied in this thread. I don’t see why this should be a concern for Tesla.
