This is where my concept of lukewarm fusion comes in. All I really need right now to make it happen is a cool name.
15 Likes
Let it go; he’s just not into you.
15 Likes
This is the classic problem with open-loop automation; I don’t know if it has a formal name but I’ve always called it the “vigilance gap”. As you point out: operator vigilance decreases as the autonomous system gets closer to being fully open-loop, but the problem is that the closer you get to a true open-loop system, the more critical the set of situations where operator intervention is needed.
The traditional solution (e.g. in airplanes and spacecraft and the like) is essentially to “invent” tasks for the operator to attend to while the autonomous system is automating, keeping the operator’s attention engaged on the overall control process. But in an autonomous car the goal is to allow full driver inattention.
The problem is that, even if full open-loop automation of driving is possible, as long as we’re not at that point yet, since our goal is to completely liberate the operator’s attention, we’ll never realistically be able to bridge that vigilance gap. By inviting the operator’s attention to divert from the operation task we’re inviting the automation process to fail – catastrophically.
9 Likes
You’re a Musk fan-person, I’ll give you that. But that’s it.
10 Likes
Driver inattention may be a goal for future systems, but in current cars, as with aircraft, we need the operator to be alert. (So I don’t think you meant to use the word “but”, which implies some tension or incompatibility.)
Yes, I think we’re in complete agreement here. Autopilot would be nice, but if the robot is doing only part of the job, then the interface should be designed to enhance, rather than diminish, operator attention. It’s just like the situation you described with aircraft (which I hadn’t known about).
2 Likes
Oh yes, not trying to disagree at all! The central lie of Autopilot’s marketing is that it’s being implicitly sold in large part as an opportunity to relieve the driver of attentional burden rather than as a safety enhancement. All of Musk’s blather about “full self-driving” belies this. There’s fine print about driver vigilance that pops up when you enable Autopilot but we all knows how effective that is.
7 Likes
9 Likes
Serious answer: fatalities per vehicle mile traveled has dropped massively every year since the 1970s. By the time self-driving cars become viable and widespread, if the current trend continues, human drivers (or more specifically, vehicle safety) will be equal or better.
Really want to save lives? Promote remote work as a federal mandate for every worker for whom it is feasible. Reducing total vehicle miles and traffic density will have a greater impact than self-driving technology and can be realized immediately.
10 Likes
And he never will be.
‘False hero worship’ rarely ever ends well for the worshipper.
11 Likes
and that doesn’t even include global warming.
if you really want to figure car deaths you probably also have to consider climate change, and even electric cars consume energy
already mentioned by others but individual autonomous electric vehicles seem like just another attempt to consume our way out of the climate crisis. when really, the boring stuff like public transportation and walkable cities will do so much more if safety and… well… staying alive is really the goal
10 Likes
One could postulate that the human being should do the routine driving and the computer should take over in emergencies, but other than a few obvious cases like slamming on the brakes if the car ahead decelerates, computers are not particularly good with emergencies
The bottom line is, if AI is not ready to take responsibility for driving a car, it should not be getting involved in steering or acceleration at all
8 Likes
Not in MY backyard !!
I don’t want THOSE PEOPLE walking in MY city !!!
Can’t I just recycle my way out of global warming?
9 Likes
i think there’s an app for that
3 Likes
I know, I’ll make it illegal to build any more housing and say it’s for the environment
4 Likes
Your ideas intrigue me, and I would like to subscribe to your newsletter. I certainly want those filthy robots nowhere near me and my beloved family. If I am to die on the road, I want it to be at the hand of my fellow human – a drunk man, or a texting teenager – and not some soulless automaton!
To the onramps!!
2 Likes
Harm reduction software is going to have to come from regulation, not car salespeople. We should be building “10% fewer fatalities” and instead we got “You can play Boggle!”
2 Likes
I think this makes a lot of sense. Perhaps the aviation industry can be a model; they take safety seriously, unlike the Teslas and (God help us) Ubers of the world.
3 Likes
Not true for 2020:
The fatality rate for 2020 was 1.37 fatalities per 100 million VMT, up from 1.11 fatalities per 100 million VMT in 2019.
My wild guess is that this is a pandemic-related blip, though.
1 Like
Not an expert, but my understanding is that in 2020, traffic was reduced profoundly for a few months. Traffic slows people down, and without it people get to drive at their preferred, more-lethal speeds.
I am starting to think the end state of autonomous driving should be a subway.
1 Like
I’d argue that it was the people who didn’t drive for many months returning to the commuting grind. Combine that with people who were already paying more attention to their phone than the road and people driving during Zoom calls and you’ve got a recipe for problems.
I’ve seen more bad driving since people started returning to the office around here than in the previous decade.
4 Likes
