Novel skill requirements include: interpreting FSD UI, anticipating common errors, recognizing intent behind FSD actions, & remaining vigilant even during periods of autonomous success.
The opposite - where the automation handles the common case, with the human watching, simply does not work. It's not worked in aviation, it's not worked in industrial systems, it's not worked in self driving cars. This isn't a novel bit of research, this is well understood.
"I'll do 99.90% of it correctly, you watch and catch the 0.1% when I screw up!" style systems that rely on the human to catch the mistakes shouldn't be permitted to operate in any sort of safety critical environment. It simply doesn't work.
Right, because it actually relies on the human to catch 100% of mistakes _anyway_. The human can already do the 99.90% of it correctly with a reasonable degree of safety.
I wouldn't be surprised if the increased cognitive load on the human in such a system is an actual overall decrease in safety and strictly worse than a human operator alone.
At least with an unreliable system that Tesla demonstrated here.
And we still get Boeing-style @&$#-ups at the UX interface.
"You go that way, at this altitude, until I tell you to do something different" is what most of them manage, and they do it well enough. Some of the high end ones have navigational features you can use ("Follow this GPS route or VOR radial"), and in airliners, they can do quite a bit, but you still are very closely monitoring during anything important.
"In the middle of the sky," yeah, you've got some time to figure things out if George goes wonky. During a Cat III ILS approach down to minimums, you have two trained pilots, paying full attention to just what this bit of computer is doing, both prepared to abort the approach and head back into the middle of the sky with an awful lot of thrust behind them if anything goes wrong. But in cruise, there's just not too much that requires an instant response.
Solving for that, even a system that was extremely conservative would be useful, as you are completely freed up when it is operating, even if that is only 20% of the time.
The horse could go crazy in scary situations endangering others but you could train that out and make a good war horse.
Basic stuff like following the road around a bend was something the horse could do but you could not tell it to autonomously go across town to pick someone up.
People were okay with horses and knew what to expect. Tesla need to rebrand autopilot to 'horse mode' and reset people's expectations of FSD.
FSD is a complete alien that might try to kill itself for the most trivial and inscrutable reason like the sensor noise excited the neural network in a certain way.
https://www.youtube.com/watch?v=2ft9OXQ03qE (you can stop after the first hit because it repeats in loop) (I chose the less shocked one, some of them are terrible).
Do you have thoughts on why this is? Is it because, almost by definition, the remaining 0.1% are the toughest problems so the hardest to catch/solve? Or is it a human-factors issue where people get complacent when they know they can count on the machine 99.9% of the time and they lose the discipline to be vigilant enough to catch the remaining problems?
Basically if you grow up with a self driving car and never learn how to handle small fuck ups, how can you ever be expected to recover from a big fuckup? (Bad sensor reading, bad weather, shredded tire, etc)
Anticipating common errors and remaining vigilant while driving is a requirement for all driving, not when dealing with FSD.
If you are capable of driving a car safely to begin with, then you're capable of recognizing then the car is going to do something wrong and turning the wheel and/or hitting the brakes.
I wouldn't personally trust a FSD beta, but only if it is properly supervised, I think it could be tested safely. But, there's the problem -- how can you really properly supervise this? How many drivers are going to try this and just let the car drive w/o paying attention? Or, using it under controlled conditions, but then giving it too much trust in more adverse conditions?
Unless the user is able to understand and anticipate common reactions, the entire system is unsafe at >25 mph or so.
Or, to put it another way, I'd love to see the statistics on how many drivers can react quickly enough to avert an unexpected turn into an exit divider at 60 mph, after a number of days of interruption free FSD.
Being a cop on duty often exempts you from distracted driving laws.
I'm not sure how they can do stuff on a laptop while driving safely but I assume the state knows best. /s