Self-Driving Cars Are Self-Driving Bullets(themobilist.medium.com)
themobilist.medium.com
Self-Driving Cars Are Self-Driving Bullets
https://themobilist.medium.com/self-driving-cars-self-driving-bullets-955dfd2c5150
18 comments
Surely there will be a tipping point when human drivers are considered riskier than AI drivers. It seems far in the future, but it's probably not when you consider all of the human errors made every day.
Not that anyone cares about that, of course, because agency is preferred over outcome.
Also, it's super annoying that now I need multiple addons (bypass-paywalls-clean, ublock origin, and 'i don't care about cookies') just to browse basic blogs these days without jumping through hoops.
Not that anyone cares about that, of course, because agency is preferred over outcome.
Also, it's super annoying that now I need multiple addons (bypass-paywalls-clean, ublock origin, and 'i don't care about cookies') just to browse basic blogs these days without jumping through hoops.
This is what gets missed. Self driving cars don't need to be flawless. They just need to be some order of magnitude better than human drivers. And assistive technology is the first step towards using that automation to make driving safer.
This idea that a computer driver should never get in an accident is flawed, IMO. Sure, we don't want "computers killing people" but the very definition of an accident is an unpredictable event and we should be looking at the entire sum/statistics of the industry, not throwing up out hands at every accident. (Not saying we shouldn't learn from each one, disect it and work to do even better, we should do that also.)
And as to those that work around safety measures so they can goof off instead of driving, congrats on the Darwin award, I have zero sympathy for them. Harsh, but c'mon. And they should be charged with a crime if they do that and injure or kill someone else.
This idea that a computer driver should never get in an accident is flawed, IMO. Sure, we don't want "computers killing people" but the very definition of an accident is an unpredictable event and we should be looking at the entire sum/statistics of the industry, not throwing up out hands at every accident. (Not saying we shouldn't learn from each one, disect it and work to do even better, we should do that also.)
And as to those that work around safety measures so they can goof off instead of driving, congrats on the Darwin award, I have zero sympathy for them. Harsh, but c'mon. And they should be charged with a crime if they do that and injure or kill someone else.
Another thing that gets missed is this: With self-driving cars every single fatal accident will be investigated, and the result of that investigation will be used to improve the software on all cars in the network.
In this way the rate at which cars kill people will decline, similar to the systematic decline in airplane deaths over the years.
So maybe it's worth initially accepting a self-driving fatality rate that is slightly above the human-driving rate?
In this way the rate at which cars kill people will decline, similar to the systematic decline in airplane deaths over the years.
So maybe it's worth initially accepting a self-driving fatality rate that is slightly above the human-driving rate?
>> we kill over 36,000 people a year with cars
What about >1 million killed in car accidents each year worldwide?
Looks like the author just criticizes Musk and his overly optimistic approach. On the other hand without it US would still be flying to space on Russian rockets.
Does his approach has its flaws? Sure. Is it working? Looks like it does.
What about >1 million killed in car accidents each year worldwide?
Looks like the author just criticizes Musk and his overly optimistic approach. On the other hand without it US would still be flying to space on Russian rockets.
Does his approach has its flaws? Sure. Is it working? Looks like it does.
Both of the fatal accidents cited in this article have human errors as proximate causes: the Uber safety driver in the bicyclist jaywalking incident was watching TV on her phone and not the road, and the teenager killed when a Tesla hit a pickup wasn't wearing a seatbelt and was thrown from the vehicle.
Humans are already the weak link, and will only become more so over time.
Humans are already the weak link, and will only become more so over time.
This is the danger of requiring humans to monitor non-L5 systems for safe operation. They are incapable of babysitting robots, and their brains shut down involuntarily. Shit automation is more dangerous than no automation at all.
Her watching TV on her phone was a
voluntary action. I'm a big proponent of machine-assisted and humane attention span monitoring systems, as someone with massive attention span issues myself, but let's call a spade a spade.
This wasn't a failure of attention, this was criminal negligence on the safety driver's part.
(I would be willing to bet money that part of her job training also explicitly specified that you're not to be watching TV on your phone while driving/monitoring.)
It's plausible that if she hadn't been watching TV, she still couldn't have conducted the monitoring task successfully due to the issue you describe, but we'll never know in this instance because she was consciously and voluntarily distracting herself.
EDIT, addendum: FSD betas with human supervisors have logged many, many miles, for years now. If what you say is true (we know it's plausible), I believe we would have seen way, way more accidents and fatalities than we have. Not everything they is plausible and fits the narrative is true. It's entirely possible that the FSD betas being tested in industry rn are already safer than human drivers, if not as useful (eg getting caught up on simple edge cases and simply stopping in the middle of the road out of caution, et c).
This wasn't a failure of attention, this was criminal negligence on the safety driver's part.
(I would be willing to bet money that part of her job training also explicitly specified that you're not to be watching TV on your phone while driving/monitoring.)
It's plausible that if she hadn't been watching TV, she still couldn't have conducted the monitoring task successfully due to the issue you describe, but we'll never know in this instance because she was consciously and voluntarily distracting herself.
EDIT, addendum: FSD betas with human supervisors have logged many, many miles, for years now. If what you say is true (we know it's plausible), I believe we would have seen way, way more accidents and fatalities than we have. Not everything they is plausible and fits the narrative is true. It's entirely possible that the FSD betas being tested in industry rn are already safer than human drivers, if not as useful (eg getting caught up on simple edge cases and simply stopping in the middle of the road out of caution, et c).
> Her watching TV on her phone was a voluntary action.
This is technically true, but completely useless when it comes to actually reducing the chance of it happening again in the future, Failures of human focus and attention are, in the aggregate, inevitable, and systems like this must be designed to accommodate that.
This is technically true, but completely useless when it comes to actually reducing the chance of it happening again in the future, Failures of human focus and attention are, in the aggregate, inevitable, and systems like this must be designed to accommodate that.
Humans are wired to need stimulation. Putting someone in a self-driving car for hours on end with nothing to do but watch the road - but not drive - is inhumane.
I don't believe this to be the case.
In any case, it is insane and negligent to watch TV whilst on the job as a safety supervisor, no matter how inhumane or boring your job may be.
In any case, it is insane and negligent to watch TV whilst on the job as a safety supervisor, no matter how inhumane or boring your job may be.
> it is insane and negligent...
Perhaps, but irrelevant.
It sounds like if you were in charge of safety for these cars, you would be just fine with firing this person, replacing them with someone else who promises not to do the same, and call it solved.
We believe that likely will not solve the problem whether or not the monitor abides by their promise not to watch TV. The conditions are inherently unsafe, because no human being can sit passively without stimulation and remain alert enough to respond in a literal split second, quickly enough to avert an accident, for hours on end.
Perhaps, but irrelevant.
It sounds like if you were in charge of safety for these cars, you would be just fine with firing this person, replacing them with someone else who promises not to do the same, and call it solved.
We believe that likely will not solve the problem whether or not the monitor abides by their promise not to watch TV. The conditions are inherently unsafe, because no human being can sit passively without stimulation and remain alert enough to respond in a literal split second, quickly enough to avert an accident, for hours on end.
I heard that nuclear plants are automated, but specifically designed to force the human operator to pay attention.
Systems that are 'good enough' and infrequently fail are a human factors conundrum. This is happening with commercial aircraft now where pilots don't have the systems knowledge to understand how the complex automation fails. Most pilots don't self train for near impossible conditions like Dennis Fitch in the Sioux City crash. https://en.wikipedia.org/wiki/United_Airlines_Flight_232#Cre...
Is there any information in here beyond
A - Self driving has been hyped B - Full self driving is hard and may not happen (without any concrete arguments beyond reality is complex)
A - Self driving has been hyped B - Full self driving is hard and may not happen (without any concrete arguments beyond reality is complex)
I don't think the author was trying to write an informative piece.
My takeaway from this article was a familiar complaint for many of us: over-hyped new features being prioritized over more beneficial work. The key quote being:
> Tesla should focus instead on improving its actually-existing AI, its autopilot system. After all, the general idea of using computers to help humans avoid collisions is very good; we kill over 36,000 people a year with cars, so anything that drives that number down is worthwhile.
I think the author is saying that Tesla keeps promising a technology that seems to always be a few years away. But they could be saving lives right now with driver-assist tech that just isn't as "flashy" as full self-driving.
We're drowning under a mountain of tech debt and security issues (car crashes) that we could alleviate in 1 sprint, but the PM wants to add "cloud" to the marketing slides, so chop chop!
> Tesla should focus instead on improving its actually-existing AI, its autopilot system. After all, the general idea of using computers to help humans avoid collisions is very good; we kill over 36,000 people a year with cars, so anything that drives that number down is worthwhile.
I think the author is saying that Tesla keeps promising a technology that seems to always be a few years away. But they could be saving lives right now with driver-assist tech that just isn't as "flashy" as full self-driving.
We're drowning under a mountain of tech debt and security issues (car crashes) that we could alleviate in 1 sprint, but the PM wants to add "cloud" to the marketing slides, so chop chop!
I would be all for having separate lanes where autonomous vehicles could drive and basically have all the vehicles sharing telemetry. I feel like this is one of the only ways to control for the amount of variables you need to account for.
I have a newer model vehicle with a adaptive cruise control and I have seen it literally slam on the brakes if someone merges into my lane. This type of stuff doesn’t give me the slightest bit of confidence considering the same company is touting its self driving cars being the way of the future.