Even After $100 Billion, Self-Driving Cars Are Going Nowhere

Thanks for that!

The incident with the tangle of fallen wires seems more significant. I mean it is exactly the sort of thing I was talking about above - how do you program (or train with patterns!) for the unexpected? The car should really also 'know' about electricity and caution tapes displayed to inform corporeal drivers.

Imagine also how these cars might behave in an earthquake.

David

Human Drivers Avoid Crashes 99.999819% of the Time, Self-Driving Cars Need to Be Even Safer

A recent tweet from Matt Farah got the Jalopnik staff thinking about this. A system that’s 99.9-percent reliable sounds nearly perfect, but in reality, that 0.1-percent error rate is enormous.

So how good does a fully autonomous vehicle need to be in order to be safer than a human driver? As Jalopnik’s resident mathematician, figuring this out fell to me — and to data pulled from the National Highway Traffic Safety Administration.

[Laird]

Human Drivers Avoid Crashes 99.999819% of the Time

Don't get me wrong, I get the point, which is meaningful, and the author's choice of denominator (one mile) is kinda reasonable and meaningful too, but I think it's worth pointing out that, strictly, given that crashes are effectively instantaneous, the choice of the interval in which a crash is deemed to have occurred is arbitrary, and thus, strictly, the author could have arrived at any percentage that he liked via his choice of denominator. Instead of distance, he could have chosen time. Instead of a quantity of one (mile), he could have chosen one thousand (miles). Instead of units of miles he could have chosen kilometres. All of these choices would have led to a different percentage (ETA: with the minor exception that in switching between distance and time, the right choice of quantity and units could replicate the percentage. ETA2: and that, similarly, with the right adjustment of quantity, the percentage could be replicated when switching between units).

Don't get me wrong, I get the point, which is meaningful, and the author's choice of denominator (one mile) is kinda reasonable and meaningful too, but I think it's worth pointing out that, strictly, given that crashes are effectively instantaneous, the choice of the interval in which a crash is deemed to have occurred is arbitrary, and thus, strictly, the author could have arrived at any percentage that he liked via his choice of denominator. Instead of distance, he could have chosen time. Instead of a quantity of one (mile), he could have chosen one thousand (miles). Instead of units of miles he could have chosen kilometres. All of these choices would have led to a different percentage (ETA: with the minor exception that in switching between distance and time, the right choice of quantity and units could replicate the percentage. ETA2: and that, similarly, with the right adjustment of quantity, the percentage could be replicated when switching between units).

Clearly in any comparison the units must match. In this case the chosen units seem to be miles^(-1). I'm not quite sure what that is being compared with.

More generally, I suspect that most human crashes involve what I would call a rogue driver. I see them from time to time, but by slowing down I let the rogue driver 'succeed' and he no doubt carries on thinking what a brilliant driver he is.

If we imagine a world where rogue drivers were stopped, I think the human safety record would go up substantially.

To me, the biggest problem with AI drivers may well be unusual hazards - such as the wire tangle, and hazards that can only be vaguely specified (so thy really count as multiple different hazards) such as encountering a crash. There is an elevated section of motorway near where I live, and a few years back a large hole appeared in it. If AI cars had been driving on it, I wonder how many would have fallen into it before the road was closed!

BTW, I finally managed to get ChatGPT to let me in. I tried it extensively yesterday evening, and it left me shaken!

David

[Laird]

Clearly in any comparison the units must match.

Yep, that's a useful insight. The calculated "percentage" is not really a percentage, because of a unit mismatch (number of crashes versus miles). It is in fact a rate (crashes per mile) and it is totally bogus to "convert" this rate into a supposed percentage by multiplying by 100, and to then subtract the result of that conversion from 100% to arrive at a supposed "crash avoidance percentage".

If we imagine a world where rogue drivers were stopped, I think the human safety record would go up substantially.

As someone you'd likely label a rogue driver, I have a different hypothesis:

Imagine a world where morons driving 10 miles per hour under the speed limit weren't in the far (left in my case here in the USA) passing lane, or
weren't using their phone while driving, or
weren't properly trained to drive period.

The safety record would go up substantially as most of the "aggressive" drivers I see are actually quite skilled.

BTW, I finally managed to get ChatGPT to let me in. I tried it extensively yesterday evening, and it left me shaken!

David

David will you elaborate on this? Shaken how? Did you try the 3.5 or 4.0 version (4.0 is not free at the moment). 4.0 is extremely advanced, a significant improvement over it’s predecessor. 

Apparently a number of industy tychoons are getting worried about this technology’s impact on society: https://nypost.com/2023/03/29/musk-exper...ciety/amp/

Here’s the letter https://futureoflife.org/open-letter/pau...periments/

Contemporary AI systems are now becoming human-competitive at general tasks,[3] and we must ask ourselves: Should we let machines flood our information channels with propaganda and untruth? Should we automate away all the jobs, including the fulfilling ones? Should we develop nonhuman minds that might eventually outnumber, outsmart, obsolete and replace us?

David will you elaborate on this? Shaken how? Did you try the 3.5 or 4.0 version (4.0 is not free at the moment). 4.0 is extremely advanced, a significant improvement over it’s predecessor. 

Apparently a number of industy tychoons are getting worried about this technology’s impact on society: https://nypost.com/2023/03/29/musk-exper...ciety/amp/

Here’s the letter https://futureoflife.org/open-letter/pau...periments/

Version 3.5. I'll discuss the rest later.

David

As someone you'd likely label a rogue driver, I have a different hypothesis:

Imagine a world where morons driving 10 miles per hour under the speed limit weren't in the far (left in my case here in the USA) passing lane, or
weren't using their phone while driving, or
weren't properly trained to drive period.

The safety record would go up substantially as most of the "aggressive" drivers I see are actually quite skilled.

I suspect you may be an aggressive driver, because the rest of us find such people a pain. Just because one person has split second reactions, doesn't mean that someone else is as capable - everyone should drive with some recognition of that fact.

If everyone were an aggressive driver, the roads would be hell!

David

How Ford and VW’s multibillion-dollar self-driving car project failed

Robert Ferris

After a rush of enthusiasm, self-driving projects have grown besieged by the challenge of developing needed technology and establishing a business model that’s sufficiently profitable to justify the billions they spend.

In fact, all kinds of mobility projects are losing money — bike-share, ride-hailing, scooters and shuttles, alike.

“The challenge for Ford and for everybody else is trying to figure out how to provide these kinds of mobility services and actually build a viable, self-sustaining business out of it while keeping the cost of the service affordable for people using the services,” said Sam Abuelsamid, principal research analyst at Guidehouse Insights. “They’re still struggling. Everybody is still struggling with that part of it.