How safe are self-driving cars?

[@jazzysyed8536]

This really helped me on a HUGE project so thanks😁

[@AllAboutEverythingTV]

Interesting, it's easy to forget how much testing needs to be done to collect a significant amount of data, especially in something regarding public safety. Great video, thanks for making this.

[@riskbites]

Thanks. I also find it interesting that humans driving cars cause less harm than we might imagine on a miles driven basis!

[@kamikazehound3243]

Also how busy are the places they are testing and how well are the drivers on average in those areas

[@NneonNTJ]

the problem is that in most deaths in car traffic a human lays at the basis of it. if there will be accidents with self driving cars who will be in fault? the company that made them? the person not taking over the steering wheel? crashes by autonomous vehicles wont be as accepted by the population as human errors

[@riskbites]

Really important point, and one that lawyers, regulators, insurance companies and others need to be grappling with now, as irrespective of how safe or otherwise the technology is, we don't have systems in place to handle the consequences that arise when things go wrong.

[@riskbites]

This is going to be one of the more interesting legal issues as completely driverless cars become more commonplace.

[@ilyassbadoch1414]

.5 How can self-driving cars reduce traffic accidents? By eliminating human errors such as distraction, fatigue, or intoxication, self-driving cars can significantly decrease the number of traffic accidents. Al systems also react faster than humans in critical situations

[@richardconto9575]

As a critique of the video, the (implicit) idea that testing 100 self-driving cars 24x7 is the same as 300 cars 8x7 is disingenuous. The risks vary depending on time of day (light, congestion, population.) One thing that the video suggests is that monitoring would be a way to "catch" problems before they become serious - but that ignores the way computer algorithms tend to fail catastrophically without warning (making this a literal computer crash.) In order to make a quick transition (i.e.: for everyone, their replacement car would be a self-driving car), simply halving the risk of injuries and fatalities isn't going to be enough. Something much more compelling would be necessary. Of course, that's likely to be outside of the area of safety. Reduced cost, greater convenience (or utility) would have to be significant too. All-electric cars offer the promise of lower regular maintenance costs (due to the simplified mechanical systems) although the replacement costs of the battery are hard for me to estimate. (And extensive adoption of electric cars are a problem for the current models of funding roads and highways through gasoline taxes.) And my description suggests that a quick transition to significant use of self-driving cars would take around half the lifetime of an existing car, which is now between 11 and 14 years. One benefit of self-driving cars is the prospect of easy-to-hire cars - and the consequent ability to reclaim the space currently being used for driveways (if not garages and carports). Of course, the draw back is that - given current work and school scheduling paradigms - easy-to-hire cars would have to be as badly over-provisioned as cars currently are.

[@riskbites]

Thanks Richard - yes, the estimates are crude and skip over what would be important details in a fuller analysis. They are however based on Rand Corporation's modeling of safety testing (see links), which is still one of the most comprehensive analyses around. On the issue of catastrophic failure, yes, also agree -- in part. What we do know from complex/non linear systems is that it's possible to build in resilience while working out the bounds and probabilities of some failure modes, and in doing so identifying early signals of possibly greater failures. By no means foolproof, but a good step in the right direction. And of course, this whole issue needs to be approached holistically, and not just as a narrow safety issue!

[@riskbites]

I agree that it's an oversimplification, albeit one that Rand used in their analysis. But it is a starting point, and not an unusual approach to making a complex issues more tractable. That said, more nuance is definitely needed to tease apart the specifics of save versus less safe conditions.