[JamilD]

> The types of accidents we’ll face in this automated future, in which these cars are meant to run together in proximity at high speed, may be fewer, but they’ll be new, different, unpredictable and, on occasion, larger and more grisly than the ones we know today.

There's absolutely no evidence to support this, is there? How would self-driving cars result in larger and more grisly accidents?

[_rpd]

> How would self-driving cars result in larger and more grisly accidents?

Algorithmic error - algorithm responds poorly to rare environmental condition (naturally or artificially generated).

Emergent behavior error - algorithm for individual vehicle safety causes unsafe conditions for groups of vehicles in certain circumstances.

Security vulnerability - mass hacking of vehicle systems by malicious actors.

Bureaucratic vulnerability - politician or bureaucrat demands mass remote disabling of vehicles or other mass action that plays well in a television soundbite, but for which the system was not designed and can not carry out safely.

There are some big challenges for automated driving. The backlashes against the inevitable disasters could define the industry. No one is going to pay for more than token research into these areas, and regulation is always at least 20 years behind. It's up to the engineers on the ground to build in mitigation mechanisms.

[ocdtrekkie]

The suggestion I've seen repeatedly is that self-driving cars can travel faster with smaller gaps between vehicles due to strong coordination between them and near-instant response time in processing compared to a human. The article also suggests safety equipment in the vehicle would potentially be lessened as well. (Google's little koala cars would crush like tin cans, for example.)

Of course, when something goes wrong with that, yes, the accident would be more severe, because the velocity is higher and the likelihood of multi-car collision would also go up.

Of course, the effect may be akin to an airplane crash: When one crashes, the survival chances are lower, and more people are killed, than say, in a car crash. However, the frequency of the incident is drastically less, and hence airplanes are generally understood to be safer, statistically.

[protomyth]

I would hope that we spend some years not packing the cars together before we have a decent set of data on the performance under nation-wide, real world conditions.

If we expect to get to that level, then the crash investigators better be closer to airplane crash level care then our current level of investigation.

[saint_fiasco]

Packing cars together and synchronizing them is probably the safer way. Less chance that a car, person or animal will get between two cars if the space between them is less.

How many pedestrians are killed by jumping between two train wagons?

[protomyth]

I would rather we take small step, and frankly deer have no problem jumping into groups of cars or even between semis and their trailers. They are not a situationally aware animal.

> How many pedestrians are killed by jumping between two train wagons?

How many tires blow out per year, or cars get slippery?

[saint_fiasco]

Fewer tires than they would if a computer checked their air pressure diligently.

Taking small steps is fine, but the smallness should be in the location, not the density. Just a small neighborhood with only self driving cars, instead of a large neighborhood with a small proportion of self driving cars.

And of course, a deer-free neighborhood will be best to start with.

[dghughes]

I think the point is if trains of several autonomous vehicles becomes common suddenly crash is worse than just one vehicle; more mass more energy.

[maxerickson]

The sensible thing would be to build the train based on the performance characteristics of the vehicles. An algorithm will probably do a better job of this than human tailgaters presently do.

[kjbflsudfb]

Not defending the argument, but I suspect the author was referring to the reduction in safety equipment installed on autonomous vehicles leading to more grisly accidents.

From the article: "One of the claims made for autonomous cars is that they can be lighter, shedding heavy metal crash cells and expensive safety gear, like airbags, saving fuel."

[sqeaky]

I would hope we wouldn't do that until they have demonstrated years of safety.

We would also need to get most of the humans off the road. No amount of self-driving AI can dodge a human with intent (or stupidity), a similar performing vehicle and a headstart.

[kjbflsudfb]

I agree. Additionally I cannot find any reference to the author's statement about autonomous vehicles without airbags, etc.

[puzzle]

I thought Google wanted to use redundant engines, like in planes...

[fullshark]

Perhaps because accidents would be due to large scale system failures instead of idiosyncratic driver-level mistakes?