June 21, 2017

How much safer?

I took my blog brother at face value when he reported here that the number of automotive-related deaths would "plummet" from self-driving cars, with "most analyses suggest[ing] that autonomous vehicles will eventually prevent over half of the 35,000 deaths that occur on American roads each year, and some reports are much more optimistic."

For its part, Tesla Motors has said "Brown's death is the first known fatality in over 130 million miles driven with autopilot, while there is a U.S. traffic fatality once every 94 million miles for cars not using autopilot."

So if the number of traffic fatalities was cut in half, or more, by autonomous vehicles, wouldn't autopilot have to log, on average, 188 million miles or more between individual fatalities? (Assuming just one person dies per Tesla crash, of course.) It's true that the one Tesla autopilot fatality is statistically insignificant, but if Brown had had a passenger who also died, autopilot would be demonstrably less safe than non-autopiloted vehicles.

And this simple analysis assumes that all of the vehicles on the road would be autonomous. And that all of the fatalities on the road are caused by vehicles that would be made autonomous, and not by the negligence of pedestrians, motorcyclists, cyclists, medium and heavy truck or light truck and van drivers, to name a few.

No, it seems like the life-saving effects of self-driving cars are only a slight improvement over the old fashioned distracted human driver, with its natural self-awareness and instinct for self-preservation, at least while sober. Although this beneficial conclusion is reached before a statistically significant number of interactions between autonomous vehicles and roadway flag men. How exactly do you make eye contact with a self-driving car anyway? Maybe the safety comparison is closer to unity after all.

Internecine Technology Posted by JohnGalt at June 21, 2017 7:14 PM

No. No. No. No. Nooooo! I mean, I disagree somewhat...

The 94 million figure is, sadly, based on sound statistical sampling thanks to those 35,000 data points. The denominator in 1/130,000,000 is borderline random. it could have been 4 or 200,000,000 -- n'est ce pas?

I refer you to "Getting Risk Right" [Review Corner] to see the pitfalls of comparing probabilities of unlikely events. One death of one woman almost spiked the promising technology of cell phones.

The Luddites were beaten back and millions of lives were saved -- and incredible prosperity and productivity unleashed. But how's this relate to our discussion? I forgot...

My larger argument is the improvement (dare I say "perfectibility?") of machine algorithms. People will drive just as badly in 50 years, but machines will be better.

Posted by: jk at June 22, 2017 10:37 AM

We haven't had a dustup in a while, and I almost sprained my tongue with it pressed so hard into my cheek for a five paragraph post. But we have to acknowledge that self-driving deaths do occur, and it didn't really take very long for the first one.

You may rightly say that it resulted from human error, but at the same time I'll point out that the autonomous driver required human intervention.

I'll concede that machines are more perfectible than human drivers, but my assertion has always been that the operating environment is not and will never be perfected. That's where I object to the car "driving itself." What in the world is wrong with just assisting the human driver? The answer, of course, is "nothing." It's just not as sexy sounding and futuristic as "autonomous vehicles." Don't misunderstand - I'm not saying they don't have their place. I'm just saying they should not be intended to replace all human self-reliance. And driver assistance features will make human drivers far safer too. Some might even be bold enough to say, "half or fewer fatalities per mile driven."

Posted by: johngalt at June 22, 2017 11:30 AM

We have a proximate technical agreement in the idea of AI-assistance in, well everything. I highly recommend Kasperov's Conversations with Tyler. Yes, man-machine partnerships will bring much of the safety benefits of autonomous vehicles.

I hate to take a side-road, but I am truly burning with the question "when will the first deaths occur from vehicles which stop themselves?" That's a popular feature, if commercials during sports are any indication. And it does not take much imagination to see its providing a bad outcome.

The answer to your "why not a driver?" is the productivity gains, not the safety gains. Y'know, cowboy, they wanted to keep elevator operators for the same reason. "What in the world is wrong with just assisting the human operator? The answer, of course, is 'nothing.'" Like the cell phone, the Luddites were beaten and we do not have to pay a union wage to a guy who punches buttons and prevents you from plummeting to your death.

True autonomy changes the landscape -- I want to reclaim commuting hours, move to a shared capital model instead of trillions sitting dormant 95% of the time, empower the disabled and blind, and turn the parking lots into wild animal refuges, where the deer and the antelope can play.

What's wrong with keeping a human behind the wheel? It precludes all those benefits I mentioned.

Posted by: jk at June 22, 2017 11:56 AM

How do you answer the closing question in my post: "How exactly do you make eye contact with a self-driving car anyway?"

Will flaggers ever feel safe again? Or pedestrians in general?

Posted by: johngalt at June 22, 2017 2:26 PM

I do not think that is insuperable. People did not feel safe in automated elevators for awhile, but they changed the technology to provide better indication.

Perhaps some lights on front, like the "Liddy Dole Lights" in the back window, could flashs to let you know you're "seen." I certainly think they'll be pretty effective at stopping at crosswalks and lights before they get too far.

Heck, we might automate the flaggers.

Posted by: jk at June 22, 2017 3:54 PM

Funny that I do not fear this, but I saw a "Roomba for weeds" video on Facebook and thought "Skynet. That's how it starts..."

Posted by: jk at June 22, 2017 3:56 PM

I can't wait to hear cockpit recordings of the human arguing with his AI driver. :-) IMO, assistive technology will wait, b/c the market will demand auto-taxis for the busy-busy and showoffs.

Honestly, the simplest way to be safe is to go slower, so there will be classic all2human resistance to the AI's control... but also human laziness can't be understated!

Posted by: nanobrewer at June 23, 2017 1:06 AM


Posted by: dagny at June 27, 2017 2:10 PM | What do you think? [8]