December 16, 2016

In Fairness

I do enjoy our running discussion about autonomous vehicles. So many of our contretempses are based on opinion and abstract theory, but on this I enjoy being 100% right and the rest of you being 100% wrong. It feels good.

But, in fairness, I am sitting on some inculpatory information. This Sunday will see a five-star, glowing Review Corner for Tim Harford's Messy: The Power of Disorder to Transform Our Lives. It is a brilliant book that affects music, art, politics, business, technology, and personal lifestyle choices.

Don't wait for Review Corner: buy it or at least listen to Russ Roberts's EconTalk podcast with the author.

I'll have a lot to say and might miss "Chapter Seven: Flight 447 and the Jennifer Unit: When Human Messiness Protects Us from Computerized Disaster."

Recall that Earl Wiener said, "Digital devices tune out small errors while creating opportunities for large errors." 21 In the case of autopilots and autonomous vehicles, we might add that it's because digital devices tidily tune out small errors that they create the opportunities for large ones. Deprived of any awkward feedback, any modest challenges that might allow us to maintain our skills, when the crisis arrives we find ourselves lamentably unprepared.

Nassim Nicholas Taleb [Review Corner] would appreciate the fragility element. I don't think it substantively undercuts the argument for autonomous vehicles, but it is a concern that the computer -- as in Flight 447 -- does the easy stuff and then hands it off when things get bad. That is a legitimate concern: "oh we're all gonna die, you better take over."

UPDATE: I meant to include one more excerpt:

With fly-by-wire, it's much easier to assess whether the trade-off is worthwhile. Until the late 1970s, one could reliably expect at least twenty-five fatal commercial plane crashes a year. In 2009, Air France 447 was one of just eight crashes, a safety record. The cost-benefit analysis seems clear: freakish accidents like Flight 447 are a price worth paying, as the steady silicon hand of the computer has prevented many others.

Still, one cannot help but wonder if there is a way to combine the adaptability, judgment, and tacit knowledge of humans with the reliability of computers, reducing the accident rate even further.

Harford, Tim. Messy: The Power of Disorder to Transform Our Lives (pp. 198-199). Penguin Publishing Group. Kindle Edition.

Technology Posted by John Kranz at December 16, 2016 4:34 PM

Indeed. Even "a long time ago, in a galaxy far, far away" the computer was an aid, or adjunct, to the human pilot. Not a replacement. And that held for ordinary run-of-the-mill X-wing Fighter pilots, not just those who were blessed with the a personal connection with "the force."

I may be mistaken but as I recall my very first objection to autonomous vehicles, and the basis for my standing as my blog brother's principal foil in the AVC (autonomous vehicle contretemps) was that the Googlemobile had NO human controls. It wasn't merely that it didn't hand over control until a fiery crash was inevitable - it didn't hand over control, period. Or even permit control. I thought I might get some sympathy from my libertarian blog brother but, and again my memory may be erroneous, I did not.

I can see where enthusiasm might have whisked him past this delicate point and into the bigger issue of automating the mundane. And, in fairness on my own part, I have held that AV's must be perfect before they be permitted to operate "in the wild." This is perhaps unfair, but at the same time may not be dismissed out of hand. Flawed as they are, human drivers have the capacity to learn from mistakes. From what I've gathered to date, AV's don't even have the power to recognize mistakes. I suspect the Ubermobile would blow that red light ten tries out of ten, under identical circumstances.

And then there is Asimov's "survival instinct." Human drivers have it, and are sympathetic to the wellbeing of other humans in their environment. AV's may have a global variable definition "bi-ped=priority 0" somewhere deep in their lukewarm, silicon little hearts, but I'm still not convinced one of them would lose any dreams of electric sheep if it were to mow down all four of the long-haired lads in the crosswalk at Abbey Road.

Posted by: johngalt at December 19, 2016 2:55 PM

Too. Much. Fun.

I'll allot some sympathy for your emergency override. I believe you'll find me very agnostic about implementation. My fulsome devotion is to automation qua automation. I want two things, and I'll let you pick the details.

First, I want the productivity. Reading The Most Important Graph

First, there is no obvious reason why growth should not continue indefinitely—although future growth will likely be more dependent on technological change than in the past. In the West, for example, we cannot replicate the growth boost that resulted from the entry of large number of women (50 percent of the population) into the labor force.

I want that slope. Watch the TeeVee News' Helicopter view of a traffic jam -- even in modest little Denver -- and imagine those man-hours returned to productivity. Gimme that and you can design the rest of the box.

Secondly, I want the reduced body count. Even Harford admits a 25 to 8 reduction in fatal commercial plane crashes per year.

As for our unfortunate pedestrian (who didn't know he was 30' from certain death until he saw the clip on YouTube...) You want 100% I think my engineer buddy knows there is no 100%. But what if we get the same better-than-3X improvement that aircraft achieved? If you were 1/3 as likely to get hit by a computer than a person -- wouldn't you take that?

(And I bet they will not be popular until they are more like 10x, but why not start at 2:1?)

Posted by: jk at December 19, 2016 5:41 PM

It's not fair. Neither of us may like it, but that's the way it is. Robots don't get a free pass for "human error."

How about a compromise? The autodrive feature automatically disengages unless you are on a freeway or an onramp? Wouldn't that solve your traffic jam issue while leaving the complicated city driving to the higher level state engine?

Posted by: johngalt at December 19, 2016 7:19 PM

Wow -- I think we have the essence of our disagreement -- the proverbial crux of the biscuit! It is "Higher level state engine."

I have zero doubt that automated drivers will be far better than a great percentage of their human counterparts. Computers will be in the 90th percentile and not susceptible to inattention when a really hot human of favored gender is visible.

I accept your concern that you're a 95th percentile driver (curiously, everyone self-reports somewhere in there, but I'm not quibbling) and fear you are "stepping down." That is actually legitimate.

I also accept the Smithian (That's Will, not Adam) concern of "Enemy of the State" where the government can drive you to the Christian White Guy interment camps instead of Starbucks.

But do you really doubt that autonomous will be a lot safer, statistically, than human drivers? Should that be the case, I don't think we have an argument, they will not be allowed until proven far superior. As long as tort lawyers need Rolexes, natural forces will keep them limited until they are far far superior.

Posted by: jk at December 20, 2016 1:16 PM

I stand by my "Summer of oh-one-four" position:

I think I mentioned I love the tech. What I don't love is the implementation chosen by Google for demonstration purposes. I want the Dodge version of this... not the PRT version as envisioned by the DAWG promoting egalitarians at google.org.

Empowerment of the less mobile? Yaay.
Faster and safer travel? Yaay.
Legalized texting while driving? Yaay.

Just put my manual controls back in. That's all I ask. Well, and maybe a turbocharged V-something with 8-speed automatic paddle shifted transmission, independent active suspension and a tuned exhaust. Or an electric power plant with a thorium battery. And voice controls.

Posted by: johngalt at December 21, 2016 6:10 PM

Okay, we're good (and I seem to hold tightly with my '14 position).

I think my favorite model is an Uber with manual and automated controls. When possible, it uses automated control but in bad weather or a very complex environment it behaves just like today's vehicle.

Posted by: jk at December 21, 2016 6:38 PM

Now were singing from the same hymnal. How about a convertible? Two-seated? Sporty looks and handling? Sort of a "Mr. Uber?" ;)

Posted by: johngalt at December 22, 2016 10:33 AM | What do you think? [7]