Review of the new book Autonomy about self-driving cars

I am about 3/4 through the new book Autonomy: the quest to build the driverless car and how it will reshape out world.

The authors claim we are going to shift to driverless electric or hydrogen-fuel-cell powered, two-seater cars that we do not own.

They further say that these will cut the cost of that which we now do with cars we own by 80%. They say you would only need about 15% as many of these new cars as the current number of cars to be able to give everyone about a one-minute wait for the ride they need. And they say a company providing such cars would only need about a 10% market share in a given metro area to be adequately profitable.

I am a little skeptical of the full 80%, but there is no question that we currently own relatively expensive five- or more-seat vehicles that are parked and doing nothing 95% of the time. That is stupid and inefficient. 

Elimination of collisions

There is also no question that collisions would be far fewer, like 90% the authors say. I say 99% if all cars are driverless. But what about the transition period? These two-seater electric cars are like Smart cars. Not protective in a collision. If they are ALL self-driving, the collisions will almost disappear. But during the transition, the fatality rate in the tiny new cars will be awful.

Low income will go for it

People on a tight budget will absolutely go for this reduction in costs. They must. Ditto many in urban areas. We already see this with ultra-short-term rental Zip cars, Uber, scooters, and bicycles.

POV is an image badge

But what about the fact that people’s choice of a car is all wrapped up in their self image and the image they wish to present to the world? This is America. Land of chrome, hundreds of different models of cars and trucks. The truck ads make it clear those vehicles are about your manhood—with the gravel voiced announcers and the one commercial that had one truck wetting itself when it heard the engine roar of a competing truck???

Cars have hardly ever been marketed for their low cost in this country. Rather it is glamor, extravagance, manhood, taste, style, and so forth. 

My wife and I each drive 2018 Lexuses. We’re going to switch to riding around in two-seater Smart cars who have transported who knows what to save a dollar a mile?

Will groceries and other stuff fit?

Can I fit my groceries in such a car? All my coaching stuff when I coached baseball and football? Will they be hidden from sight of the thieves in California? Suppose I forget something in the car?

Local driving only

What about long trips like the one we just did from San Francisco to Sedona, AZ to Palm Springs back to San Francisco? The authors seem to only be considering local driving.

I was in a group of Harvard grads in 2010 in San Francisco. Most of them were grads of the liberal non-profit employer grad schools. In other words, they were rather poor for people trying to live in San Francisco. One said she just bought a car for the first time in a while because she got tired of having to rely on Zip cars. She did not give details, but the breezy certainty that we will all use other people’s Smart cars instead of our own POV is belied by such anecdotes.

Body odor, pickpockets, vomit

San Francisco has the second highest utilization of mass transit—after NYC—in America. When I lived in the city, I took the Hyde Street Cable Car to work. No problem. It was empty at that hour of the morning. But I had to take the bus home, because there were too many tourists taking the cable car. I would get squashed against people with BO. My wife got pick-pocketed. Drunks throw up in mass transit, or they spill stuff. You have to wait for them to arrive. When you do not get into the car in your garage, you get wet on a rainy day riding in a public self-driving Smart car.

We already have this stuff—sort of

We can already live this future vision and we do to an extent. My wife and I use Uber in San Francisco, Vegas, and NYC. We rent cars on other trips. Sometimes Uber takes too long to get there. Sometimes we have trouble figuring out how to use a rental car’s climate control or audio equipment. Public vehicles are not perfect, They have disadvantages.

POVs can be electric or fuel cell powered

Some of the advantages of these futuristic cars can be had in a POV, like electric power or fuel cell power. We can also buy a two-seater tiny car. We don’t need to use those owned by others. We already use short-term rentals when we travel to other cities by plane.

Can get higher utilization just within your family

Similarly, a family owning its own driverless car could benefit from the vehicle not being parked 95% of the time just with use by different members within the family during the day.

Renting is not always better

We already have the option of renting or owning on all sorts of things from clothes (tux) to vehicles to houses to furniture to cell phones. Renting rarely wins all the time.

Self-driving is coming

Self-driving is surely going to take over, if only because it would eliminate collisions and the resulting deaths, injuries, and dollar costs. Electric power may also be coming if they can speed up recharging and create more power stations. Ditto fuel cells.

Tiny cars? Maybe not.

But all vehicles being like Smart cars size wise? I am very skeptical about that. Ditto about willingness to ride in public cars that cater to the same polyglot group that now takes buses.

Cars can be hacked

Also, these cars can be hacked. You could be kidnapped by a hacked car and taken to a bad place. Maybe anesthetized. 

Summary

Self-driving? Absolutely.
Urban areas where many already do not own a car? Sure.
Poor people? No question.
Electric? Probably.
Tiny? Only if they all are.
Rented short-term rather than owned? Only when you fly too far to take your POV.

Not being able to use your vehicle to prove your manhood, attract “chicks,” show how well you are doing, show how much of an outdoorsman you are, show off your great taste, etc.? I will believe that when I see it.

Why not neural network programming?

I was unpleasantly surprised to see how the self-driving car people have taught the cars how to drive correctly. Basically, they told the car to memorize every situation they told it about: roadside construction, dogs on the road, pedestrians suddenly changing their mind and crossing the street unexpectedly.

When I was a cadet at West Point, I had a company mate and classmate named Tom McNaugher. He was the smartest guy in our company. He is now a professor at George Washington University. I just saw him at the 50th reunion. 

He used to tutor other guys in the company, including me. My need was his ability to program the big computer at West Point. My efforts at it always produce a result of “manual abort” or “auto abort” or as I called them at the time: the Mexican guy Manual Abort and the German guy Otto Abort.

Anyway, Tom was famous for an admonition he kept repeating to all of us. “Don’t spec it. Understand it.” “Spec” is cadet slang for memorize. The self-driving car guys are telling their cars, “Don’t understand it. Spec it.” Actually, I just checked his yearbook write-up where I thought that phrase had been immortalized. It was, but the phrase was, “Just derive it—don’t spec it.”

I have written 36 books, 35 of which are how-to books. Writing how-to books is arguably the hardest type of book writing. Some of your readers are the top guys in the field. Most of your readers will use what you say in their lives and either it will work or fail.

In the course of writing 35 how-to books, I not only became an expert in the seven fields I wrote about, I also became an expert in acquiring expertise and imparting expertise.

Get the big-picture principles, not just memorization of an encyclopedia

In all the fields I have acquired expertise in, there are about five or six big-picture principles—McNaugher’s “derive it” or my “understand it.” For example, get-rich-quick real estate gurus love to teach you a large number of seemingly discrete, unrelated techniques. They might title a book 101 Creative Finance Techniques. And when I first started studying real estate, it seemed like I had to memorize a million different things.

But then the light bulb went on. Real estate finance is a sort of soup and there are six possible ingredients you can use in all sorts of proportions:

1. How much do you have to pay back? (payment size)
2. When do you have to pay it back? (monthly? balloon? both?)
3. What are the credit histories of the various persons agreeing to pay back the loan? (borrower, guarantor, co-signer)
4. What are the financial capacities of the various people agreeing to pay back the loan? (income, net worth)
5. What asset(s) are pledged as security for the loan? (e.g., house, car, stocks)
6. How can the lender acquire possession of or sell the asset pledged as security? (Foreclose, repossess, hold in escrow)

All of the 101 techniques are just variations on those six principles. 

Similarly, in football coaching, which I wrote eight books about, I initially tried to memorize every offense and defense and special team. Then the light bulb went on and I understood there were two big-picture principles: The basic principle of offense is “strength against weakness.” The basic principle of defense is “strength against strength.” By the way, many, maybe most, football coaches NEVER learn those principles.

So it seems to me that the self-driving people should have used NEURAL NETWORK programming rather than artificial intelligence (AI) memorization of everything. I am not sure what artificial intelligence means. And I very much doubt that most people understand it as well as I do. Is a motion-sensing light switch AI? It is less intelligent than a roach which runs away from vibration and light.

Neural network programming teaches the computer X is good; Y is bad, then teaches the computer the way you teach a small child or a pet. Child or pet does X and you say “good boy.” Child or pet does Y and you say, “No! BAD boy!”

Dollar’s worth of gasoline

Years ago, the LA Police put out a bulletin saying credit-card thieves were buying one dollar’s worth of gas after stealing cards to see if they had been reported. Mellon Bank in Pittsburgh said, “We already knew that. Our computers saw that reports of card theft ‘Y’ closely followed purchases of one dollar’s worth of gas.” The computer knew probability and statistics and saw the correlation.

Mellon’s computers had NOT been told buying a dollar’s worth of gas meant the card was stolen. It was told reports of stolen cards were bad and saw the correlation between the one-dollar sales and such reports. The self-driving car guys are telling the cars that buying one-dollar’s worth of gas is bad. They are telling the cars, “Spec it, don’t derive it.” Mellon’s computers derived it.

Dumber than a horse

A horse will not walk into a wall even if you tell it to. It will not fatally crash into a semi truck the way a self-driving Tesla did. Are horses smarter than Tesla cars? In that respect, yes, because they were neural-network trained, not given a massive encyclopedia of Y’s.

I am not a computer engineer or an automotive engineer, although I studied both in college. The guys in the book are apparently the smartest guys on earth at this. But it still seems to me that they should have just loaded the car with sensors, then driven it on the simplest course first and moving onto more and more complex courses, saying “Good boy” and “Bad boy” as appropriate all along the way. I wish I could get an explanation as to why they did not.

True, a computer can memorize a zillion things and search them fast. But neural networks and horses can interpolate and fill in gaps in the list of things to memorize and indeed be SMARTER than the human programmers like finding the dollar-of-gas predictor before the humans did.