« How close to reality is the driverless car? », demande la quatrième de couverture. Davies attend l’épilogue, pas plus long qu’un gros article, pour commencer à esquisser les prémices d’une réflexion sur le sujet. Cette collection d’anecdotes sur les « grands défis » de la DARPA et les premières années de Chauffeur/Waymo est fort sympathique, mais elliptique et superficielle. C’est le tribut des ouvrages qui prétendent écrire une histoire qui n’est pas encore faite.


Hans Moravec, qui pensait que les robots évolueraient pour former une nouvelle espèce d’ici à 2030, p. 17 :

The first robot that could move around and “think” about its actions was Shakey, built by the Artificial Intelligence Center at Stanford Research Institute (which has since split from the university and changed its name to SRI International). But the first machine widely recognized as an autonomous vehicle came from nearby Stanford University’s Artificial Intelligence Laboratory, or SAIL. The Stanford Cart had been built in 1961, part of research into how well a human on Earth might be able to control a rover on the Moon. This cart, which looked like a card table riding a quartet of bicycle wheels, spent the next twenty years being passed from one researcher to another, each using the platform for his own application. By the time the Austrian-born computer scientist Hans Moravec adopted it in the early 1970s, it could use a camera to follow a wide white line painted on the ground, in very specific conditions, at not quite 1 mph.

Mais la question demeure : peut-on vraiment remplacer le cerveau ? La conduite autonome est-elle un problème d’intelligence artificielle généraliste, et une intelligence artificielle spécialisée suffisamment avancée peut-elle être distinguée d’une intelligence artificielle généraliste ? La question à cent-milliards, p. 39 :

Building an autonomous vehicle is a matter of replacing, with machinery, the three things humans need to drive: the hands and feet that operate the controls; the senses that perceive their surroundings; and the brain that turns data into decisions.

Le dieselgate est vraiment la meilleure chose qui soit arrivée à Volkswagen, qui a bien failli devenir le Xerox de l’industrie de l’automobile, p. 84-85 :

Dupont was an engineer with the Electronics Research Laboratory, the Palo Alto research and development center Volkswagen had established in 1998 to dip a ladle into Silicon Valley’s pool of talent and ideas. The lab’s few dozen engineers were investigating things like haptic touchscreens, electronically tintable glass, and handwriting and voice recognition. When Dupont heard that Stanford was joining the Grand Challenge, he secured permission to get the lab involved in a supporting role that played to its strengths. He presented Thrun with a whopper of an offer: Volkswagen would build Stanford a car, taking on all the work of translating a computer’s commands to the working of the throttle, brake, and steering wheel.

Et les constructeurs ne peuvent plus produire une publicité sans montrer CarPlay et Android Auto, p. 204 :

The flagship example was the “infotainment” systems that ran on cars’ center screens, with features like navigation and connection to one’s phone. Automakers struggled to make systems that were capable and easy to use, frustrating customers who’d paid extra for them. Like everything in the auto industry, the systems spent years in development, so whatever made it into a car was far from the latest and greatest tech. The difference between the screen in a driver’s hand and the one in his car was glaring.

That difference took on new importance in 2014, when Google and Apple introduced their automobile entertainment alternatives. Android Auto and Apple CarPlay let drivers project their phones onto their car’s center screens, giving them access to maps, phone contacts, messages, music, and other apps. They brought to the car operating systems that users already knew and liked, and whose quality would keep up with their phones. The popular response was summed up in the title of a report by research firm Strategy Analytics: “Consumer Interest for In-Car Smartphone Mirroring Is Almost Universal.” What scared the automakers wasn’t just that these options sapped customer interest in paying extra for their own systems, but that they ceded control of a major part of the in-car experience to Google and Apple. In the age of the smartphone, when drivers spent far more time in traffic-a full week every year, by some accounts-than on winding backroads, tech like infotainment mattered more than horsepower. That was a problem for companies that had spent decades and millions of dollars establishing brand identities with slogans like “Go Further” and “Find New Roads.” And it was just a preview of what might happen next.

De la même manière que le vélo électrique sera bientôt le vélo, et que le vélo est en train de devenir le « vélo musculaire », la voiture autonome deviendra la voiture, p. 228 :

It was impossible to know what that chase would produce, but the development of the human-driven vehicle offered a lesson. When automobiles first hit the road, people called them horseless carriages, and for good reason: They worked just like horse-drawn carriages, without the clopping and the manure. But as auto technology evolved, they become capable of much more, and people started calling them cars. The common understanding of driverless cars-as taxis without the taxi driver, or personal cars whose owners don’t have to hold the wheel-displayed the same lack of imagination. This latest technological shift was sure to produce changes nobody could foresee, and likely outgrow a name pulled from the paradigm it replaced.

La voiture de Sisyphe, p. 240 :

But none were surprised to see Waymo continue to rely on its humans. If they had realized anything in the decade since they’d first joined forces, it was that they were pursuing a problem that put a cruel spin on Sisyphus: The higher they pushed their boulder, the steeper the climb became, the more opaque the clouds that blocked their view of the summit.