Thursday, June 03, 2021

What happened to self-driving cars?

In 2014, I wrote a blog post about a fake company Elfdrive.

With a near record-setting investment announced last week, the self-driving car service Elfdrive is the hottest, most valuable technology start-up on the planet. It is also one of the most controversial.

The company, which has been the target of protests across Europe this week, has been accused of a reckless attitude toward safety, of price-gouging its customers, of putting existing cabbies out of work and of evading regulation. And it has been called trivial. In The New Yorker last year, George Packer huffed that Elfdrive typified Silicon Valley’s newfound focus on “solving all the problems of being 20 years old, with cash on hand.”

It is impossible to say whether Elfdrive is worth the $117 billion its investors believe it to be; like any start-up, it could fail. But for all its flaws, Elfdrive is anything but trivial. It could well transform transportation the way Amazon has altered shopping — by using slick, user-friendly software and mountains of data to completely reshape an existing market, ultimately making many modes of urban transportation cheaper, more flexible and more widely accessible to people across the income spectrum.

It was a spoof on Uber but now it looks more like Tesla, expect that Tesla's market value is over half a trillion, about six times larger than General Motors.

The post was really about self-driving cars which I thought at the time would be commonplace by 2020. We are mostly there but there are issues of silent communication between drivers or between a driver and a pedestrian on who goes first that's hard to duplicate for a self-driving car. There is the paradox that if we make a car that will always stop if someone runs in front of it, then some people will run in front of it.

There is also the man-bites-dog problem. Any person killed by a self-driving car will be a major news item while the person killed by a human-driven car while you've been reading this post will never be reported.

We'll get to self-driving cars eventually, it just won't be all at once. We're already have basically self-driving cars on highways and in many other roads as well. As the technology improves and people see that it's safe at some point people will say, "So why do we even need the steering wheel anymore?"


  1. I've always been skeptical about the promises made for self-driving cars, particularly after talking to a software engineer who worked on the Google car software (before Waymo). On limited access highways or limited access lanes, I think we'll see them soon. On ordinary streets I think we are still at least ten years away, for a number of reasons:
    1. insurance liability is unclear
    2. conditions on uncontrolled highways are way more unpredictable than people realize, and machine learning is only able to accurately predict based on things the algorithm has seen before.

  2. "Any person killed by a self-driving car will be a major news item." I used to think this too. But now I'm not so sure. We are already 3 years past the first pedestrian death from a self driving car (, and I don't think I've ever seen much about it on the news.

  3. Self driving cars are 10 years away and always will be.

  4. I was about to say that the self-driving car doesn't solve any real problem (the private car is a really bad/inefficient way to move people around), but then I realized that I had missed something. While the private car is an amazingly fun and liberating toy, society has come up with something even more fun: the cell phone. This means that the car industry is in serious trouble, since people would rather read their cell phones than drive, which makes public transportation more attractive. In this brave new world where people would rather ride the commuter train than drive, people will suddenly realize what an insane waste of money the private car is. Instead of owning two cars, a family could take cabs/Ubers instead of the second car, and take vacations (renting a car at/near the destination) instead of the first car. And still save a ton of money.

    Without the self-driving car, the car industry is doomed.

  5. But, but, but, in the 1960s we were promised flying cars by 2000.

  6. (Where's my jetpack!)

    Lance- here is your chance to be wrong again- give a prediction of when
    1/2 of all cars in the USA will be self driving.

  7. Honestly the computer industry required three decades after the invention of the transistor though the ideas of universal computation had arrived in the early 30s. We do not see a 1) working model for AI as a mathematical entity 2) implementable model of AI as in Von Neumann's architectural paradigm. Arguably 1 is the problem and perhaps the future generation Gauss and Terence Tao would be a mathematician adept and computing paradigms who can really foresee things in automatic proofs and converting mathematical proofs to computer representability and contribute immensely to AI. Until it happens we cannot get the required. I really do not think we are just deep networks without formal AI while driving a car. Driving a car requires about as much intelligence as doing highest order mathematics. Only thing is an academician and mathematical researcher is 1) trained 2) assimilated lot of patterns to discern and 3) retains much information of relevance to solving partiular patterns. So having a self-driving car would be as distant as having proper formal AI in a mathematical sense.

  8. Road fatalities are about 1 for every 100 million miles driven.
    Last time I checked, all self-driving companies combined have had less than that many test miles driven on actual roads, yet there has been at least one death (by Uber). They have probably surpassed 100m miles since then.
    That suggests that the tech is just not "there" yet, and I don't know if anyone knows when it will be there, if at all. By "there" I mean human-level performance under generic driving circumstances. Carefully controlled environments or environments where a lot of extra data is available are probably trackable at some point.

  9. Related: