Onstage at TED last week, Elon Musk made the very bold proclamation that in about two years, Tesla owners will be able to go to sleep at the start of a drive and wake up at their destination.
While I applaud his showmanship and acute engineering recruiting tactics, I think he is dead wrong—and am willing to very publicly take the “over” on his bet by many, many years.
The reason for Musk’s miscalculation—either willful or not—is that the Valley loves nothing more than extrapolating the future based on exponential J-curves with accelerating limitless growth and improvement. While insanely seductive, the problem is that the J-curves almost never actually appear in nature. Rather, nature is full of S-curves, where powerful compounding growth meets environmental limitations. Things improve rapidly for a while, and then you hit a point where each incremental bit of growth or improvement becomes dramatically more expensive and time-consuming.
The many-billion-dollar question you have to ask yourself about self-driving cars is whether the technology will be good enough to be extremely valuable before the incremental cost of improvement skyrockets. Or will the rapidly scaling improvements taper off before the technology is fundamentally valuable?
Sadly for many technologists of the past and today, if you mistake an S-curve for a J-curve you are liable to be very far off in your predictions of the answer to that question and of the future.
A History of Over-Optimism
Over the years, so many well-respected and famous technologists have made the repeated mistake of overestimating the closeness of the future. Dreams of flying cars within a decade or two date back to the 1950s. There was a period after the discovery of DNA where people were sure we were close to unlocking full human cloning. More recently, we went through a phase where many thought we were within a few years of a cascade of life-extending techniques so they would live forever. Now many people think that immersive VR and AR, human-equivalent AI assistants, and full real-world autonomy are right around the corner.
It is easy to understand how this thinking happens in the tech community. In a sense, it is an overcorrection by technologists for what is considered normal human linear thinking. The physical world we occupy has very little compounding or exponential experiences. We don’t experience much calculus firsthand. So, the story goes, most people get poor training in future estimation and don’t understand how to think about compounding growth.
Technologists get it wrong on the other end. They might see or experience a moment of compounding growth, but few have the staying power and historical context (and incentive, for that matter) to see the broader picture containing that moment. Simply put, many things can look like exponential growth when you are looking at exactly the right moment a little too closely. Focusing on that moment can lead you far away from reality.
Great Companies Are Built on Successive S-Curves, not J-Curves
Most great companies have a single valuable mission—and then leverage successive waves of technology to continue to grow.
The textbook example of this is, of course, Gordon Moore’s company, Intel. Moore’s original exponential doubling prediction on transistor density in the 1960s was supposed to last for a decade or so. What made Intel a powerhouse was that it was able to ride an early S-curve wave to deliver valuable products to the market, and then consistently find new technologies and approaches that extended its overall growth.
The same might be said of a company like Facebook. If you look too closely at Facebook at the right time you might think you are looking at a J-curve. But the reality is that the company has successfully ridden a series of S-curves of adoption, technology, and use over time. It is also fundamentally limited to an S-curve pattern by the population of the world.
The key, however, was that each stage—colleges, U.S., web, mobile, profile, feed, ads—delivered enough value and growth against the overall mission to allow the company to invest in building the next piece of the equation.
Amazon is also a good example to draw from. Its long-term mission might not have changed much, but it set itself up as a platform that could harvest successive technologies as they became valuable, and then move on to the next tech as that became commercially valuable to apply.
All these great companies have a single theme. They ride successive waves of technological improvements, and rapidly hop to the next solution once further improvements in any one technical component get too hard to be worth further investment. They take advantage of the “easy” part of the technology curve, and ignore the last few painstaking percentage points of improvements.
The Self-Driving Car S-Curve Trap
Self-driving cars are different. They need to be very, very close to perfect before they are valuable at all. There is no 50% credit.
A self-driving car that works 90% of the time, or even 99% of the time, might be a nice safety addition, but it doesn’t deliver the true dream of not needing a driver at the wheel.
True autonomy requires several “nines” of reliability. And, the big question is, will the S-curve on the AI that goes into self-driving technology lose its rapid compounding growth and revert back to slow and expensive progress before or after the cars are reliable enough to use fully autonomously?
Pretend that a good college team with a shoestring budget can get to 90% reliability, and a great professional team with lots of resources can today get to 99.9% reliability. How incrementally expensive is each nine of reliability beyond that? It might be easy in a few years—or it might be nearly impossible.
This, of course, is the challenge in prediction. We are currently in a period of rapid improvement, but will that improvement taper off before or after vehicles are reliable enough to be really autonomous?
If the answer is after, Musk may be right. We will get there quickly and then companies will have the luxury to hunt out the next successive technology to improve them.
If, however, the going gets tough before fully autonomous vehicles are reliable enough, it will be another moment of extreme over-future-prediction on the level of the flying cars we are still waiting for.
Technorealism
I am deeply optimistic about technology and our ability to collaboratively build the future. But I also believe that we owe each other and the world a sober and accurate picture of where we are, and the hard work that needs to be done to move forward.
I also believe that problems where you don’t get to harvest real value until you have solved it to many degrees of completeness are far more difficult than problems where you get extreme partial credit for being on the right path.
I would love Musk to be right. But I am pretty sure he is wrong, and that while we might have very nice cruise control, we will be nowhere near full autonomy in a few years—because there is no partial credit in real autonomy.
I understand why Musk strikes the tone he does and sets the audacious timelines he does. It is inspiring and exciting to the general public. It sells more cars and it is great recruiting. It also perpetuates the cult of the founder, to his benefit. If we are making exponential predictions on an exponential curve, then all the more credit should necessarily go to the person who kicked it off in the first place.
But I am not holding my breath for him to deliver on this promise. It is both less dramatic and far harder than landing rockets back on a platform most of the time.