Autonomous Cars Everywhere by 2025? Not so Fast!




​The excitement over self-driving cars has accelerated in the past few months, with increasing promises of very near (10 years or less) implementation of these systems.  Lyft CEO John Zimmer, in his interesting and visionary piece promised the Third Transportation Revolution and suggested that “by 2025 car ownership will all but end in major U.S. cities.”  Shahin Farshchi of  Lux Capital suggested that we should even accept a level of error and fatalities with self-driving technology and let it sort out its bugs.  While I am equally excited about the safety and efficiency that these systems will bring, I believe they will take significantly longer to flourish than most people think.  There are three reasons for this.


First, a problem with the above arguments is that they ignore the reality of human nature.  Here in Silicon Valley we apply technology mindset to all the problems – an approach that has worked well for many cases, but not all.  Human drivers kill many people each year (about 30,000 in U.S.), but it will not be acceptable if we had but a few technology glitches that caused fatal results, even if the percentage of those was far below human errors.  The society will simply not accept this because we are programmed to accept and deal with human risk, but we cannot assess machine risk. 


​As a result, people cannot be convinced easily to get into a driverless car that they perceive is less than perfect, even if we can prove to them that it is generally safer than human drivers.  We should note that the assumption that driverless cars will be safer than human drivers has not yet been proven yet, but it is logical and we assume that it is within reach of our technology in the next say 20 years. ​


People will not get into driverless cars unless we prove they have near-perfect safety record.


​There is also another aspect of the consumer psychology that is relevant here. Cars are purchased not just for transportation but as they provide a level of freedom and a sense of prosperity to the owners.  Thus the fact that they are used only 4% of the time (assuming that statistics counts for the time we are asleep) is irrelevant.  We don’t buy cars just as a utility to provide us with transportation.  Consumer like to have cars, if they can afford them. 

This consumer tendency will be a headwind against the ride-sharing with driverless cars, potentially limiting their growth rate, though not the eventual success of these platforms.   But it is an important headwind that is often ignored.



Second, we have to consider the fact that ride-sharing services really work best in highly dense areas of the major cities.  The entire system is based on frequent use and frequent availability of rides.  As aggregate demand drops in suburban and rural areas while at the same time distances increase, the ability of the ride-sharing platforms to maintain the required availability of cars diminish substantially, regardless of driver-based or driverless cars.  


​Now consider the fact that only for the 10% of the U.S. population lives in the major cities while the remaining 90% lives in the areas that will not be easily served by ride-sharing or autonomous cars. These includes suburban areas of the metropolitan districts which may be easier than rural areas to penetrate, but still pose significant challenges as car ownerships remain very high there, while traffic conditions are usually less congested.



​Only 10% of the U.S. population lives in the top 20 major cities.  It will be far more difficult to make self-driving ride share systems work for the remaining 90%.


While it is true that car ownership in the cities has declined somewhat, on average it is still a whopping 83% ownership, even in the big 20 cities (it is over 90% across the country).  For the 83% to drop to almost zero, it will take a lot more than a decade. 


Third, we have to consider the level of technology required for the fully autonomous (level 5) is pretty significant, especially when we go beyond safety and required high speed and efficiency. The promise of autonomous systems and ride-sharing based on self-driving cars only flourishes with Level 5 (no human driver required), and that in turn will require the vast majority, if not all, the cars on the street to be autonomous and communicating with each other. By the time we can develop such complex technology we are likely to find other breakthroughs for transportation that doesn’t require navigating our difficult streets and minding the building, people, and other obstacles.



​So while I share the enthusiasm for a safer and more efficient driving system, I am reminded of a number of previous predictions of technology take-over that proved to be many decades off.  Let’s remember that back in 1975, BusinessWeek (any many others) predicted a paperless office by the 1980s, at the latest.  Only recently, in 2016, we have seen the first sign of potential leveling off of paper’s use in the offices.   Artificial Intelligence (AI) was supposed to take over well before the begging of 21st century, and again only recently we are seeing some meaningful improvements in AI systems.  These two cases alone should be a reminder to us that technology can take a much longer to be useful and even when it is, the adoption of it will not be automatic.



​The best way to advance the growth of autonomous cars is to be transparent with the risk and the technology hurdles, while pushing ahead with the innovation.

As Scott Keogh, CEO of Audi, recently wrote in the Wall Street Journal (“The Danger of ‘Self Driving’ Car Hype”) rushing the adoption of self-driving cars can actually turn off the consumers. The best way we can advance the growth of driverless cars is to be transparent with the risk and the technology hurdles we face, while we push fully ahead with the innovation.