Uncanny Valley

On Aug. 16, 2016, Ford announced its intent to produce a high-volume, fully autonomous SAE level 4-capable vehicle 2021 suitable for ride-hailing or ride-sharing service. Ford Chief Technical Officer Raj Nair said in an interview that Ford had decided to leap to full autonomy “because we have not found a technology that can ensure driver engagement [when not in control]”. <https://media.ford.com/content/fordmedia/fna/us/en/news/2016/08/16/ford-targets-fully-autonomous-vehicle-for-ride-sharing-in-2021.html>
SAE level 4 autonomy corresponds to the following scale:
0 = no autonomy
1 = driver assistance
2 = partial automation
3 = conditional automation
4 = high automation: steering, acceleration, deceleration, monitoring of driving environment, fallback performance of dynamic driving risk are all accomplished by automation, as well as some additional driving modes, even if a human does not respond appropriately to a request to intervene.
5 = full automation
SAE chart can be found at: < http://cyberlaw.stanford.edu/loda>

What’s notable about this announcement is the statement that Ford does not think there is a comfortable middle ground in which the human can be trusted to remain alert and engaged when automation is performing some driving tasks. Google also took this view, very early on, and headed directly to totally autonomous cars, deliberately and visibly taking the human out of the loop. Some commentators have called out Tesla as short-sighted for settling in the level 2 to 3 automation scale. There is a term, “the uncanny valley,” which describes when a humanoid robot bearing a near-identical resemblance to a human being arouses a sense of unease or revulsion in the person viewing it, because the robot appears human, except for some strange attribute that makes it definitely not human. People are freaked out and feel repulsion against those robots. There is a similar “uncanny valley” for autonomous cars and safety, where the automation is good enough that many people trust it to be infallible, when in fact it isn’t. Some safety experts call it a “chasm…” of risk, or danger.

The vote is out on whether autonomy has to be complete to be useful. The following table presents a score card:table voting L234

Like many disruptive technologies, autonomous and semi-autonomous driving has not yet settled on a singular path to development; there are many competing ideas and technologies that right now appear to all have strong advantages. Some are stronger in some markets and weaker in others; performance in snow and rain provides an excellent example for autonomous cars; or whether the drive involves construction, or going off-road. There is no clear single path right now.  There is, however, the potential for a slippery slope out of the uncanny valley.
Last week, Uber and Otto began autonomous Uber driving in Pittsburg. Note that many of the autonomous driving engineers with Uber got their start when their Carnegie Mellon automobile finished third in the DARPA Grand Challenge, so it’s not really a coincidence that those engineers went back to Pittsburgh for their autonomous taxi launch. Toyota and GM have also declared their intent to market autonomous cars for hire in the 2021 time frame.
After all, why would someone own a totally autonomous car? Semi-autonomous cars enhance the driving experience for the driver by automating many of the tasks that are tiresome: driving in heavy traffic in a straight line, finding a parking space, parallel parking, hours and hours of driving on highways. In a totally autonomous car, the human is always a passenger, akin to riding in a taxi. As long as the passenger is comfortable, the human shouldn’t care how much acceleration is available. When riding in an Uber or Lyft, I’ve found that all I care about is the comfort and cleanliness of the back seat, and I’ve yet to be disappointed in either regard.

thrill of racing
If autonomous on-demand cars were cheap and always available, would the average person own one? The IRS mileage re-imbursement rate for personally-owned automobiles is 55 cents per mile, meant to emulate an average cost of ownership over average driving. The last Lyft I took was $10 for 8 miles, (a trip that I’d previously taken by taxi for $25). That’s $1.25 per mile in the Lyft. Assuming half that fee was for my driver, an autonomous Lyft could be nearer the 60 cents per mile mark. If you could be driven to and from work daily for 60 cents per mile, be assured a ride wherever you needed it, never have to pay for parking and never get a driving or parking citation, a lot of people in cities would find that a cost-effective alternative. These folks wouldn’t need driver’s insurance, or even a driver’s license.
There will be people who will need to own cars, such as anyone who needs a pickup truck to haul cargo; people who travel in rural areas where car share is not ubiquitous, the thrill of driving (see photo) and other cases . But totally autonomous car share has the potential to dramatically reduce the household demand for owned automobiles, millions of which sit idle 80% of the time. The automobile manufacturers aren’t panicking yet, but I’d expect to see increasing numbers of them pulling a “U-turn” as Toyota did, when it realized it was in danger of being left out of the autonomous car market completely.  If totally autonomous cars for hire take over just half the market, and reduce the number of cars sold for that half by 80%, the result will be far fewer car companies, approximately 40% fewer.  Only the manufacturers of the most efficient, safe, and reliable cars will survive.   And those in the uncanny valley will slowly disappear as well, driven by the lower costs.Model T