Some people think autonomous vehicles (AVs) must be nearly flawless before humans take their hands off the wheel. But RAND research shows that putting AVs on the road before they’re perfect improves the technology more quickly—and could save hundreds of thousands of lives over time.
Why Waiting for Perfect Autonomous Vehicles May Cost Lives
Michael DeKort is an engineer who is the Founder of Professionals for Safe and Responsible Autonomous Mobility. He’s also a member of the SAE On-Road Autonomous Driving Verification and Validation Task Force and a vocal critic of Tesla and Elon Musk’s approach to implementing autonomous vehicles. A month before the Rand article (linked above) was published he and I had a conversation on this very subject. Here is one of my comments to him.
Thanks for reaching out. I have read some of your articles and while I am not a technical expert, I believe I understand the gist of your argument.
Before I give my opinion, it might be helpful to provide you with a little background about my perspective. Yes, I am the president of a large Tesla enthusiasts club in Florida and not surprisingly I am an avid supporter of what Elon Musk is attempting to do in a number of areas, including autonomous vehicles.
Paraphrasing the Chinese proverb, we Floridians in particular happen to live in interesting times with regard to the implementation of AV. I have never thought of Florida legislators and transportation officials as being progressive in their outlook. However, a few years ago I had the opportunity to hear one of our Secretaries of Transportation speaking on the subject of AV and I was frankly surprised to hear how much they are promoting its development.
He went on to candidly state that due to the aging driving public that FDOT had essentially failed in its basic mission to reduce highway accidents and fatalities and they viewed AV as a means of reversing that trend. Likewise, the Chairman of the Florida Senate Transportation Committee and the Governor are also big proponents of AV. As a result, Florida has passed AV legislation that is the least restrictive in the country. I view that as a good thing in accelerating the development of AV, particularly for Tesla owners in Florida.
Last year, immediately following the death of Joshua Brown in that terrible crash here in Florida, I was interviewed by a couple of national news reporters. They wanted to know if I had changed my attitude regarding Tesla’s Autopilot. I told them no, that in fact I had a Model 3 on order with Autopilot and I had no intention of cancelling that order.
From early reports it appeared that Autopilot had encountered a situation in which it couldn’t “see” the truck. I told the reporters that even conceding that specific situation in which an engaged human would have done better than Autopilot, in terms of overall safety we need to consider the totality of driving situations. When using Autopilot as a driving assist feature, in my opinion there were going to be more situations whereby the system sees better, reacts faster, steers more accurately such that, in totality a Tesla with Autopilot engaged in conjunction with an engaged driver, is going to be safer than if it were turned off.
As you know, months later, after reviewing the statistics, the NHTSA reached the same conclusion.
I also told the reporters that Joshua was unfortunately too much of a Tesla enthusiast and was misguided in his trust of the technology. I explained that despite the questionable selection of the name, “Autopilot”, it definitely wasn’t self-driving and Joshua made a fatal mistake in treating it as if it were a fully autonomous vehicle. I said that a better name for the feature would have been “Driver Assist”.
Both the NHTSA and the NTSB also agreed with my initial assessment that the driver was at fault in his unfounded, over-confidence in the level of installed technology. After recently reading the NTSB report, I also agree with them that Tesla’s implementation of Autopilot, with limited driver warnings, played a major factor in the fatality.
Earlier this year I traded my four-year-old Tesla in for a new model with Autopilot. I also paid the additional $8,000 for the hardware and the eventual software for promised full autonomy. I have no illusions about the rudimentary current version of Autopilot. I use it exclusively in highway situations with no intersections. It excels in stop-and-go traffic. I don’t use it on exit ramps, or on secondary roads.
Last week while my wife and I were driving on the highway she asked me, “Is she [meaning Autopilot] driving?” My response was, “No, the three of us are driving.” What I meant was that before I upgraded my car to Autopilot, driving was a collaboration between my wife and myself. We are both over 70 and now at our stage of life it frankly helps to have more than one set of eyes on the road. So even before Autopilot I was already driving in Shadow Mode with my wife providing “Driver Assist”.
Returning to your basic question, even with the current basic version of the Autopilot software, I view it as a valuable safety feature when used as it was intended and for certain currently limited driving situations (highways for now). Inherent in this view is that the driver must take full responsibility for his/her actions and be fully engaged in their driving just as if Autopilot weren’t engaged. Some may say, “Well if I can’t relax behind the wheel what is the point of engaging Autopilot?” The answer is simple, there are going to be driving situations, especially for older drivers, whereby Autopilot is going to react faster than a human and mitigate an accident.
Some may say, “But won’t some people misuse the feature or fall asleep and die in fatal crashes along with their children?” Yes, absolutely. We can’t help that. The relevant question is, “When applied to the entire population of drivers, statistically is it safer to engage Autopilot or not? Experts can make recommendations to our legislators as to what that threshold should be, 2 times safer, 5 times safer, 10 times safer? However, everyone should recognize that many adults and children will die while we wait to implement “perfect” solutions.
Regarding AI Simulations versus Shadow Driving versus Test Tracks, I don’t view this as a pick-one-approach situation. I believe that multiple approaches should be used. As a layperson I am not qualified to speak to the ideal mix, but I believe that it is likely, especially here in Florida, that more lives will be lost by unduly deferring the implementation of AI by relying solely on simulations.
President, Florida Tesla Enthusiasts