A blog about the social impact of AI
18 September 2019
Karim Jebari

On the alleged safety of autonomous vehicles

An autonomous vehicle is a vehicle that can autonomously execute a high-level transportation task – for example, “get me from A to B” – after human instructions. Autonomous vehicles have been touted for a while as “the next big thing”, are said to bring about great benefits to society. Among these benefits, safety is often cited as the most important. The argument is that with autonomy, the number of accidents could be drastically reduced. Considering that vehicle-related accidents are a significant cause of mortality for young adults in most countries, this is a major consideration.

Many have accepted these claims at face value and have moved on to consider other problems, such as whether self-driving cars should embrace kantian or consequentialist ethics. Some philosophers argue even that regulation that could make autonomous vehicles safer should be delayed or not implemented, if it would also delay the deployment of autonomous vehicles on a mass scale, under the assumption that such a delay would cause many additional deaths. However, on closer inspection, these safety benefits seem uncertain, at least in the near term (10 years). 

While autonomous vehicles may be more safe than an average ordinary vehicles, this is not a relevant comparison. Rather, autonomous vehicles should be compared with the safest cars in their price class. Manufacturers of such cars are rapidly deploying AI technology to improve safety. Thus, autonomous vehicles are chasing a moving target with regards to traffic safety. 

The argument for superior safety in autonomous vehicles goes as follows. Human drivers have reaction times of 0.5 to 2 seconds. At high speeds, this time makes a significant difference in many potential collisions. For example, at 50 km/h, 1 second is about 14 meters. If a human driver with that reaction time would see a person 20 meters ahead in its path at that speed, it would be very difficult to avoid a collision. Most cars have a stopping sight distance of more than 6 meters at that speed. A machine has a reaction time that is orders of magnitude faster than most humans. A machine seeing a person on the road in this scenario (driving at 50 km/h) would immediately hit the brakes and likely avoid a collision, or at least prevent a deadly collision. 

The difference in reaction time accounts for a significant part of the difference in expected safety between an autonomous vehicles and a vehicle that relies on human reactions. But proponents of autonomy seem to be falsely assuming that only autonomous vehicles can make use of the speed of machines in the context of collisions. However, autonomous emergency braking is a feature that is already integrated in many new high-end car models. In the EU, legislation is being prepared to make it mandatory in all new cars from 2022[1] But automated emergency braking is not the only new technology that can greatly improve safety in new cars.

For example, according to the U.S Dept of Transportation’s National Motor Vehicle Crash Causation Survey, 6% of pedestrian fatalities are caused when drivers slow down when approaching a red light, and making a right turn while looking at their left.[2] This common error is to a large extent mitigated by a relatively novel feature; adaptive headlights, that shift the headlights in the direction where the driver is steering. 

About 7% of car crashes and 21% of fatal crashes (in the US) are caused by people being drowsy or falling asleep while driving. These accidents could be to a significant extent prevented by driver drowsiness detection software, which monitors eye and head movements and alerts the driver when these indicate dangerous levels of drowsiness. Adaptive cruise control is another feature that can adjust the distance between vehicles to prevent rear-end collisions, a common category of accidents when drivers are drowsy.

Other situations that are common in accidents involve vehicles running off the edge of the road and/or a failure to stay in the proper lane, also often in combination with drowsiness. This could be mitigated by lane assist controls, software that warns a driver if the vehicle is departing from the lane, and/or adjusts the vehicle course if the driver does not respond. This feature is already integrated in many of the newest car models. 

Combining these already existing automated safety features makes these cars much safer. While it may be true that a self-driving car will be safer than the average car in the US, that claim is not very interesting. The relevant comparison is between an autonomous vehicles and the safest and most expensive cars when autonomous vehicles are introduced, because they are likely to be at least as expensive. 

Finally, while traffic fatalities are important, we should not omit the fact that the number of fatalities caused by traffic related pollutants (ozone, NOX, bensen, particles, carbon monoxide etc) are an order of magnitude higher in Sweden (250 deaths/year vs 5000 deaths/year). In the US,  the number of transportation-related deaths caused by pollution is about double the number of deaths caused by traffic accidents.[3] Reducing pollution should be a priority. Unfortunately, autonomous vehicles are not likely to make any significant difference in this regard.

In conclusion, while autonomous vehicles may be safer than the average car in the US and EU, they are not likely to be safer than the newest and most expensive cars in the market in the near future. This means that we should disregard the safety argument for rapid implementation regulations that allows AVs on public roads.

footnotes

  1. Automakers agree to make auto braking a standard by 2022. Endgadget, Mariella Moon. 2016-03-19.
  2. U.S dept of transportation national motor vehicle crash causation survey
  3. Caiazzo, F et al. "Air pollution and early deaths in the United States." Atmospheric Environment Volume 79, November 2013, Pages 198-208

0 comments

Add comment