Will autonomous cars de-skill their human drivers? In a thoughtful presentation MIT researcher Bryan Reimer points out the dangers of letting cars drive themselves autonomously part of the time. As people rely on automated driving more, they drive less themselves and their experience shrinks which may make them more likely to err at the steering wheel. He also dismisses the idea that humans would be effective at monitoring an autonomous car’s actions and take over in difficult situations: Besides having to be constantly alert, they would need a much deeper understanding of the autonomous car’s capabilities and limitations to be effective in such situations.
These are important insights for the evolution of autonomous vehicles. They have direct implications on the way that driverless vehicles are conceptualized and for the legal frameworks. Current driverless car laws are are based on the idea that a human is in control or should be able to take over immediately in critical situations. The reality will be different. The laws will need to address truly autonomous operation (where no occupant can be held liable for the car’s operation).
Reimer proposes to increase human-centered research and developoment to improve the interface between driver and autonomous vehicle. But it is hard to see how this could overcome the dilemma he has sketched. Improving the autonomous capabilities of these cars to the point where they perform verifiably better than almost all human drivers seems to be the only realistic alternative.