Please reload


Please reload

AI Products in a Human World



The State of California allows car manufacturers to test their autonomous vehicles (AV) on public roads through an application process handled by the Department of Motor Vehicles (DMV).  One of the requirements is that reports of all AV accidents be submitted to DMV so that it can made available to the public [1].  A recent analysis of the AV accident rate in California found the following [2]:


“It was found that rear-end collisions, with the AV standing in front of the conventional vehicle, are the most frequent type of collision, happening with a frequency that doubles that of rear-end (fender-benders) for conventional cars.”

Surprising. Not really.  In his book, The Design of Everyday Things, Don Norman talks about the seemingly simple door and how to this day, we still see people who approach a door struggle to open it, not knowing whether to push or pull.  This is a design issue.  When a door (or any) design does not provide any visible clues as to its operation, people will interact with it using their current mental models. 


What does this mean for the introduction of AV into the general driving public?  Well, it means that there are now a small group of vehicles (AV) operating in a different way than most of the other vehicles with human drivers who have a certain mental model of how cars will generally behave.  And when there is a different behavior encountered by the human driven vehicle than the norm then an accident can occur. 


Now low-speed rear-end accidents are not serious but these accidents do reveal the challenge of trying to anticipate the safety of products with intelligence and autonomy operating in a world of humans.



Please reload

Recent Posts

Please reload

©2017 by Consortium for Safer AI.