Imperfect Perfect AI

Alex Gorton

January 28, 2020

A few weeks ago, I was in my hometown of Tempe, Arizona and saw a bizarre-looking car with multiple cameras perched atop its roof. It was cautiously making a right hand turn off of a main thoroughfare and moving at what seemed like a glacial pace at best. The word “Waymo” was prominently displayed on the passenger side door. Although I was surprised to see an autonomous car functioning in everyday traffic, the sight was common to the other passengers in my car.

Arizona has become home to test driving autonomous cars[1]. Companies such as Uber and Waymo have taken to the streets of Phoenix and its surrounding suburbs in an effort to test their algorithms. The governor of Arizona, Doug Ducey, has passed executive orders in an attempt to pave the way for companies to use Arizona’s roads to perfect their technology[2]. However, taking autonomous cars out of secure testing sites and putting them on roads occupied by other drivers, bicyclists, and pedestrians has come with a variety of safety issues. Last year, a pedestrian was killed by a driverless Uber while jaywalking outside of a designated crosswalk. After an investigation, it was determined that the car could not tell whether the victim was a pedestrian, bike or car, because jaywalking had not been factored into the car’s algorithm. Unfortunately, the safety driver who was also in the car at the time was distracted and not watching the road at the time of the crash. Authorities ultimately determined that Uber was not criminally responsible for the accident[3].

The case of the Arizona Uber accident raises multiple questions and requires us to take an honest look at artificial intelligence and the role that we choose to give it in our everyday lives. Although jaywalking is against traffic laws and, thereby, was not factored into the autonomous driving algorithm, it is a common occurrence on busy streets across America. In fact, each of us encounter identical multiple traffic violations everyday on our commutes. We’re human and sometimes, either deliberately or unconsciously—we break laws and we mess up. In response to these traffic faux paus drivers around us break, move aside, and sometimes out of exasperation give the occasional honk of their horn. These interactions are imperfect and yet we as humans accommodate.

The interactions between drivers in traffic highlights a disconnect in the way we view AI. There is a general misconception that AI removes human error. The reality is, however, much to our chagrin, that AI has faults that even perfect data cannot cure. We are coming to discover that despite our best efforts to create a fully automated world free of human error, AI will always require some sort of human oversight.

Hindsight is often 20/20, and in this case, it makes it easy to say that jaywalking should have been factored in to Uber’s algorithm. On the other hand, the fact that it wasn’t speaks volumes about the way we view AI. As a society, we have chosen to view computer based learning as superior to human judgment. It is precise and calculating. It does not factor in what we see as the weaknesses of the human condition—hesitation, emotion, and uncertainty, for example. We like to think that AI is the most objective form of decision making. But maybe what we think makes AI superior is really what makes it inferior.

We will never be able to avoid the human element of AI. Instead of viewing that as a detriment, perhaps we should view it as a benefit. AI has the ability to create a lot of positive impacts in our world, but we need to remember that our responsibility to one another cannot be off-loaded onto machines. In our quest for perfect AI, we have to remember that we are trying to use it in an imperfect world. And who knows imperfection better than us humans?

 

[1] Ryan Randazzo. “Waymo’s Driverless Cars On The Road: Cautious, Clunky, Impressive.” The Arizona Republic, December 5, 2018.

[2]  Daisuke Wakabayashi. “Self-Driving Uber Car Kills Pedestrian In Arizona, Where Robots Roam” The New York Times, March 18, 2018.

[3] Daisuke Wakabayashi. “Self-Driving Uber Car Kills Pedestrian In Arizona, Where Robots Roam” The New York Times, March 18, 2018.