In the wake of the Tesla accident that resulted in the first fatality for self driving car technology, there was a USA Today article on self driving cars that presented some new views and options on how to prevent these crashes in the future. While the article makes some really good point, there are still some items that I want to address in order to provides a data science perspective.
I have already written on self driving cars and the inevitable accidents that will happen. The USA Today article ignores a lot of the data science technology that make these cars possible. I can only assume that this is because (1) the author doesn’t fully understand it or (2) the truthful answer is just not something that we are accustomed to.
Neither the other truck, Tesla or even Harry Potter killed this man. An algorithm killed this man. How does a math formula kill someone?
Think about that for a second – how do you tell someone that a death happened because of a math formula? That’s just not something that is in our collective consciousness and we will have to get more accustomed to that if we are a society that is going to proceed with more and more data science automation in our lives.
The articles goes on to present vehicle-to-vehicle communication technology that should help eliminate these accidents. In fact, they cite the 35,000 plus motor vehicle fatalities in the U.S. that can be eliminated once we perfect automation along with vehicle-to-vehicle communication. I believe that both of these views are mis-guided and unattainable.
Vehicle-to-vehicle communication can indeed have a improved effect on vehicle automation. But on what scale and how long will it take to achieve those scales? Does the political will and finance exist to keep those plans alive for the long term? The infrastructure and communication across multiple government agencies that it would take to manage that is almost mind-boggling. Take the example of a road intersection that leads to an interstate on-ramp. The signage may be funded by the city, the roads are a combination of funds from the city, county and state and then the interstate on-ramp is most likely the same combination of funds, with the addition of federal funds. So now you have a large collection of agencies that have to work together to bring this technology to just one intersection. That’s a large effort that is measured in multiple years, not months.
A 4,500 pound car traveling at 65 MPH is always a dangerous proposition, no matter whether the car is automated or not.
Finally, we have to provide ourselves with a realistic view of what self driving vehicles will do to the vehicle accident rates in the United States. I think it’s reasonable to expect a reduction in the 35,000 (approx.) deaths per year, but unreasonable to expect an elimination of this number. We have to be prepared for scenarios where we are just trading one type of accident (say, sleepy drivers) for another type of accident (say, automated cars with faulty sensor readings in rainy conditions). We have come to a point in (relative) vehicle safety where we overlook the fact that we are still in massive metal objects of 4,500+ pounds moving at high rates of speed – the physics of that can lead to accidents regardless of who/what is driving your vehicle. Who’s to say that the accident rate won’t actually rise for a period? The cognitive process that cars have to go through in order to learn how to drive make both positive and negative outcomes a part of the data science process. The vehicle death rates may not be a straight line down once automate cars are more popular; the death rate may actually fluctuate over time for any number of years, or even decades, to come.
We are living in an interesting period for data science as more and more cognitive technology will enter our lives. It’s possible that “death by algorithm” will enter our collective consciousness and that will, in turn, reduce the shock of self driving car fatalities. But the time between now and then (if that time ever comes) will be an interesting period for everyone and will play a big role in the role that cognitive technologies will play in the world.