Tesla response to fatal accident

July 1, 2016

AI & Humans
I remember hearing a while back from some tech people that the first time an automatic driving car kills someone, it will be a huge deal.

This is because all of us will expect an AI to not just be as good as a human driver, or slightly better, but exponentially safer.

Regarding that point, I just heard the news that there was the first known fatality for a driver in an automatic driving Tesla.

I want to point you towards the Tesla blog post responding to the accident.

A quick point before that.
One thing I’m learning to do more often is that we should always ask “compared to what” when hearing any type of statistic.

If a new prescription drug is touted for its results, we compare it with a placebo. If someone says we’re paying too much for health care in our country, we would calculate it as a percentage of GDP and compare it with other industrialized nations. If the stock price of a company has risen 15% during the year, we would compare it with the S&P500. If the S&P rose 35% during the same time, then our stock is not doing so hot.

With fatalities caused by AI driven cars, we compare it with human drivers. Tesla vehicle computers collect data so that a black box type examination can be conducted.

From Tesla:

This is the first known fatality in just over 130 million miles where Autopilot was activated. Among all vehicles in the US, there is a fatality every 94 million miles. Worldwide, there is a fatality approximately every 60 million miles.

That means the AI has been able to drive 38% more miles before a fatality compared to American drivers, and 116% more than drivers worldwide.

When looking at all types of accidents (not just fatal), Elon Musk has stated previously:

“The probability of having an accident is 50% lower if you have Autopilot on. Even with our first version. So we can see basically what’s the average number of kilometers to an accident – accident defined by airbag deployment. Even with this early version, it’s almost twice as good as a person.”

Mathematically this is a huge improvement, but the general public will demand almost fail-proof AI before trusting a computer to drive for us. The technology will get better, but we can expect many more bumps along the way.

Resources:
A Tragic Loss – Tesla Blog
Tesla’s Autopilot lowers probability of having an accident by 50% based on early data, says Musk

Update:
Joshua Brown, Who Died in Self-Driving Accident, Tested Limits of His Tesla – NYTimes