Could The Biggest Problem With Self Driving Cars Be The Engineers Developing Them?

Cirrus Aircraft's SR-22 is the most popular single-engine piston plane. It has a technically advanced "glass" cockpit -- advanced avionics, benign handling, and lots of safety features. And yet, a few years ago, they had a troubling safety record.

It turns out that a lot of rich guys with poor flying skills were buying SR-22s. They had the cash to buy the plane, but not the skills to safely operate it, and therefore, the poor safety record.

The solution was not to throw more tech at the already tech-rich planes. It was to offer really good pilot training programs, and make them accessible to new Cirrus pilots, and to offer recurrent training programs for existing pilots. A few years after the new training programs were put in place, Cirrus enjoyed an enviable safety record.

The same applies to cars. If your lack of driving skill and your dependency on automation is part of your "way," then that's a setup for risk and bad experiences when the automation fails, or when it's unavailable to you. That's what happened to Vahid.

On a separate note, I doubt that a programmer's driving skills play any part in how good the autonomous software is in a car. That's because the programmer codes according to specifications, which he likely didn't write. He does NOT write code according to his driving style, his understandings of road rules, driving etiquette, etc.

One of the problems the engineers face -- the guys and gals defining and designing autonomous driving system -- is the conflicting instructions they get, such as (1) make it flawless, (2) Your budget is $X, and (3) you have to be finished by {Y-date}. There's a saying in the software business: "There's good, there's fast, and there's cheap -- you can pick only two." These real-world constraints come from the business unit or "head office." The engineers do not define them. They'd rather take their time and pull off "amazing."

I think the title nailed the irony.