The Rapidly Increasing Scope for Software Quality Improvement

The nature of a computer is fundamentally different than that of any other manufactured device. No other device has software. This is a game changer.

For example, the headline in msnbc today reads:

Cars of today are essentially computers on wheels — and that’s a problem. Locating the source of electronic glitches is like “looking for a needle in a haystack.” Full story.

But the story doesn't mention another important consequence. In Alan Cooper's book The Inmates are Running the Asylum, he poses the following questions:

What do you get when you cross a computer with an airplane?
Answer: A computer.
What do you get when you cross a computer with a camera?
Answer: A computer.
What do you get when you cross a computer with an alarm clock?
Answer: A computer.
What do you get when you cross a computer with a car?
Answer: A computer.
What do you get when you cross a computer with a bank?
Answer: A computer.
What do you get when you cross a computer with a warship?
Answer: A computer.

The point Cooper is making is that software quality assurance (SQA) is becoming increasingly more important. And just because Toyota knows how to make a quality car, it does not mean they also know how to make a quality computer. And their failure to fully appreciate that cars are now "computers on wheels" is costing them enormously. Perhaps even the existence of the company.

This enormous risk potential is there for almost every manufacturer. Worse(?), science and even mathematics are becoming more and more dependent on computer software. For example:

What do you get when you cross climate science with a computer?
Answer: A computer.

Our confidence in climate forecasts depend in an ever increasingly fundamental way on how well they do their climate model software quality assurance.

2 comments:

  1. Worse(?)

    I don't think so, but I think Gelman's take,

    With great power comes great responsibility. [...] A Bayesian inference can create predictions of everything, and as a result you can be much more wrong as a Bayesian than as a classical statistician.

    applies here in a slightly modified form:

    With a physical simulation you can create predictions of everything, and as a result you can be much more wrong as a computational physicist than as a classical physicist.

    ReplyDelete