Working Up

by Dwayne Phillips

First you work on a small problem set. Once you learn from that, you expand the problem set.

Google recently started hiring speakers with accents to help train its speech recognition software systems. Why didn’t they do this sooner? Why did they only use middle-America, white-bread Americans, or some other Johnny Carson, no-accent accent cliche you like? Several years ago the creators of a a face-recognition system were embarrassed (actually scorned and derided) when their system failed to recognize faces of persons who didn’t fall into the middle-America…and so on. You get the picture?

Are tech companies run by racists who disdain those with accents or facial features they don’t like? I don’t think so. I think these companies are run by problem solvers who have some experience in attempting to solve difficult problems.

A common, experience-born approach is to attempt a subset of the entire problem set first. Why? Because it is easier. You learn. You don’t stumble so hard. You move on to a larger subset when you have an idea of what you are doing.

It appears unfortunate to some that middle-America, etc. persons—who are about one-third of the world’s population, a good subset—have about two-third’s of the world’s money. That is a public relations problem, not a technical problem.

Perhaps tech companies would be better served if they asked a public relations expert for some tips.