click to enlarge
Composite from Bigstock images
It's the early 1990s. Imagine a U.S. high-tech company has made a quantum leap in face recognition software. The product is so far ahead of the competition, it corners the market and becomes the industry standard.
In this imaginary scenario, the company is owned and staffed by African American techies.
What are the chances the weakest link in the software would be its inability to recognize dark-skinned faces? Almost zero, right?
When early incarnations of the software were tried out on the company staff, any problems with differentiating and recognizing people with dark skin would jump out at them immediately. The company would make it a priority to pinpoint and correct the problems, because who wants to create a program where they, their coworkers, their mothers, fathers, boyfriends, girlfriends, wives, husbands, sons and daughters can be mistaken for someone else? The recognition system for dark-skinned faces would be as close to flawless as the company could make it.
Of course, before bringing the product to market, the company would make sure the system worked with lighter skinned faces as well, because that's where the money is. Still, its accuracy might drop a bit because Anglos were excluded from the initial testing and tweaking process, and they weren't in the room prodding their fellow techies to work out the bugs.
But if, instead of it being an all-African American company, a quarter of the top people were Anglo, white faces would have gotten more attention from the beginning. If you add Asians and people from other ethnic groups into the mix — and women, let's not forget about women — you're likely to have as close to an equal opportunity face recognition product as any company could create.
Now let's jump back into the real world. One of the weakest links in today's face recognition software is recognizing darker faces. In a grotesque example, in 2015 when Google users asked to see images of gorillas, some African American faces were likely to be included. Google apologized and promised to fix the problem. Instead of fine-tuning the software though, Google simply blocked gorillas from its image search options to take care of that one specific problem.
An article in Thursday's Star is about Amazon's face recognition software which appears to have trouble
detecting African American women's faces. As you might imagine, the problem wasn't discovered in house. It was pointed out by a female, African American researcher at MIT.
Both Google and Amazon tried to make excuses for their errors. A Google spokesperson said its software made mistakes with other people's faces as well. Amazon attacked the methodology the MIT researcher used in her recent study, though a group of Artificial Intelligence scholars defended her work. Both companies thought it was perfectly fine to downplay the problem and minimize the effort they put into finding solutions.
Problems like these arise regularly in the private and the public sector because too few members of minority groups are sitting at the table. If Silicon Valley had more African Americans in positions of power, they could throw a spotlight on racially insensitive aspects of the corporation's work, which are invisible or unimportant to others, and work to find solutions to the problems.
Here's another recent example, or two or three, from Hollywood.