And that data may be seriously flawed, according to a recent report in the Wall Street Journal.
Among the most cringe-inducing databases and AI programs is one that identified Africans as gorillas. And one AI program being tested by Amazon.com systematically rejected any resumes including the word “women’s” — as in women’s colleges or women’s professional organizations. In other cases, dark-skinned women were identified as men.
What’s the problem? The US Bureau of Labor Statistics says that programmers who write artificial intelligence programs are mostly white and male. Women and minority groups are not adequately represented. And a lack of diversity among programmers results in a lack of diversity in datasets. For example, one widely used dataset is more than 74% male and 83% white, meaning algorithms based on this data could have blind spots or biases built in.
The solution is fairly straight forward: eliminate bias upfront among the “guys” writing the code. If want diversity in data, you need diversity in the team designing the algorithm, commented Affectiva co-founder Rana el Kaliouby.
You’ve heard it before, and we’re sure you’ll hear it again: we need more women and minorities in computer science. Unfortunately, achieving that goal remains elusive.