We have all read stories of facial recognition software that fails to recognize dark-skinned faces, or robo-loan officers that deny mortgages to certain groups. As a growing body of research has made clear, algorithms created by non-representative groups have resulted in AI that perpetuates the inequities already prevalent in our society. As more companies rely more heavily on data and AI, these problems of algorithmic discrimination may only become worse.
Most companies know this by now. What they’re trying to figure out is: how can they avoid becoming yet another bad example?
The short answer is, thinking critically about the data you’re collecting and how you’re using it needs to be everyone’s job. Expanding the circle of who is in the room helping to question, build, and monitor algorithms is the only way that we will develop responsible AI. Doing that work requires data literacy — the ability to parse and organize complex data, interpret and summarize information, develop predictions, or appreciate the ethical implications of algorithms. Like math, it can be learned in beginner and advanced modes, spans multiple disciplines, and is often more practical than academic.