In Cannes this week, IBM is showing off an AI system that purports to address bias in advertising. The system IBM uses deploys 75 fairness metrics and 13 algorithms that help uncover bias in advertising.
In theory, everyone is against bias. The dictionary defines bias as “prejudice in favor or against one thing, person or group compared with another, usually in a way considered to be unfair.”
Yet bias isn’t always bad. The Positivity Bias operates in such way that we recall only the positive aspects of a memory, like remembering the good aspects of a first date rather than the nervousness and awkwardness of it.
When it comes to advertising, though, bias is hard to shake. If you’re selling a razor, you will naturally target men, but how? An interesting comparison on YouTube compares a Gillette ad from 1989 with one from 2019. The older ad has no ambiguity and presents manhood as something to be proud of. The latter shows men absorbing the news about allegations of sexual assault and then features men reaching out to be peacemakers.
In other words, the bias has shifted dramatically since 1989.
It’s possible that a well-maintained AI system can stay on top of such shifting biases, but such maintenance will likely always need a human checking over content on pretty much a daily basis to ensure that no examples of bias slip though. That’s because for advertising to work well, it has to have a strong grasp of the way consumers are thinking. That grasp can be aided by computers, but no supplanted by them -- at least, not yet.