I’ve spent the past few months collaborating with others on the UKAI trade association national industry guide “AI’s Impact on People: Protecting Workforce, Customer and Brand.” The guide will be launched this autumn.
In the meantime, I’ve been thinking about how much of the discussion on bias is about addressing the bias in AI models and working to be sure that it isn’t allowed to proliferate through agents or automation. It’s also important to realise that there’s already bias in our systems. (This is why it’s baked into AI models in the first place.)
I spent more than 10 years working at Mercer. One of the major initiatives led by the EVP I worked for was “When Women Thrive”. This was an initiative in partnership with the World Economic Forum. The goal was to address gender bias across multiple factors. Women in work, in education, in government, in health, in financial security. There were individual countries and individual organisations that made progress, but the overall picture was depressingly consistent. It would take more than 100 years at the current rate of change to reach real equality.
I also had an experience a couple of weeks ago at an event where I was surprised at the comfort some people have with expressing racist views. There was the one overtly racist individual, who I called out and condemned. There were the more subtle comments, though. There were comments that are based in unconscious bias. There are a lot of people who have never addressed the biases that they have absorbed throughout their life. This doesn’t make them bad people, but it’s the result of having lived through the recent past in certain places and never having thought critically about it.
As we think about how to address bias in the context of AI adoption, it’s important not just to think about the new bias introduced with AI models. There is bias already living in existing operating models. Unless something is designed with the explicit intention of countering bias, it is likely to be vulnerable to unconscious bias.
Many organisations have worked on their hiring, promotion, and compensation processes and structures to explicitly address bias, but it can sneak in other places.
Have you thought about how goals are set, how work is allocated across team members, what is and isn’t measured, or who picks up ad hoc work? Maybe you have and everything is great, but if you haven’t, it’s important to look carefully at what you have in place today before you point AI at it.