All AI is Biased: 4 Ways to Fix It
MIT’s Renée Richardson Gosline debunks the idea of AI neutrality: Adding friction — and — humans can be a cure
By Paula Klein
In the rush to adopt digital technologies, marketing leaders must also recognize the importance — and challenges — of behavioral science and the study of human nature.
In her presentation at the recent CMO Summit@MIT, Renée Richardson Gosline, IDE Human/AI Interface Research Group Lead, explained the intersection of AI and cognitive bias and how bias can be introduced when AI is used for decision-making and media campaigns.
One big buzzword of digital marketing transformation is reducing customer “friction” points by using AI and automation to ease the customer experience. “In fact, it has become quite a popular and robust approach,” Gosline said, “and we’ve all caught what I like to call frictionless fever.” Retailers such as Aldi, Hudson, and Amazon Go, “allow you to skip the checkout and use facial recognition or hand-scans as your wallet. The future of investing is frictionless and autonomous,” too, she noted.
Yet, most executives don’t understand what’s going on with AI, Gosline maintains. They don’t understand that 100% of AI is biased.
“Since all AI is trained on historical data, and humans program AI, and organizations have very clear goals, AI is not neutral; it is biased toward those goals and toward what has come in the past,” she said.
Gosline suggests that marketers have “avoided friction at their own peril. Have we adopted all of this AI and this automation and lost the central human behind the data points?”
Going forward, she said that
“rather than thinking about friction as bad and something to be eradicated, we also should think about when friction could actually be helpful” and when adding friction in the use of AI could improve trust and reduce harm and bias.
Gosline presented research that demonstrates the relationship between cognitive style and trust in AI versus humans. “We need to think about who is using algorithms and when it makes sense for us to potentially add friction to interrupt the automatic and potentially uncritical use of algorithms.”
Four Considerations for Better AI Outcomes
At the IDE-sponsored CMO Summit@MIT, Gosline offered four points to consider before introducing AI tools and ‘good friction.’
“Adding these points of friction in our processes may help us center our customers and not just talk about customer centricity,” she told the CMO Summit attendees. In turn, that will increase trust and reduce harm. Gosline strongly believes “using AI in a way that’s human-centered as opposed to exploitative will be a true strategic advantage” for marketing. The considerations are:
1. Ask: Should AI even be doing this or can it be done better by humans?
2. Audit user journeys — whether they are customers or employees — to identify touchpoints that would benefit from the addition of friction.
3. Embrace inclusivity in terms of who’s developing these models and what data is being used to train these models. “Who is in the room when we decide who’s going to be affected by these models?”
4. Identify asymmetric friction. Don’t create a “lobster trap” that is frictionless to get in, but creates a lot of friction full to get out.