AI isn't biased, but you might be
We've all seen the headlines suggesting that AI is racist and sexist. However, many people overlook one important fact -- that AI is simply a tool, incapable of being inherently biased. That’s not to say AI isn’t capable of producing biased outcomes -- as the headlines show, it certainly is. But it can only ever be as biased as the data upon which it relies.
So how can developers and marketers avoid deploying biased AI? Unfortunately, there is no magic one-size-fits-all solution. As with any successful technology deployment within a business, it requires a thorough understanding of the datasets you are working with, and the outcomes AI can produce with said data. The first step is knowing what to look for.
Moving away from demographic data
In order to avoid AI bias, it is paramount that developers and marketers do not allow AI-driven outcomes to be influenced by data which reveals any kind of demographic pattern. The obvious data points to exclude are of course categories such as gender, ethnicity, age, sexuality and geographic location. But the issue is far more complex than this.
Take the retail industry for example. Technology deployment in the sector skyrocketed during the pandemic as retailers rushed to move sales online, while highstreets around the world were forced to close their doors to customers. In order to maintain a high level of customer experience in the digital space, many businesses are adopting AI-driven personalization.
When done right, personalization can significantly decrease sales lead times, while supercharging customer loyalty and engagement. However, if not implemented properly, there is a risk of creating unintentional bias. For example, on the one hand AI-powered personalization can analyze data collected from your browsing and purchasing history to recommend the next product you might want to buy online. But on the other hand, the AI can use the same data to unintentionally provide different outcomes for clients based on demographic patterns, thus discriminating against certain demographic groups.
Again, this is not because AI is inherently biased: it isn’t capable of comprehending wealth or consciously discriminating against people based on their affluence. Instead, it is because AI is trained to identify patterns and produce outcomes which encourage more sales. Even with overtly demographic date omitted, data correlating with demographic patterns can still be identified and used to influence AI’s decisions.
In this instance, AI may observe that some shoppers spend more time browsing but ultimately make fewer purchasing decisions. In comparison, it could determine that others spend less time browsing and yet make more purchases. AI is then capable of offering greater discounts to slow browsers to encourage more sales, while holding back on offers to faster shoppers to protect profit margins.
What AI isn’t capable of in this situation, is understanding that faster browsing times and making more purchases are behaviors often linked to more wealthy clients. Meanwhile, those who spend more time researching products, and who are ultimately less inclined to rush or complete a purchase, are often less affluent. Without having the consciousness to understand the implications of the outcomes it produces, AI inadvertently makes an ecommerce site guilty of price discrimination.
The solution
To avoid AI bias, marketers and developers need to not only have a thorough understanding of the data used (or deliberately not used), but also the patterns that data may correlate with. Ultimately, however accurate or effective your AI may be, to remain ethical -- and to avoid a PR disaster -- it is paramount that it is not influenced by any kind of demographic pattern. And this requires human skill.
But with 37 percent of UK marketers suffer from a talent shortage in AI and machine learning, the industry is in the midst a skills gap, presenting a significant challenge. To tackle this head on, companies need to embrace automation technologies in other areas, giving teams more bandwidth to focus on more complex, high value tasks -- like deciphering data.
At a time when businesses are relying on online sales more than ever, understanding your customers and their data is only going to become more important. Utilizing AI to personalize content for customers is undoubtedly invaluable in recreating some of the positive customer experience lost by making sales virtual. To do this ethically, businesses must learn to identify and avoid datasets which unintentionally reveal demographic patterns.
Image Credit: Mopic / Shutterstock
Omer Artun is Chief Science Officer at digital experience specialist Acquia,