Holistic Data Used to Track AI Bias
Bias creeps into AI models as they are trained. Hence, experts warn that data sets and model training techniques must to be cleansed of bias at each step in the process. Among the tools promoted to eliminate bias are holistic data sets, so-called because they incorporate historical, out-of-domain and previously unused data.
Another approach involves what one vendor bills as an “ethical bias check” designed specifically to eliminate bias hidden in AI models used for customer engagement. Pegasystems Inc. (NASDAQ: PEGA) said its bias checker capability has been added to its Decision Hub platform as a way to flag discriminatory messages before they reach customers.
Unlike other bias detectors, Pegasystems said Tuesday (May 19) its approach simulates an entire customer engagement campaign across sales channels. Hence, all AI decisions can be screened simultaneously for bias, including online marketing pitches and email promotions.
The tool utilizes predictive analytics to simulate the likely outcome of a campaign, determining for example whether the audience for a specific pitch skews toward a particular demographic. Once alerted, users can pinpoint and adjust troublesome algorithms to eliminate bias.
The tool also allows users to set thresholds for elements such as age, gender or ethnicity. Those thresholds can be “slanted” for certain types of promotions such as healthcare providers pitching seniors about relevant Medicare services.
“High-profile incidents have made businesses increasingly aware of the risk of unintentional bias and its painful effect on customers,” said Rob Walker, Pegasystems’ vice president for decisioning and analytics. The monitor is intended to “empower businesses with tools that help reduce AI bias to improve how businesses interact with customers,” Walker added.
A Pegasystems’ study found that businesses still have a long way to go in using AI and other automation tools to improve customer engagement. Among the findings, the “AI and Empathy” study revealed that most customers believe businesses are more interested in sales revenue than service. Meanwhile, less than half of the 6,000 consumers surveyed remained unconvinced that AI can improve their dealings with retailers while human interaction is preferred over automated call centers and chatbots.
Hence, considerations like eliminating AI bias in customers relations represents one of the first steps in winning over consumers at a time when many traditional retailers are struggling to survive.
Along with predictive analytics, the Pegasystems AI platform uses “customer decision management” as a real-time recommendation system for each step in a customer interaction. It also includes a tool to provide greater transparency into AI models used for customer engagement, the company said.
