Katie Bakewell -NLP Logix
Artificial Intelligence and Machine Learning are making their way into the mainstream for Debt Collection. AI models can predict which consumers are most likely to pay, the channel that is most likely to influence a consumer, identify potential agent strengths, and monitor compliance from speech data, all with relatively high accuracy. You would not employ a model that determined who paid based on race or gender, but what if the model you utilize implicitly segments by national origin? In addition to the negative publicity protected class infringement can cause, implicit bias can be costly - recent lawsuits based on settlement letters sent with an unseen bias have resulted in settlements costing millions of dollars. The goal of this session is to inform the industry on safe, compliant use of machine learning. To achieve this goal, this session will: Look at the potential dangers of employing machine learning models and ways to avoid them. Discuss different model types, from completely transparent to black box, and how they affect usability. Walk through how to measure the level of implicit bias across protected classes in ML models and strategies for mitigation.