Enterprises today pursue AI applications to replace logic-based expert systems in order to learn from customer and operational signals. But letting AI learn from the wrong datasets can be disastrous for a company’s business and reputation. Because we are solving new problems, the right training data often does not yet exist, and because we don’t know the question we are asking, we can’t judge whether or not the answer we get back is right.
To overcome initial data limitations, teams can use existing knowledge of their business to jumpstart initial systems. Drawing on her experience working on enterprise recommendation and decisions support systems with Global 10 companies and the federal government, Elsie Kenyon explains how to harness institutional human knowledge to augment data in deployed AI solutions by keeping humans in the loop across all stages of AI projects, including experiment design, feature engineering, training data generation, and results evaluation.
Elsie Kenyon is a senior product manager at AI platform company Nara Logics, where she works with enterprise customers to define product needs and with engineers to build implementations that address them, with a focus on data processing and machine learning. Previously, Elsie was a researcher and casewriter at Harvard Business School. She holds a BA from Yale University.
©2017, O'Reilly Media, Inc. • (800) 889-8969 or (707) 827-7019 • Monday-Friday 7:30am-5pm PT • All trademarks and registered trademarks appearing on oreilly.com are the property of their respective owners. • firstname.lastname@example.org