The Business Integrity Leadership Initiative welcomed Alyssa Simpson Rochwerger as a guest speaker for the Let’s Talk About Ethical Artificial Intelligence program. Alyssa Simpson Rochwerger is the co-author of the program's book, Real World AI. Rochwerger serves as the Director of Product Management for Blue Shield of California and has over a decade of experience in building data-driven products in numerous leadership roles, including at IBM Watson as its Director of Product.
Rochwerger began her presentation by explaining that she did not join the AI space with a technical background, but rather she majored in American Studies during college and her background was in liberal arts. She noted that a piece of advice she received during a flight from London to LA inspired her to take an opportunity to work in a division outside of the one she had been working at IBM. The advice was to “Go solve the hard problems, everything else will sort itself out.”
Shortly after that, she began working on IBM’s Watson portfolio which features business-ready tools, applications, and solutions designed to make AI adoption responsible and successful. It was through this role that she started seeing risks in the design and implementation stages of development that threatened the success of AI programs. She jumped in on the product management side and started addressing the risks in a practical, cross-functional way.
As Rochwerger noted, AI isn’t new. It was first introduced in the 1950s and has evolved significantly since that time, to include subsets such as machine learning and deep learning. She also discussed the importance of reliable, unbiased business data in the AI design phase. For machine learning she described “good data” as data that contains thousands or millions of examples, and data that is organized and labeled into categories or outcomes with diverse viewpoints included.
Rochwerger discussed examples from her book of AI programs that missed the mark, where their creators went wrong, and what ethical risks any business person should be aware of. Some of these risks include not having enough guardrails in place to prevent bias from skewing your data set and algorithmic model drift, which happens when a model gets used in the real world and its accuracy declines over time. To avoid these issues, one must be aware of potential gender, racial, ethnic, and age bias (to name a few) that could be hiding within the data your model is being trained on, as well as continually refreshing that training data to ensure optimal model performance so it doesn’t become outdated.
Lastly, Rochwerger explained how AI teams are built and what components make them great. It is necessary for the technical team members and business team members to collaborate together and have the same goals and incentives so everyone stays aligned. According to Rochwerger, the ideal team environment embraces psychological safety, trust, humor, constructive feedback, honesty, and cross-functionality. Even if you’re working with AI, your team is more dynamic and successful when you know your collaborators as humans, and you can be open and vulnerable with one another. That’s how trust is created between team members, and what allows the team to perform at an optimal level.
