Connected Spaces: Responsible AI
Because Microsoft is committed to building AI responsibly, the Connected Spaces product team collaborated with Ethics & Society to uphold our responsible AI principles from the start. As an internal product and design agency, Ethics & Society guides technical and experience innovation towards ethical, responsible, and sustainable outcomes. Together, we were able to deep dive into key considerations around privacy and transparency, build a UX that gives store employees more control over the AI system, and leverage cross-disciplinary expertise to help us achieve better outcomes for individuals and companies.
Initiated
Early on in our product life-cycle, I identified that there were many risks of building a computer vision based system. Our team was new to thinking of this problem space and didn’t have the expertise needed to think through all the challenges. I asked the Ethics & Society team to host an initial workshop for us, kicked off initial collaborative research studies, and brainstormed ways we could work together long-term. This eventually expanded into a 6 month depth engagement with buy-in from leadership on both teams.
Engaged & Activated
I drove the depth Ethics & Society engagement from the product’s design-side. It was essential for me to ramp the Ethics & Society team up on product truth, shape which focus areas would make impact, and give timely feedback on any progress progress. My consistent drumbeat was the guidance that “if we were able to immediately implement all the things learned through the engagement, we hadn’t pushed hard enough.” It was essential to balance the need for concrete artifacts with the opportunity to push boundaries. Once we solidified our focus areas and built momentum, we were able to activate three work streams across our product team feature crews.
Shipped
Meaningful Human Control
With the variability in environmental factors across stores, we created mechanisms for calibration, configuration, and monitoring for users to adjust the AI system to their preferences and configure for their unique environment.
Outcome: Shipped pieces of the design and influenced the product roadmap
Model Documentation
We conducted research to gather needs of internal Machine Learning developers to serve as a foundation for future extensibility plans. This also helped streamline the existing team’s communication methods.
Outcome: Repeatable research methodology to learn from ML developers
Evangelized
During this engagement, we had key learnings that we thought would be beneficial for internal and external teams to learn from. We hosted talks at internal Machine Learning & Design conferences and published a Medium article.