Abstract: 

In many contexts, an AI is being to support a human decision maker (rather than making a decision on its own). For example, an AI may provide a clinician with treatment recommendations for a particular patient. In these settings, the hope is that human+AI team makes better decisions than either human or AI alone.

However, this complementarity is rarely reached. People will often over rely on AI systems, even when they are wrong; they may also under rely on AI systems when they are right.

In this short talk, I will cover some principles from our research on building AI systems that provide effective decision support. I will discuss how the right kinds of explanations can help people determine how to use and trust an AI. I will also discuss how the right support varies based on the context---not only the person's task, but also factors such as whether the task must be performed quickly. It also varies based on the qualities of the person, such as their tendency to over rely on AIs.

While my work is motivated by applications for decision support in healthcare, these principles apply broadly to many contexts in which an AI is providing information to help a person make a decision.

Bio: 

Finale Doshi-Velez is a Herchel Smith Professor in Computer Science at the Harvard Paulson School of Engineering and Applied Sciences. She completed her MSc from the University of Cambridge as a Marshall Scholar, her PhD from MIT, and her postdoc at Harvard Medical School. Her interests lie at the intersection of machine learning, healthcare, and interpretability.