Balancing our relationship with AI

There is increasing recognition that many situations can benefit from humans and AI working together. AIs far surpass humans in finding patterns in large quantities of data, but especially in complex, real settings, humans tend to have a deeper, more nuanced understanding of the context and the objectives.

Interpretable machine learning—in which AIs make aspects of their decision-making process transparent to humans—has been touted as a way to enable better human+AI decisions. The explanations that come from an interpretable machine learning system are supposed to help calibrate the human’s trust in the AI, as well as empower them to perhaps identify a solution neither would have alone.

However, this promise is rarely achieved. The way in which the explanation is presented has a large effect in how it is used. For example, when given an AI recommendation, there is likely to be a confirmation bias: people will tend to look for reasons for the recommendation to be sensible, rather than consider all options. If an AI system performs well most of the time, people may no longer engage critically and begin to over-rely on the AI.

In this talk+open discussion/brown bag seminar, Prof Finale Doshi-Velez will describe some of the findings of her lab and related work on in the context of how people respond to AI recommendations and explanations, including ways to increase critical engagement. We will then broaden the discussion to implications relevant to broader decision-making and how to avoid these cognitive biases.

Prof Finale heads the Data to Actionable Knowledge group at Harvard Computer Science. 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.

The talk will be moderated by Dr Yiyun Shou, Lead scientist (Health and Lifestyle) at the Institute for the Public Understanding of Risk.