As Artificial Intelligence (AI) and machine learning (ML) grow increasingly accurate, they simultaneously grow more complex, threatening to undermine their progress in fields such as healthcare, law and finance because of a lack of transparency and trust. In an effort to contribute to the trustworthiness of AI systems, the National Science Foundation (NSF) and Amazon partnered together in 2019 to start funding projects through the Fairness in Artificial Intelligence program. Since the program first launched, 21 projects have been funded, including “Towards Adaptive and Interactive Post Hoc Explanations,” a collaboration between UCI, Harvard University and the University of Chicago.
The project aims to develop a novel framework for generating adaptive explanations of ML models, allowing for interactive communication, to help people better understand the decisions being made for and about them.
“UCI will be focused on creating explanations that adapt to the user’s need — that is, automatically creating custom explanations depending on what subgroups they are interested in, or what their level of expertise is in the ML application,” says Sameer Singh, assistant professor of computer science in the Donald Bren School of Information and Computer Sciences (ICS).
The team will be focusing on multiple real-world application domains. For example, they plan to explore two healthcare applications: one for the early detection of seizures and another for diagnosing tumors. “For each, we will create ML explanation systems that aid medical professionals in their decision-making, adapting to their queries, domain expertise and history of interactions,” says Singh.
The work builds on previous collaborations between Singh and Assistant Professor Hima Lakkaraju of Harvard University. “Hima and I have been collaborating on papers on machine learning explanations for a while,” notes Singh. “However, one aspect we were missing were explanations that actually adapt, interact and customize to each user, which would make them much more useful.” Singh is excited that they are now partnering with Professors Chenhao Tan and Yuxin Chen from the University of Chicago to further explore this important area.
— Shani Murray