Lab Vision

Joseph Jay Williams is an Assistant Professor of Computer Science at the University of Toronto (by courtesy, Psychology and Statistics), and a faculty affiliate at the Vector Institute for AI

How can you turn any user interface into an intelligent, perpetually improving system?

Our answer is to use adaptive experiments that quantify the impact of alternative interventions on people, combining human intelligence (collaborations between designers, scientists, and users) with artificial intelligence - machine learning algorithms and statistical tests that automatically use data from experiments to enhance the interventions future users receive.

Our research agenda is to transform components of any real-world user interface into intelligent, adaptive systems that are perpetually enhancing and personalizing interventions to help people. For example, we've published papers on technology for education, learning, and mental health, by testing competing ideas about how to design components of online homework, apps, text messaging interventions, and other interface components. 

A distinctive focus is micro-experimentation 'in the wild', where we conduct and build tools for randomized A/B comparisons that can be used by practitioners and scientists to test design decisions and hypotheses about how to help people. Another focus is adaptive experimentation – applying and modifying machine learning algorithms and statistical tests for more rapidly yet reliably using data from experiments to change which interventions future participants receive. 

I use my cognitive science theories in deciding the target actions for experimentation (e.g. explanations, motivational messages) and the metrics to optimize (e.g. student ratings, response rates). To generate new actions I design crowdsourcing workflows, leveraging my human-computer interaction research. 

Data from experiments is analyzed using methods from Bayesian statistics, and algorithms from machine learning are used to turn data into dynamic enhancement and personalization of users' experiences. My self-improving systems are powered by combining human intelligence – in generating hypotheses that can be tested with data – with statistical machine learning – to automate rapid iteration and improvement.

One example of my research is a system I created for automatically experimenting with explanations, which enhanced learning from math problems as much as an expert instructor [LAS 2016]. Another system boosted people's responses to an email campaign, by dynamically discovering how to personalize motivational messages to a user's activity level [EDM 2015].

My TEDxPortofSpain talk explains how I use this approach in education, using MOOClets to intelligently adapt explanations for how to solve math problems. 

Slides and recordings of overview talks can be found on the Representative Talks page.

More about my research

In blending scientific research and applications I collaborate and consult with domain/topic experts, instructors, designers, and researchers from diverse disciplines like education, psychology, and computer science

I draw on theories and methodology from research I have done, as well as synthesizing findings from other scientists, work on behavior change, reviews of evidence-based best practices for teaching and learning, practical experience as a statistics tutor, evaluations of educational technology products & authoring tools for e-learning, experience as an ed-tech consultant and science & technology advisor, and my own active development of interactive learning and behavior change resources