Joseph Jay Williams

Adaptive Experimentation for Better AI Interfaces

My CV provides links to papers.

This year I'm a Visiting Scholar at UC Berkeley & Stanford.

I build AI systems that help people change their beliefs and behaviors for the better – in business settings, education, mental & physical health, and other domains.
I integrate AI and HCI (Human-Computer Interaction) to develop tools and methods for using Adaptive ("A/B/N") Experimentation to: (1) Transform everyday interfaces – text messages, online homework systems, and emails – into real-world AI agents. (2) Continuously enhance AI interactions with people.
Because AI can appear effective for some people yet fail others, I introduce workflows for generating alternative actions (A, B, … N) through co-design among LLMs, users, designers, and scientists. Reinforcement Learning algorithms adapt A/B experiments to learn which actions work, for whom, and when. This enables continuous testing and personalized improvement of interfaces to AI.
These methods enabled systems that coach like therapists via text messaging, tutor like teachers in online homework, and turn emails into persuasive agents for marketing. The work has impacted 500,000+ people.

I have published 85+ papers, resulting in 2 Best Paper awards, 7000+ citations, h-index of 44+. Top venues include CHI, NeurIPS, AAAI, Nature Human Behaviour, and PNAS. I led the teams awarded a $1M XPRIZE for the future of experimentation, and $3M in NSF funding – to create tools for Adaptive Experimentation.

My AI for Social Good Research Statement is at tiny.cc/williamsresearch. [Google Scholar]

See Education Focused CV, Research Statement, Health Focused CV, Research Statement.

Paper Highlights:
-Identifying 3 minute emails that boost students' performance on high-stakes tests by 4% [AC.34].
-Inventing the AdaptEx (Adaptive Experiment) framework [AC.29] to transform technology touchpoints used by billions of people (e.g. text messages, emails, websites, apps) into self-improving AI systems.
-Williams-Thompson Sampling Algorithm for accelerating discovery and practical impact in experiments.
-One Minute Interventions for exercise, managing stress, healthier smartphone use.

Disciplines Span: (1) Human-computer interaction; (2) Theoretical and applied work in social-behavioural sciences (e.g. experimental/cognitive/clinical/educational/health psychology, medical sciences, mental & physical health, education); (3) Applied GenAI (LLMs); (3) Applied Machine/Reinforcement Learning; (5) Applied Statistics.

Our 10 year target is to investigate how to scale access to adaptive experimentation techniques, and generalize to many kinds of personalized interventions: Millions of people and organizations can benefit from changing their behaviour.

Over 4 graduate students have gone into research positions (2 with faculty positions, 1 a research scientist, and 1 a prestigious independent postdoc at Stanford).

Biography

Joseph Jay Williams is director of the Intelligent Adaptive Interventions Lab. He is an Associate Professor with courtesy appointments at Rutgers and University of Toronto, and a visiting scholar at UC Berkeley and Stanford. He has appointments or supervises PhD students in Computer Science (Human-Computer Interaction, Applied AI, Reinforcement Learning), Statistical Science, Psychology, and the Vector Institute for Artificial Intelligence. He also has courtesy appointments in Economics, Industrial Engineering, Mathematics, & the Faculty of Information. His PhD Students span HCI (Human Computer Interaction), Cognitive/Social/Clinical/Health Psychology, applied ML (reinforcement learning), applied AI (LLMs), & Statistics.

Joseph was previously an Assistant Professor in Information Systems & Analytics at National University of Singapore, Research Scientist at Harvard, Postdoctoral Scholar at Stanford, and did his PhD at UC Berkeley. He is originally from Trinidad and Tobago.

Undergraduates & interested graduate students/postdocs interested in joining or collaborating, contact: iaiinterest@googlegroups.com.

Joseph can be contacted at
williams[at]cs[dot]toronto[dot]edu.

You can follow Joseph on LinkedIn, Facebook, Instagram, TikTok or any other social media.

Page updated

Report abuse