Alan Chan

Alan Chan

PhD Student



I am a PhD student at Mila, advised by Nicolas Le Roux. Previously at Amii + UAlberta, I was an MSc student with Martha White. My research interests broadly include value alignment and AI governance. Recently, I have been interested in interpretability and assessing the implicit knowledge in language models. Please feel free to reach out to chat about anything!


  • Value alignment
  • AI governance


  • PhD in Computer Science, 2024

    Université de Montréal

  • MSc in Computing Science, 2020

    University of Alberta

  • BSc in Mathematics, 2018

    University of Alberta


I am broadly interested in ensuring a just transition to the integration of AI systems into society, given the potential for exacerbation of existing harms and the creation of new ones. I approach this goal from a sociotechnical perspective: it is essential not only to reason about the mathematical properties of our systems, but also to construct governance frameworks to constrain them. My (mathematically) technical research revolves around value alignment, with a recent focus in using ideas from game theory and mechanism design to analyze and control algorithmic systems. My governance interests revolve around algorithmic regulation, antitrust, and global inequality in AI development.

Recent Papers

Loss of Control: "Normal Accidents" and AI Systems

A thread in recent work on the social impacts of AI systems is whether certain properties of a domain should preclude the application …

The Limits of Global Inclusion in AI Development

Those best-positioned to profit from the proliferation of artificial intelligence (AI) systems are those with the most economic power. …

Approaching Ethical Impacts in AI Research: The Role of a Graduate Student

Extant societal challenges and the problematic applications of algorithmic systems so far have motivated broader consideration of the …

Inverse Policy Evaluation for Value-based Sequential Decision-making

Value-based methods for reinforcement learning lack generally applicable ways to derive behavior from a value function. Many approaches …

Training Recurrent Neural Networks Online by Learning Explicit State Variables

Recurrent neural networks (RNNs) allow an agent to construct a state-representation from a stream of experience, which is essential in …

Recent & Upcoming Talks

MSc Seminar

My seminar for my MSc thesis.

Problems with Fair ML

This talk will be a mostly non-technical dive into problems that I find with a lot of fair ML research today. I will begin with some …

Recent Posts


One of the defining experiences of my intellectual development was the entirety of high school social studies, two years of which I spent with somebody who has shaped my thought the most critically.


I always struggle with the languages box on applications: what do I put for Cantonese and Mandarin? I’m sure that I don’t speak them fluently or even moderately well, but I’m not a beginner either.

Talking to Strangers

My last daily chat with a stranger was several months ago. Last year, motivated mostly by the positive experiences of a friend who had done a similar challenge, I decided to find somebody new—usually a student on campus—to talk to every day.

Coping and Learning

When I was in junior high school, I had already grown a shell of isolation. When we feel lonely, especially after a prolonged period of time, it is more difficult to reach out, even accept, the social interaction we need to warm the frigidity in our hearts.

Conflicts in Categorizing

Disclaimers I use the word category quite loosely in this piece. I do not refer to the mathematical field of Category Theory, although there may be some interesting, unexplored connections there.