Junpeng Lao

I am a psychologist, neuroscientist, and Bayesian statistician.

During my PhD, I studied how culture shapes human visual perception and the neural processes behind it, using eye-tracking, EEG, and fMRI. My deepest interest is how people combine prior experience with new information when they reason about the world. That question drew me into Bayesian inference, first as a practical tool for analysing data, and then as a way to think about how beliefs change when new evidence arrives.

Over time, that interest grew into broader questions about knowledge and epistemology: how knowledge is formed, how we judge whether it is reliable, and how one scientist can build on what another has learned. This is now a big part of my work as Founding Scientist and Chief Epistemologist at rekursiv.ai, where I work on autonomous science discovery with AI. Research leaves behind more than a final result: experimental evidence, reasoning, new hypotheses, failures, and dead ends. My focus is building a process that integrates these artifacts into knowledge that helps human and AI scientists tackle the next question.

Before joining rekursiv.ai, I spent eight years at Google, most recently as a Staff Data Scientist. I worked on large-scale forecasting problems for Trust and Safety, where decisions had to be made from incomplete and constantly changing information.

I have also spent much of the past decade contributing to open-source Bayesian statistical software and the wider Python data science community. I am a core developer of BlackJAX (1.1k GitHub stars and around 220k downloads each month), a library for composable Bayesian inference in JAX, and I curate its ecosystem and community. I am also a core developer of PyMC (9.7k GitHub stars and around 1.8 million monthly downloads). I started the PyMC community on Discourse, where I spent years helping people use the library and think through modeling questions. I watched it grow from fewer than ten people to more than 4,000 members, and it now receives over 300,000 human page views each month.

I learned a lot about mentoring and scientific collaboration through this wider community. It led to work in chemistry, toxicology, and cosmology, and to co-authoring Bayesian Modeling and Computation in Python with Osvaldo Martin and Ravin Kumar. At Google, mentoring became a more deliberate part of my job when I managed a small data science team. At rekursiv.ai, I am now working on how to teach AI to discover. The mentoring continues.