Joëlle Pineau

Some people change the field of AI with a brilliant model, and some change it by forcing everyone to play fair. Joëlle Pineau is the second kind, though she also did the first. Born in Ottawa in 1974, she completed her doctorate in robotics at Carnegie Mellon, and has since dedicated her career to an unglamorous but deeply necessary goal: making AI actually work, not just on paper.

In 2018, while co-directing the Reasoning and Learning Lab at McGill University and serving as a core member of Mila (Quebec's AI institute), she gave a talk at NeurIPS that shook the entire community. Her message was simple and blunt: most machine learning experiments cannot be replicated. Nobody shares the code. Results change if you change the random seed. That is not science, it is alchemy. From that talk, Pineau led the NeurIPS 2019 reproducibility program, which introduced the first mandatory checklist for every paper submitted to the conference. Today that checklist is a standard across the field.

But her impact did not stop at academia. As global VP of FAIR (Fundamental AI Research, Meta's research lab), she oversaw from 2023 the projects that millions of researchers and builders use daily: PyTorch, LLaMA, SAM (Segment Anything Model), FAISS, and more. She was one of the key drivers behind distributing LLaMA openly, giving world-class language models to universities, startups, and labs around the planet. She also received Canada's Governor General's Innovation Award in 2019, and is a Fellow of AAAI (since 2018) and of the Royal Society of Canada (since 2023). In 2025 she left Meta to join Cohere as Chief AI Officer.

What inspires me most about Joëlle Pineau is that she did not settle for doing good things: she insisted those things be verifiable. In a field where hype often runs faster than evidence, she chose the longer, right road. If you use an AI model today and trust that the results you see are real and not magic, part of that trust you owe to her.