Margaret Mitchell

There are people in AI who build models that are bigger, faster, more capable. And there are people who ask the uncomfortable question: for whom? Margaret Mitchell, known in the field as "Meg," spent her career asking exactly that, out loud and in peer-reviewed papers. She studied computational linguistics, earned a PhD from the University of Aberdeen and a postdoc at Johns Hopkins, then joined Microsoft Research where she worked on Seeing AI, an app that describes the visual world to people with visual impairments. That early experience of building AI that literally sees for others shaped everything that followed.

She joined Google in 2016 and soon after co-founded the Ethical AI team, one of the first in the industry dedicated to fairness, inclusion, and transparency in machine learning models. Her most lasting contribution came in 2019: with her Google colleagues and researcher Timnit Gebru, she published "Model Cards for Model Reporting" at the FAccT conference (January 2019). The proposal was simple but revolutionary: every AI model should come with a documentation card describing what it does, what data it was trained on, where it fails, and what known limitations it has. Today, model cards are the de facto standard across the entire industry, from Hugging Face to Google, Meta, and Anthropic.

In 2021 came the hardest part of her story. After Google fired Timnit Gebru for publishing research critical of large language models, Mitchell tried to document what happened using corporate email. Google locked her out and in February 2021 fired her, citing policy violations. The incident sparked protests among Google employees worldwide and ignited a debate about research freedom inside the big AI companies. Months later, Hugging Face hired her as Chief Ethics Scientist: the organization that has done the most to democratize access to AI models wanted its most articulate ethical voice working from the inside. There she contributed to the BigScience project that produced BLOOM (2022), the first large, fully open and multilingual language model.

What I find most valuable about Meg Mitchell is not just that she created model cards (though that legacy is enormous), but that she showed transparency in AI is not a luxury or a marketing extra: it is an obligation that sometimes costs something. Every time you see a model's technical card, a note about its limitations, or an evaluation broken down by demographic group, a little of her work is behind that honesty. Using AI today with awareness, holding it accountable when something breaks, demanding to know what data trained what you use: that is also the legacy of people like her.