Field notes from AI's crowded frontier.
About
Hi! I'm Mikey. I'm trying to work out how to make AI go well, which it turns out is less than straightforward.
I'm a strategist, evaluator, and researcher. For more than 16 years, leaders in philanthropy - and previously, governments and nonprofits - have relied on me to figure out what they're trying to achieve, how to do it, and whether it's working. In practice, this means I design and lead rigorous, mixed-methods evaluations of hard-to-measure initiatives, and translate research results into strategies people actually use to achieve impact. I've worked all over the world, from California (AmeriCorps) to Bulgaria (Peace Corps), and in countries across Africa, Asia, North America, and Latin America, where I've advised and led programs seeking to strengthen governance, improve climate resilience, and reduce poverty.
Transitioning to AI Safety
In mid May 2026, I came across this letter by George Rosenfeld. Reading it crystallized something for me: the questions I've been asking throughout my career, about how to do good in contexts where knowing what good looks like - and how to measure it - can be extremely challenging, now most urgently need to be answered in the field of AI safety.
Social scientists have spent decades building methods for working out not just whether something succeeded or failed, but how, why, and under what conditions, and using evidence to develop strategies for shaping complex systems. Much of this work has yet to be adapted to the field of AI. That matters, because AI capabilities are fast outpacing our ability to understand how frontier systems behave, and why. My hypothesis is that we need a richer toolkit for assessing AI systems, making evidence-based governance decisions, and testing and implementing rigorous, impactful strategies, at model, organization, and field levels - and that other disciplines can help.
I'm now focused on helping AI go well, including by exploring how to apply social science insights and evaluation methods to AI safety. This site is where I test my thinking about what that might look like, and what it might make possible.
See my writing for more, or find me on LessWrong. You can also check out my CV, or get in touch at mjm469@georgetown.edu / on LinkedIn.