Regenerative agriculture & ecovillages
A decade in intentional communities — Twin Oaks, Living Energy Farm, and others — building soil, water, and food systems by hand.
About
How a sustainability educator ended up working on legible AI
Before the research, there was a decade of building a humane future by hand — regenerative farms, off-grid systems, classrooms, and community. I went into AI safety because that future only works if its intelligence is something we can read, audit, and steer.
I'm an independent researcher working on legible verification — but I didn't arrive from a machine-learning lab. For roughly ten years I worked in regenerative agriculture and intentional communities, designed and taught education programs, built off-grid infrastructure, and helped run events and care work. The through-line was always the same instinct: understand a system well enough to maintain it yourself — to read the soil, the water, the wiring, the room.
That instinct is why I work on AI the way I do. The future I want to live in — decentralized, humane, life-serving — increasingly runs on intelligence we can't see into. So the keystone question became: can we make that intelligence legible — inspectable, externally grounded, governable — the same way you'd want to read a microgrid before you trust it to keep the lights on? That is the research on this site.
The foundation
Off-grid systems & sustainable construction
Solar AC/DC microgrids, natural building, compost and biochar infrastructure; a cargo trailer converted into a mobile workshop and pop-up gallery.
Education design & facilitation
Years designing and teaching — science and chess workshops, custom curricula, and community workshops across the US and abroad.
Disaster response & care
American Red Cross disaster-action and mass-care training; Nonviolent Communication and community mediation.
The other half of the work
Alongside the research I build Values Commons, a privacy-first, open-source public ecosystem for choosing and organizing by values. Its first working instance, Conscious Consuming, is live: 85 datasets, about 24,500 entries, every curated claim source-cited. The two projects share one conviction, that systems people depend on should be inspectable by the people who depend on them. I keep them on separate pages so each can be judged on its own evidence.
Get in touch
Open to research collaboration, funding conversations, and reviewer access to the full evidence. A focused research résumé is here.
bentleymoonperkins@gmail.com