The Problem Genome

The Problem Genome is a human–AI collaboration project at Dartmouth that points researchers’ and project-based learners’ attention to worthy problems, domain knowledge, and what it will take to solve them.

It is built for the first weeks of a project-based course, when teams usually re-derive what experts already know. Two taps take a student from ‘what are you into?’ to six real problems; every problem is drawn as a card you can read — the title carries the data, the ground carries its constraints — and each brief page works as a ready-made assignment sheet: what’s stuck, why it matters, what’s been tried, and entry points sized to a semester. Instructors can assign it cold; teams can adopt a problem, or bring the collection a new one.

The collection holds 600+ problems and counting across 30+ domains, each tagged along eight dimensions — about 80 genes — and sorted into 19 cross-domain families with names like ‘It worked in the lab’ and ‘The missing yardstick.’ Verification is the spine: every brief cites an expert source that was fetched and checked, new problems pass a gate before they enter, independent readers re-tag samples blind so the genes can be trusted, and when a citation fails a later check, the repair is logged on the brief itself.

Begun at Harvard in 2013 with HILT funding — right theory, stalled on expert supply — and made possible at scale in 2025–2026 by AI collection with human review at the Altringer Eagle Lab, Dartmouth. The March 2026 site’s five hundred round glyphs are kept as period work, regenerated as the print 500 Problems.