Skip to content
Citizen Infrastructure

Technology's place is education

A recurring theme among people who’ve spent years helping groups govern themselves: the limiting factor is rarely a missing tool. It’s a missing literacy — the human capacity to share power, facilitate, and decide together. Get that backwards, and you build “boxes” that don’t let people do what they actually need to do.

Education first, then tools

Section titled “Education first, then tools”

If you treat self-governance as fundamentally an educational project rather than a mechanical one, technology’s role changes. It stops being the thing that produces good democracy and becomes a way to make education motivated, well-structured, and consistent across people — a tool people use to come together and practise, not a replacement for the practice.

Seen this way, even AI looks different: not an engine that generates good decisions, but something that can help more people build the skills and connections that good decisions require.

The offline test

Section titled “The offline test”

There’s a useful heuristic here.

The advice to builders: don’t start from a single tool or feature. Go big, meta, and comprehensive first — understand the literacy you’re trying to grow — and only then get specific about tools. It’s a close cousin of the lesson that the hard part of civic tech isn’t the tech.

A worked example: citizens who build evals

Section titled “A worked example: citizens who build evals”

One vivid version of “education first” comes from the Stanford political economist Andy Hall, who argues that the answer to AI in the classroom isn’t to ban it but to make it an object of study. In his teaching, students don’t just use AI; they build their own evals — small tests that probe what a model actually does: where it lands on a moral dilemma, whether it can predict an election, how it handles a sensitive data scenario. Building the test is the literacy. A student who has written an eval understands a model’s limits in a way no amount of using it conveys — and, Hall’s civic point, is now equipped to hold a system to account against their own values rather than take its outputs on trust. It’s the offline test in a new key: the capacity (knowing what to measure, and why) comes first; the tool only amplifies it. Scaled up, it’s a route to the kind of broadly-distributed AI literacy that lets a public, not just a lab, judge whether a system is worthy of trust.

Sources

Section titled “Sources”