AI Governance Isn't Boring — It's the Most Important UX Problem

Here's the premise of PublicAI Pulse, the project that made me think most seriously about governance:

The people who design AI governance systems are making UX decisions.

They're choosing who gets to interact with the system, what feedback they receive, how transparent the system is about its reasoning, what happens when it fails. These are UX decisions. We just don't call them that.

The Slider Problem

PublicAI Pulse is a browser-based simulator. It has two sliders:

1. Transparency — how much of the AI's reasoning is visible to affected parties
2. Bias Correction Intensity — how aggressively the system attempts to correct for historical biases

Move these sliders and watch what happens to simulated healthcare access, transit routing, housing allocation.

Building this taught me something I didn't expect: the slider positions are not neutral. There's no "default." Every deployment of an AI system in public services has implicit slider settings — usually set by engineers and policymakers who never had to explain them to anyone.

Why Governance Looks Boring (And Isn't)

Governance looks boring because it's discussed in the language of policy. Compliance frameworks. Ethics guidelines. Audit requirements.

This language obscures what governance actually *is*: decisions about power. About who gets to know what. About who gets to appeal when something goes wrong. About what counts as going wrong.

When a transit authority deploys an AI routing system and it routes fewer buses to lower-income neighborhoods, that's not an algorithm problem. It's a governance problem. Someone decided what "efficiency" meant, whose travel time mattered, and whether disparate impact was acceptable.

That's UX. That's design. That's governance.

The Healthcare Dimension

The healthcare slider in PublicAI Pulse is the most sobering.

When transparency is low and bias correction is weak, the simulation shows reduced appointment access for demographic groups historically underserved by healthcare systems. When transparency is high and bias correction is strong, overall efficiency metrics drop slightly — but distribution becomes more equitable.

This tradeoff is real. It exists in deployed systems today. Administrators are making this choice right now — usually implicitly, usually without realizing it.

The goal of PublicAI Pulse is to make that choice explicit. To force people to see the slider they're already moving.

Design Principles for AI Governance

Through building this project and thinking about it since, I've come to believe that good AI governance in public services requires:

1. Legibility — Affected parties should be able to understand, in plain language, what factors the system considers about them.

2. Contestability — There must be a mechanism for challenging decisions. Not just an email address. A real process with defined timelines and escalation paths.

3. Disparity monitoring — Outcomes must be tracked by demographic group. You can't correct for disparities you're not measuring.

4. Sunset clauses — Deployed systems should require active renewal, not passive continuation. "Nothing has broken" is not evidence of "everything is working."

5. Participatory auditing — The communities affected by algorithmic systems should have representatives in the audit process.

None of these are technically difficult. They're all design decisions.

The Most Important UX Problem

We're at an inflection point. AI systems are entering healthcare, transit, housing, criminal justice, education. The governance decisions made in the next five years will shape who benefits and who is harmed.

These decisions look like policy. They're really design.

And designers — engineers who build things for people — should be in the room.

That's why I built PublicAI Pulse. Not because I have all the answers, but because I think more people need to see the sliders.

PublicAI Pulse is live at shraavanitople.github.io/publicai-pulse — try moving the sliders.