AI Policy - Making Progress Using a Human-centered Approach

I've spent time in rooms where government and Silicon Valley had to find common ground on technology policy challenges issues that were politically charged, newsworthy, and contentious. Social media. Election integrity. Platform accountability. In fact, I created the room, or the space for strategic dialogue to take place, when none existed.

What I learned: the hardest part was never the "what." The issues were mapped. Everyone more or less knew what the problems were.

The hardest part was the "how" — building enough institutional trust across a deep cultural and incentive divide to actually make progress.

To have credibility to create the room, I had to do the deep research to understand the issues (the “what”), but I was also a third party that was able to see the problems in new and different ways, and importantly challenge conventional wisdom in how to think about and solve for the problem. I did not come with the baggage of an entrenched perspective. I made a point to see it from various perspectives.

We're in the same place with AI policy now. And the diagnosis is worth being clear-eyed about:

The gap is structural. Policymakers can't evaluate what they're being told. Technologists can't translate it into governable frameworks. The people who can do both are spread too thin across too many urgent problems. And both sides can’t see the value of bridge builders, or the connective tissue across the divide.

Regulatory capture is happening before regulation exists. The entities with the deepest technical knowledge have the strongest financial interest in shaping the rules. The lobbying apparatus around AI has already outpaced government's capacity to respond.

Timescales are mismatched. AI capability develops in months. Good governance requires frameworks that outlast election cycles. Democracies aren't built for that.

And the apparent consensus is shallower than it looks. "Safe, trustworthy, responsible AI" is a value statement, not a mechanism. The gap between values and implementation is where governance efforts go to stall. CSET has taken a step toward this with their Operationalizing AI whitepaper.

We're not stuck because we lack ideas. We're stuck because we haven't built the institutional infrastructure — translators, shared empirical frameworks, durable convening structures — that serious governance demands.

I've helped build versions of that infrastructure before, in a related context. It requires treating the "how" of convening as a discipline in its own right. I'm not suggesting it's easy. I am suggested it's possible.

That's the work ahead.