Why the AI Safety Community Needs State Capacity and Democracy Reformers

The unsexy work of fixing democracy, rebuilding public trust, and improving government capacity might be a neglected foundation that makes other AI safety goals actually succeed.

Photo by Unseen Histories on Unsplash

We're about to see a lot more attention and funding directed toward AI governance. But not every group needs to focus on AI-specific bills and regulations. Some should work on public trust, state capacity, voting reforms, faster drug approvals, better clinical trials, and improved science funding.

These all matter for transformative technologies. They determine bioresilience, crisis response, and fast coordination on hard problems.

This is AI policy, too. It's just not the prestigious kind. But it might determine how well we navigate the next decades.

Why Institutional Work Is a Bottleneck

Right now, AI policy and institutional work exist in separate silos. Grant-makers treat AI governance as distinct from democratic resilience. The broader Effective Altruist community's lukewarm response to Open Philanthropy's abundance program hints at less appetite for broader institutional work (IMHO).

This separation is not ideal because most tech policy failures aren't about the tech.

Today's bottleneck might be convincing people that transformative AI is coming. But even when evidence becomes undeniable, structural problems will still block effective responses.

Congress passes almost no major legislation. Rulemaking takes many months while tech capabilities change every few months. Government can't pilot Universal Basic Income or test reskilling programs at scale. We lack DARPA-style authorities where talented leaders get 3-4 year terms with minimal oversight to fund and run high-reward programs.

These governance failures show up everywhere: pandemic response, climate policy, infrastructure, and biodefense. We need the foundation that makes complex coordination possible—to respond to AI mishaps, gradual disempowerment, and global catastrophic risks.

When Government Works, People Don't Want to Tear It Down

If you dig into democracy research—the work done by mainstream think tanks and detailed in books like "How Democracies Die" and "Why We're Polarized"—you find that public trust keeps coming up as a key variable. Public trust, which has dropped by 50% in the last decades, affects whether people seek out reliable information and maintain faith in shared institutions.

This matters for AI governance because trust determines societal resilience. When people trust institutions, they're more likely to engage, vote, and support non-populists.

This explains why state capacity on basic services matters so much. A government that makes people's lives better, by delivering basic services, keeping housing costs under 30% of income, and promoting national projects that make people hopeful, creates stability. When institutions work, people are less attracted to politicians who promise to tear everything down.

Institutional changes happen more than people think. The YIMBY movement has had success in many states. The Abundance movement will likely follow suit. Examples like DARPA prove the government can move fast when you give managers authority and accept some risk. On election reform, Alaska and Maine implemented ranked choice voting statewide.

Making Complex Coordination Possible

In a world where institutions can't function, even excellent AI and societal resilience proposals will struggle since they'll depend on who is in power and how policy is decided. Who is in power and how policy is decided depends on our broader institutional issues.

These foundations help domains that require complex coordination.

Some AI governance funds should go toward state capacity and responsive governance efforts because they make every other intervention work.