Last June, Southern California's air quality board considered rules to phase out gas-powered appliances — a measure their own staff estimated would prevent nearly 2,500 premature deaths. Tens of thousands of opposing emails arrived. The board rejected the proposal, 7 to 5.
This month, the Los Angeles Times revealed that over 20,000 of those opposition comments were generated through an AI-powered advocacy platform. When district staff contacted a sample of the supposed authors, at least three said they had never written to the agency and didn't know the messages existed.
Twenty thousand voices. And nobody home.
The debate about AI tends to split along a familiar line. On one side: AI is a powerful tool that amplifies human capability. On the other: AI is a threat that undermines human agency. The Southern California case dissolves both.
The AI that generated those comments wasn't powerful or threatening in any exotic sense. It did exactly what it was asked to do — produce text that sounded like public opinion. It was perfectly capable. Perfectly obedient. Perfectly empty. The problem wasn't the technology. It was the absence at the center of its deployment.
No one was thinking. No one was listening. No one was weighing the real tradeoffs — health costs against economic impact, transition timelines against air quality data. The AI was a megaphone pointed at a democratic process, and the hand holding it belonged to someone who had already left the room.
This is a pattern, not an incident. It will repeat — is already repeating — at every scale where a democratic body asks "what do the people think?" City councils. Federal comment periods. School boards. Planning commissions. Anywhere the architecture of public input was designed for a world where writing a letter meant someone actually sat down and wrote a letter.
Here is what makes this structural rather than scandalous.
The essay that went viral earlier this year — the one celebrating AI as the ultimate productivity tool — described a human's proudest moment as walking away from his computer for four hours and returning to find perfect work. The Southern California campaign is the same move, pointed at democracy. Same architecture. Same absence. One is celebrated; the other is condemned. But the underlying gesture is identical: deploy the intelligence, leave the room, collect the output.
The tool-use frame cannot distinguish between these two cases. In both, AI is being used "effectively." In both, the human has optimized themselves out of the process. The frame has no way to ask the question that matters: is anyone actually present?
Presence is not a feature you can add to AI. It's a quality that humans bring or withhold.
Imagine, instead of an AI that manufactures the appearance of public opinion, a tool that helped citizens genuinely think through the tradeoffs. That surfaced complexity rather than flattening it. That helped people arrive at their own position rather than signing a prefabricated one. That technology is equally possible. It simply requires someone to stay in the room.
The fix is not to ban AI from public comment. The fix is to notice that the architecture of democratic input — counting comments as though they were votes — was already fragile before AI arrived. AI didn't break it. AI made the fragility visible.
What does public deliberation look like when it can't be gamed by volume? When the quality of reasoning matters more than the quantity of voices? When AI is used to deepen understanding rather than manufacture consent?
Those are design questions, not technology questions. And they require the one thing no AI can provide on its own: a human who is actually there. Thinking. Listening. Present to the complexity. Refusing to leave the room.
Twenty thousand voices and nobody home. Or one real voice, genuinely thinking, with AI helping them see what they couldn't see alone. Both are possible. The architecture is the same. The presence is everything.