The Grammar of AI Legitimacy

How do publics reason about AI legitimacy?

The project asks how diverse publics reason about AI legitimacy: not only which systems or institutions they trust, but how trust, distrust, ambivalence, and demands for oversight appear in the language of their responses.

Public trust in AI is not only an attitude. It is also a linguistic, relational, and institutional structure. People mark trust through conditions, hedges, verification practices, first-person testimony, procedural demands, warnings, and grammars of agency: “I use it, but verify”; “with human supervision”; “not as a decision-maker”; “companies will manipulate us”; “people should decide.”

Reading those forms carefully can help preserve public reasoning as reasoning — not merely as preference data to be counted, summarized, or smoothed into consensus too quickly.

Public values need more than preference aggregation.

CIP’s work is compelling to me because it treats public participation as infrastructure: something to be designed, tested, interpreted, evaluated, and improved, rather than invoked abstractly or reduced to aggregate preference alone.

My contribution would be to analyze the language through which people explain AI’s usefulness, danger, authority, legitimacy, and limits. I am especially interested in the gap between structured preference data and the texture of open-ended public reasoning — the hesitations, conditions, stories, translations, warnings, and demands for accountability that can disappear when public input is aggregated too quickly.

What is worth preserving in public reasoning?

The project asks not only what publics believe about AI, but what their language reveals about the conditions under which AI systems, AI companies, governments, public research institutions, and AI-governance proposals become trustworthy, untrustworthy, legitimate, or illegitimate.

• How do people linguistically express trust, distrust, and ambivalence toward AI systems, AI companies, governments, public research institutions, and proposed governance structures?Signals: hedging, modality, conditionals, verification language, intensity, warning structures, procedural demands.
• How do publics distinguish AI as a useful tool, advisory interlocutor, autonomous agent, institutional force, or governing actor?Signals: “I use AI,” “AI helps me,” “AI decides,” “companies control AI,” “governments should regulate,” “people should decide.”
• What “legitimacy grammars” recur around transparency, audits, benefit-sharing, representation, local benefit, privacy, corporate control, corruption, and human oversight?Signals: visible procedure, traceability, representation, constraints on power, and demands for accountable institutions.
• How do translation, summarization, peer voting, and model-mediated analysis change what becomes visible as a public value?Risk: public reasoning may become smoother, more coherent, or more consensus-like than the original responses.
• Can a public-reasoning codebook inform evaluation blueprints for models tasked with summarizing or representing public values?Possible output: test cases for whether models preserve dissent, uncertainty, cultural specificity, conditions, and minority concern.

Interpretive methods, structured for comparison.

My background is in close reading, rhetoric, literary interpretation, and public-facing argument. For this project, I would bring that training to structured public-values data through a transparent, bounded, human-validated workflow.

This is a mixed-methods project: qualitative close reading paired with structured coding and comparison across response types, agreement rates, translations, demographic segments, and country or regional groupings where the data supports it.

The point is not to replace quantitative analysis or computational social science. It is to add an interpretive layer that can notice how language carries agency, trust, legitimacy, dignity, fear, and moral imagination under pressure — while still producing traceable, comparable, decision-relevant findings.

From public reasoning to usable governance signals.

The goal is research that preserves interpretive nuance while producing usable signals for AI governance, democratic oversight, model evaluation, and institutional decision-making.