collective-alignment-public-input

---
name: collective-alignment-public-input
category: ai-safety
description: Methodology for incorporating public input into AI model alignment through large-scale surveys and democratic value aggregation.
---

# Collective Alignment: Public Input on AI Model Behavior

## Overview

Methodology from OpenAI's work on collective alignment — surveying over 1,000 people worldwide on how AI should behave and comparing their views to the Model Spec. Demonstrates how to align AI defaults to diverse human values.

## Core Methodology

### 1. Survey Design
- **Demographic Sampling**: Ensure global representation across cultures
- **Behavior Scenarios**: Present concrete AI behavior dilemmas
- **Preference Elicitation**: Quantify what people want AI to do in edge cases

### 2. Analysis
- **Compare to Model Spec**: Measure gaps between public opinion and current AI behavior
- **Identify Consensus**: Find areas of broad agreement vs. cultural divergence
- **Update Defaults**: Use results to refine AI default behavior

### 3. Iteration
- **Regular Re-surveying**: Capture evolving public sentiment
- **Transparency**: Publish findings and methodology
- **Accountability**: Link survey results to actual model updates

## Key Findings

- Global public opinion often differs from existing AI defaults
- Cultural variation requires nuanced, context-aware alignment
- Democratic processes can inform but not fully determine AI behavior

## When to Use

- AI alignment research
- Building democratically-aligned AI systems
- Understanding cultural differences in AI preferences
- Designing user-configurable AI behavior

**Activation**: collective alignment, public input AI, democratic AI alignment, Model Spec, AI behavior preferences