Policy Maps: Tools for Guiding the Unbounded Space of LLM Behaviors

Abstract

AI policy sets boundaries on acceptable behavior for AI models, but this is challenging in the context of large language models (LLMs): how do you ensure coverage over a vast behavior space? We introduce policy maps, an approach to AI policy design inspired by the practice of physical mapmaking. Instead of aiming for full coverage, policy maps aid effective navigation through intentional design choices about which aspects to capture and which to abstract away. With Policy Projector, an interactive tool for designing LLM policy maps, an AI practitioner can survey the landscape of model input-output pairs, define custom regions (e.g., "violence"), and navigate these regions with if-then policy rules that can act on LLM outputs (e.g., if output contains "violence" and "graphic details," then rewrite without "graphic details"). Policy Projector supports interactive policy authoring using LLM classification and steering and a map visualization reflecting the AI practitioner's work. In an evaluation with 12 AI safety experts, our system helps policy designers craft policies around problematic model behaviors such as incorrect gender assumptions and handling of immediate physical safety threats.

UIST 2025

References in corpus (15)

• Improving fairness in machine learning systems: What do industry practitioners need?
• Supporting Qualitative Analysis with Large Language Models: Combining Codebook with GPT-3 for Deductive Coding
• Everyday algorithm auditing: Understanding the power of everyday users in surfacing harmful algorithmic behaviors
• ChainForge: A Visual Toolkit for Prompt Engineering and LLM Hypothesis Testing
• EvalLM: Interactive Evaluation of Large Language Model Prompts on User-Defined Criteria
• Collective Constitutional AI: Aligning a Language Model with Public Input
• Understanding Practices, Challenges, and Opportunities for User-Engaged Algorithm Auditing in Industry Practice
• Identifying and Improving Disability Bias in GPT-Based Resume Screening
• Deliberating with AI: Improving Decision-Making for the Future through Participatory AI Design and Stakeholder Deliberation
• Zeno: An Interactive Framework for Behavioral Evaluation of Machine Learning
• Participation in the age of foundation models
• Concept Induction: Analyzing Unstructured Text with High-Level Concepts Using LLooM
• Model Sketching: Centering Concepts in Early-Stage Machine Learning Model Design
• Wikibench: Community-Driven Data Curation for AI Evaluation on Wikipedia
• Where Does My Model Underperform? A Human Evaluation of Slice Discovery Algorithms