美国人工智能权利法案蓝图（英文）-75页_11mb

Blueprint for an AI Bill of Rights Summary

Introduction

This document, published by the White House Office of Science and Technology Policy in 2022, outlines a framework of five principles to guide the design, development, and deployment of automated systems, ensuring they protect American rights, opportunities, and access while promoting safety, fairness, and transparency.

Key Principles

• Safe and Effective Systems: Automated systems must undergo rigorous testing, monitoring, and risk mitigation to ensure they are safe, effective, and do not cause harm, including measures to prevent misuse or unintended consequences.
• Algorithmic Discrimination Protections: Systems should avoid bias and unfair treatment based on protected characteristics, requiring equity assessments, representative data, and ongoing disparity testing to ensure fairness for all communities.
• Data Privacy: Protections include built-in safeguards for data collection, user consent, and limitations on surveillance, emphasizing privacy by design and giving individuals control over their data.
• Notice and Explanation: Users should be informed about automation use and receive clear explanations of how decisions are made, enabling transparency and accountability.
• Human Alternatives, Consideration, and Fallback: Options to opt out automated systems and access human alternatives or appeals must be available, especially in high-risk contexts, with processes for timely remedy.

Scope

The Blueprint applies to automated systems with the potential to meaningfully impact civil rights, equal opportunities, or access to critical resources. It addresses issues in sensitive domains like employment, education, healthcare, and criminal justice, with focus on underserved communities.

Legal Status

As a non-binding guide, it does not supersede existing laws, regulations, or international instruments. It aims to inform policies and practices without mandating compliance, adapting to specific sector needs.

Public Engagement

The framework was developed through a year-long process involving public input from communities, industry, and experts to address harms and promote responsible AI use.

Implementation

These principles can be integrated into policies and technical standards, as illustrated by examples like government executive orders, industry practices, and ongoing research, ensuring protections for the public.