⚡ Quick Summary
Published by Partnership on AI, this guidance sets out tools for improving AI’s impact on jobs and directing AI development and use towards shared prosperity. It is designed for AI-creating and AI-using organizations, policymakers, labor organizations and workers, and applies to AI systems across industries, technologies and use cases. The document defines AI’s impact on shared prosperity as its impact on the availability and quality of formal-sector jobs across skill, demographic and geographic groups.
Its three-step process is to learn about the Guidelines, apply a high-level Job Impact Assessment Tool, and follow stakeholder-specific recommendations. The assessment identifies ten Signals of Opportunity (OS1–OS10) and eighteen Signals of Risk (RS1–RS18), then requires users to examine how prospective benefits and harms are distributed and to repeat the analysis for upstream and downstream markets. The document provides Responsible Practices for AI-creating organizations (RPC1–RPC15) and AI-using organizations (RPU1–RPU17), alongside suggested uses for policymakers and labor organizations. Signals are early indicators rather than guarantees; each identified risk should receive a mitigation strategy, prioritised by severity of potential impact.
🧩 What’s Covered
The guidance proceeds from its rationale and scope to assessment criteria and recommendations for distinct stakeholders.
- Purpose and origin: The opening sections describe anticipated AI-driven changes to jobs, wages, tasks and job quality. They situate the guidance in two years of applied research, including economic modelling, engagement with frontline workers, and work on responsible sourcing of data-enrichment labor.
- Scope and definition: The document frames shared prosperity through labor demand, covering availability and quality of jobs. It lists relevant factors including labor’s share of revenue, task composition, skills, market concentration, job stability, privacy, autonomy, fair treatment and human recourse in employment decisions.
- Application principles: Users are asked to engage affected workers, especially those with least decision-making power; assess signals comprehensively; avoid treating opportunity signals as offsets for risk; and use analysis to support collective decisions and deeper investigation.
- Job Impact Assessment: The tool instructs users to assess every signal rather than cherry-pick, record explanations, mark a signal present where likelihood is above zero, examine distribution across groups and time, and consider effects on competitors, suppliers and clients.
- Opportunity signals: OS1–OS10 cover widely distributed benefits, worker productivity, new paid tasks, egalitarian labor markets, suitability for lower-income geographies, access to work, workers’ revenue share, worker-identified needs, co-development and job quality. Several signals include explicit caveats on displacement, inequality, uncompensated work and trade-offs.
- Risk signals: RS1–RS18 cover elimination or reallocation of tasks, geographic job displacement, concentration, outsourced data labor, non-consensual training data, wage prediction, work intensity, scheduling, breaks, monitoring, autonomy, mentorship, satisfaction, employment decisions and discrimination. The guidance treats elimination of more than 10% of a job’s weighted core tasks as significant enough to warrant attention.
- Responsible practices and suggested uses: The RPC and RPU lists address disclosure, worker participation, mitigation, responsible data-enrichment sourcing, non-discrimination, explanations, data transparency and opt-outs, human recourse, red teaming, worker protections, recognition of extra work and sharing productivity gains. Policymakers can integrate the tool into standards, risk frameworks and conformity assessments; worker organizations can use it for audits, negotiations, training and policy positions.
💡 Why it matters?
Workplace AI decisions can alter employment, pay, working conditions and worker power even where a system functions as intended. This guidance gives organizations a structured way to surface those effects before and during development, procurement and deployment, rather than limiting scrutiny to technical malfunction or misuse.
It is particularly relevant where AI affects performance targets, monitoring, scheduling, hiring, pay or task allocation. The tool connects impact assessment to concrete actions: worker participation, transparent data practices, human-decided grievance mechanisms, risk mitigation and mechanisms for sharing productivity gains. It also gives policymakers and worker representatives a common inventory of issues for regulation, procurement, dialogue and collective bargaining.
❓ What’s Missing
The document expressly describes workplace-AI practice as still preliminary and says the Responsible Practices will be refined through community testing and feedback. It provides a high-level assessment rather than a scoring model, threshold methodology or standard template for estimating the magnitude of impacts; the exact steps for distributional analysis are described as highly case-specific. It does not present completed implementation case studies, although it proposes developing a library of examples. The guidance deliberately leaves global benefit-redistribution mechanisms, such as universal basic income, outside the scope of this version and instead focuses on AI’s effect on labor demand. It is global in framing rather than a jurisdiction-specific account of legal obligations.
👥 Best For
AI product, procurement and workplace-governance teams assessing the labor effects of a proposed or deployed system will find a structured inventory of questions and practices. It also serves policymakers considering AI or worker-protection measures, and unions, worker organizations and representatives preparing audits, negotiations, workplace training or policy positions.
📄 Source Details
Guidelines for AI and Shared Prosperity: Tools for improving AI’s impact on jobs is a 50-page English guidance document published by Partnership on AI in June 2023, marked V.1. The content was iteratively developed by PAI’s Labor and Economy team: Katya Klinova, Stephanie Bell and Sonam Jindal. No URL for this document is printed in the supplied PDF.
