Drawing From Worker Insights to Chart a Better Path for Workplace AI
Across industries and around the world, AI is changing work. Millions of workers are already encountering AI at their jobs and feeling its effects, for better or worse. In public discussions about AI and labor, however, the real and ongoing impact on workers is often overshadowed by speculation about the far future. Drawing from an international study of on-the-job experiences with AI, Partnership on AI’s (PAI) new report, “AI and Job Quality: Insights From Frontline Workers,” identifies ways that AI is impacting job quality today and what can be done so that everyone shares the benefits of this technology.
Previous work by PAI’s AI and Shared Prosperity Initiative highlighted the need to better understand AI’s impacts on job quality, including by learning from the workers who experience these impacts firsthand. “AI and Job Quality: Insights From Frontline Workers” is intended to start addressing that need.
Through journals and interviews, customer service agents in India, data annotators in sub-Saharan Africa, and warehouse workers in the United States shared their stories about workplace AI. Our new report identifies five important themes that emerged from these workers’ experiences:
- Executive and managerial decisions shape AI’s impacts on workers, for better and worse. This starts with decisions about business models and operating models, continues through technology acquisitions and implementations, and finally manifests in direct impacts to workers.
- Workers have a genuine appreciation for some aspects of AI in their work and how it helps them in their jobs. Their spotlights here point the way to more mutually beneficial approaches to workplace AI.
- Workplace AI’s harms are not new or novel — they are repetitions or extensions of harms from earlier technologies and, as such, should be possible to anticipate, mitigate, and eliminate.
- Current implementations of AI often serve to reduce workers’ ability to exercise their human skills and talents. Skills like judgment, empathy, and creativity are heavily constrained in these implementations. To the extent that the future of AI is intended to increase humans’ ability to use these talents, the present of AI is sending many workers in the opposite direction.
- Empowering workers early in AI development and implementation increases the opportunities to attain the aforementioned benefits and avoid the harms. Workers’ deep experience in their own roles means they should be treated as subject-matter experts throughout the design and implementation process.
The report also offers guidance for those who want to make a positive impact, separated by stakeholder group. These opportunities for impact can be downloaded individually as audience-specific summaries below:
- AI-implementing companies, which can commit to AI deployments that do not decrease employees’ job quality.
- AI-creating companies, which can center the participation and well-being of worker end-users in their values and practices.
- Workers, unions, and worker organizers, who can work to influence corporate policy as well as purchase and implementation decisions.
- Policymakers, who can shape the environments in which AI products are developed, sold, and implemented.
- Investors, who can account for the downside risks posed by practices harmful to workers and the potential value created by worker-friendly technologies.
The actions of each of these groups have the potential to both increase the prosperity enabled by AI technologies and share it more broadly. Together, we can steer AI in a direction that ensures it will benefit workers and society as a whole.
The Partnership on AI is leading an effort — in continued collaboration with workers at the frontier of AI implementation — to develop commitments and targets for AI’s impacts on job quality as well as tools to support stakeholders in implementing these targets and commitments. To learn more about this effort and stay updated on our work, visit the AI and Shared Prosperity Initiative page on PAI’s website.
This work was generously supported by the Ford Foundation.