UC San Diego Labor Center and researchers urge worker-centered AI adoption, flag cases of harm

Satomi Rausch Ziegler, executive director of the UC San Diego Labor Center, and Dr. Lilly Irani, associate professor and faculty director at the labor center, warned the San Diego County AI Ad Hoc Subcommittee that AI implementations can produce operational failures, safety risks and inequitable outcomes unless workers and subject-matter experts are involved from planning through deployment.

The presenters summarized machine-learning basics and described several case studies intended to show both risks and mitigation steps. Their recommendations emphasized transparency from vendors, human oversight, inclusive design involving front-line workers and phased testing.

Key case studies and regional implications

- Hershey's ERP rollout: The presenters said a rapid enterprise resource planning (ERP) deployment resulted in major operational disruptions, a reported drop in sales and a substantial stock decline; the presentation cited roughly $100 million in lost sales and an approximately 18'19 percent decline in key metrics tied to the rushed implementation. Presenters said the failure illustrated the danger of insufficient stakeholder input and training.

- Sepsis detection tool: A health system deployed an AI sepsis-alert system that produced many false positives and did not include nurse override mechanisms; staff had to perform unnecessary interventions and faced potential professional and liability risk. Presenters said this case shows the need for human override and clinician involvement in design and deployment.

- Hotel room scheduling: An algorithm that prioritized elite guests' rooms without worker input forced room attendants to travel between floors with heavy carts, reducing productivity and increasing physical strain.

- Automated background checks: Presenters said automated screening products used by landlords and employers often contain outdated or mismerged records. Because scores are automated and opaque, individuals can be wrongly denied housing or work; presenters cited National Consumer Law Center findings and noted FTC enforcement activity.

- Algorithmic rent pricing: The presenters described algorithmic price-setting tools used by landlords to set rents and noted City of San Diego council attention to related competitive and housing-cost impacts. They said centralized data and automated matching can exacerbate rent increases and homelessness trends; presenters cited a 26 percent rent increase in San Diego over three years as part of the context discussed.

- Data center pollution: Researchers raised environmental and public-health concerns from large data centers that power large-language models and AI services, citing academic estimates of regional public-health costs (a Northern Virginia study estimated $190'$260 million in regional health costs at one data center in a year) and urging consideration of non-carbon pollution reporting and equitable burden sharing.

Recommendations

The presenters urged local governments to: involve workers and subject-matter experts at the outset; define success metrics and test tools in limited pilots before full rollouts; demand data access and transparency from vendors; build human-override mechanisms and accountable governance; and support training and upskilling for employees. They also asked the county to partner on a regional worker survey the UC San Diego Labor Center launched in English, Spanish and Tagalog to gather local workplace experiences with AI.

Ending

The labor center asked the county for help distributing its survey and said the results will guide tailored policy responses for the region.