Panel: AI offers speed and scale for health care but governance, testing and transparency are required — CalPERS begins pilots

CalPERS hosted external guests and internal presenters to discuss the role of artificial intelligence in health coverage and operations. Presenters distinguished predictive AI (risk scoring, sepsis prediction, resource forecasting) from generative AI / large language models (summaries, drafting and natural‑language interaction). They described clinical uses — automated drafts of patient messages, population‑health analytics, virtual care augmentation and image‑analysis tools — and cautioned that early randomized evaluation at UC San Diego showed pre‑drafted clinician messages did not reduce clinician time and increased message length while improving perceived empathy.

Speakers said governance is essential and described CalPERS’ work: an organizational AI governance committee, an updated Acceptable Use policy, responsible‑AI training for staff, and pilots such as Persi (CalPERS’ internal GenAI assistant) and Microsoft 365 Copilot assessments. Planned GenAI use cases included automated summary of contact‑center calls, extraction and analysis of unstructured investment research, an employer‑policy analyzer (“PERLbot”) to speed labor‑code reviews, and software‑development assistance. Staff emphasized human‑in‑the‑loop safeguards, third‑party testing where feasible, transparency about AI uses that affect members and coordination with state and federal regulators on disclosure standards and liability frameworks.