Get Full Government Meeting Transcripts, Videos, & Alerts Forever!
Get email alerts on the Ai Governance topic
No spam. Unsubscribe anytime.
CalSTRS board hears Carnegie Mellon expert on AI governance; emphasis on sandboxes, red‑teaming and human accountability
Summary
Carnegie Mellon’s Dr. Param Bir Singh told the CalSTRS board that trustees must build board-level ‘‘fluency,’’ insist on sandboxes and red‑teaming before production use, and assign clear human accountability whenever AI materially affects member services or investment decisions.
Dr. Param Bir Singh of Carnegie Mellon told the California State Teachers' Retirement System board that trustees should focus governance on questions of accountability and risk where AI materially affects decisions.
"What is the role of the board, particularly when AI starts having an impact on decision making?" Dr. Singh asked, framing the presentation as a nontechnical governance session intended to help trustees identify what to ask management before AI implementations proceed.
He said boards need collective AI fluency rather than relying on one technical director, noting industry research showing boards with multiple tech‑savvy directors outperform peers. "You need at least three tech savvy directors on the board," he said, adding that a single expert can become an overrelied-on adviser rather than producing effective oversight.
Dr. Singh urged practical safeguards before enterprise implementation: use sandboxes for testing, run comprehensive red‑teaming exercises to surface failure modes, and require auditable records of use. He stressed the need for a named human accountable for outcomes: "AI can be used, but humans should be the ones who should be deciding eventually ... who is going to be responsible for that." He illustrated consequences with high‑profile examples including biased triage systems in health care, the UK Post Office prosecutions tied to flawed systems, Zillow’s iBuyer losses when models failed at scale, and the Apple‑card bias controversy.
Board members raised questions about oversight structures, asking whether a dedicated committee or integration into existing committees works best. Dr. Singh recommended a small committee to reduce diffusion of responsibility and regular board education sessions to build common fluency; he also recommended external technical advisors to support but not supplant board responsibility.
On metrics and evaluation, Dr. Singh urged trustees to judge AI initiatives on nonfinancial outcomes and decision quality — for example, whether AI produces earlier risk insights or improves member service response times — rather than only short‑term ROI. He warned that many pilots succeed technically but fail during scale and integration, and that boards should require evidence that pilots will embed safely into organizational decision processes.
The presentation concluded with a Q&A in which trustees discussed workforce impacts, training and onboarding for staff, red‑teaming cadence, and how to balance technical detail with governance oversight. Trustees broadly endorsed continued education and asked staff to present governance recommendations at future meetings.
The board’s discussion positions AI as a strategic tool that requires tailored oversight where it affects member benefits, investments or other material decisions; the board asked staff to return with recommended governance definitions and educational material in coming meetings.
The board heard the presentation in open session; there were no public comments on the education item.
