Project Description

Santa Clara County, CA is a prominent local government serving nearly two million residents. The County faces an increasing challenge in evaluating and implementing AI systems across its numerous departments. As AI technology becomes increasingly prevalent in government operations, the County receives a growing influx of proposals for AI implementation that span diverse use cases - from streamlining administrative processes to enhancing public services. Currently, the County's approach requires creating individual guidelines and evaluation criteria for each AI technology proposal, resulting in a fragmented and resource-intensive process. This approach has become increasingly unsustainable as the volume of AI proposals grows. The lack of a standardized framework for evaluating AI technologies across departments makes it challenging for the County to maintain both efficiency and thorough risk assessment in its technology adoption process.

Project Aims

This document recommends a streamlined multi-stage framework for generating AI use case guidelines for the Santa Clara County government. We offer a template of use-case guidelines to approve future technology usage proposals to streamline the time-consuming process of producing guidelines for each individual proposal. We propose a singular, standardized document to evaluate AI usage applications utilizing a two-tier template, first focusing on technology and then on specific departments and divisions within the government. By implementing this structured methodology, Santa Clara County, CA can more effectively assess and approve AI systems across departments, while ensuring both technological consistency and departmental specificity.

Methodology

We focused on analyzing Santa Clara County's existing technology evaluation processes to develop an efficient two-tiered template for AI use case guidelines. We reviewed current departmental use case guidelines to identify common elements and department-specific requirements. We conduct a systematic literature review of relevant state legislation, including California's AB 302 and SB 313, as well as federal frameworks such as NIST's AI Risk Management Framework.

To validate our approach, we mapped existing County use case guidelines to identify patterns in departmental requirements and conducted stakeholder consultations to understand specific implementation needs. We also benchmarked against other municipal AI governance frameworks, particularly from jurisdictions of similar size and complexity. This comprehensive strategy enabled us to determine the essential components for both technology-specific and department-specific tiers and maintain adaptability across county departments.

Project Deliverables

We created a two-tiered template for generating use-case guidelines that simplifies the guideline creation process to ensure that each department's guidelines are both thorough and aligned with organizational standards.

Project Impact and Future Work

Our two-tiered template for use-case guidelines represents a novel approach to streamlining technology application proposals and mitigating risks from new GenAI technologies. By introducing a modular framework with two distinct levels, our method ensures a more efficient, scalable, and adaptable procurement process for evaluating and approving the use of various rapidly emerging technologies across a wide range of county operations. We recommend a phased approach to implement this system. The county should begin by creating a robust set of Tier 1 guidelines for any broad pilot technology, such as generative AI tools like GPTs. Concurrently, departments can start drafting Tier 2 guidelines specific to their operational needs, using the Tier 1 framework as a base. The current existing guidelines can be retrofitted into our two-tier structure in order to ensure consistency and alignment with the new framework. By leveraging our recommendations, Santa Clara County can enhance its capacity to adopt and regulate technologies while safeguarding security, privacy, and ethical use.

Contributors