Core Concepts
The authors developed the Situate AI Guidebook to support public sector agencies in making informed decisions about developing or deploying AI tools, focusing on reflexive deliberation and practicality.
Abstract
The Situate AI Guidebook aims to scaffold early-stage deliberations around the development or deployment of proposed AI innovations. It addresses societal, legal, data, and organizational factors crucial for responsible AI governance in the public sector. Through co-design activities with stakeholders, the guidebook offers structured processes and key questions to facilitate informed decision-making.
Public sector agencies are rapidly adopting AI systems but face challenges in ensuring responsible design and implementation. The failures of existing AI tools highlight the need for systematic processes to support decision-making at early stages of ideation and design. The guidebook emphasizes reflexive deliberation on goals, societal impacts, data constraints, and governance factors to enhance ethical considerations in AI projects.
Key questions in the guidebook cover topics such as community needs, ethical considerations, data quality, model selection, long-term maintenance, and organizational policies. By promoting reflective discussions and practical decision-making processes, the guidebook aims to improve outcomes and mitigate risks associated with public sector AI projects.
Stats
"We conducted co-design activities and semi-structured interviews with public sector agency workers (agency leaders,
AI practitioners, frontline workers) and community advocates."
"A growing body of work documents how these AI systems often fail to improve services in practice."
"Many failures in public sector AI projects can be traced back to decisions made during the earliest problem formulation and ideation stages of AI design."
Quotes
"We conducted formative semi-structured interviews and iterative co-design activities that guided the content and process design of the Situate AI Guidebook."
"Participants shared that they did not currently have structured opportunities to proactively discuss social and ethical considerations surrounding AI tool design."
"Participants who had experience developing AI tools often underscored the importance of ensuring that they had the computing resources and data needed to develop their proposed AI tool."