Harnessing Multi-sector Collaboration in Statistics and Artificial Intelligence for Social Welfare

This paper explores a range of management topics that can be important for multi-sector collaboration in the production, dissemination and use of high-quality statistical information. Principal attention focuses on statistical information historically produced through government agencies, or other public-stewardship institutions.
First, the paper considers factors that can be important in defining goals for statistical information intended for social welfare. These include identification of high-priority information needs of key stakeholders; and the translation of those needs into particular statistical information products that meet specific criteria for quality, risk and cost.
Those criteria lead to the second topic: exploration of data sources, methodology and technology to carry out the needed statistical work on a sustainable and cost-effective basis. Extensions beyond traditional work with sample surveys and administrative records can lead to in-depth reconsideration of quality criteria, and of related communications with stakeholders.
Third, the paper highlights a range of questions that are important for management of multi-sector collaboration to address the goals described above. These involve:
- Notable contributions available from each sector, including specialized technical and managerial capabilities; stakeholder networks; data sources; and discretionary funds

- Differing sector-level approaches to management of intangible capital; intellectual property rights; stakeholder relationships; and multiple dimensions of quality, cost and risk

- Related implications for transparency on methodology, technology and empirical results

- Other similarities and differences in institutional culture and incentive structures
 

Companies in every sector of the economy are attempting to leverage the latest advances in artificial intelligence (AI). While this brings exciting opportunities and innovation, quickly incorporating such cutting-edge technology into user-facing systems brings obvious risks. I will summarize my group's recent work on using principled statistical procedures to ensure degrees of reliability in large, black-box AI models (e.g. neural networks, transformers). Applications range from quantifying the uncertainty in visual systems for autonomous driving, to guaranteeing speed-vs-performance tradeoffs for models run on low-resource hardware, to ensuring users cannot prompt a large language model to go beyond behavioral boundaries. All work discussed has been done in collaboration with industrial partners and with contributions from their research scientists.