CHANCE: Enhancing Public Understanding of Science, Society, and Technology

The Nuremberg Code states ten directives for human experimentation, including directive six - "the degree of risk to be taken should never exceed that determined by the humanitarian importance of the problem to be solved by the experiment." The Nuremberg Code was followed by the Belmont Report, and the Menlo Report. Published in 2012, the Menlo Report directly considered the impact of technology. The Belmont and Menlo Reports were US governmental initiatives. The Toronto Declaration is a more recent statement of the impact of technology, specifically Artificial Intelligence, on human rights and was conceptualized under the leadership of Amnesty International. Collectively these works provide guidance for the ethical treatment of human beings during the conduct of scientific investigations. This presentation will review these important works in the context of our contemporary world where data science and artificial intelligence practice and research must be viewed as human endeavors. Using the reports to provide an ethical lens by which to view and understand both the development and impact of technology will provide a unique perspective of our current worldview. 

The statistical aspects of trustworthy machine learning (ML) and artificial intelligence (AI) have not been extensively studied, though Statistics has long played a critical role in the arena. When there is lack of consideration of interpretability, uncertainty quantification, limited/incomplete data, and selection bias, statistical methods may offer reasonable solutions. Popular and influential as ML/AI has become, some of the breakthroughs in statistics in the last 50 years have fueled this revolution, such as bootstrap, causal inference, deep learning, and exploratory data analysis, among others. Much effort from the statistical community is needed to tackle the open problems of trustworthy machine learning. We will review the exchanges of researchers at a Banff International Research Station workshop on statistical aspects of trustworthy ML, which was held in February 2024. Specific topics include interpretability, privacy-preservation, robustness, and fairness. 

Artificial intelligence and other computationally intensive tools are leading to extraordinary changes in environments for the production, confidentiality protection, dissemination, and use of public-stewardship statistical information. Truly groundbreaking and productive responses to these changes will require highly creative, thoughtful, and sustained collaboration across, and trust among, many stakeholder groups. This paper highlights ways in which successful collaborations will depend on serious, sustained and practical applications of ethical principles in technical, managerial and stakeholder-relationship settings that are complex, rapidly evolving, and often unpredictable. Four underlying principles receive primary emphasis: respect for – and obligations to - individuals, institutions, communities, and the environment; respect for facts; the importance of contextual factors; and realistic, actionable and timely communication. We illustrate these general concepts with three use cases: (1) confidentiality protection; (2) changes in production processes for current statistical information series; and (3) fundamental changes in the estimands for statistical information suites.