Investigator: Kenneth Goodman (co-PI).
Summary
The use of predictive analytics has long raised challenging ethical issues:
- Prognostic scoring systems in critical care medicine: The ability to predict death with accuracy could inform patient choice, improve quality, and reduce unnecessary cost ? at least in principle. Though all such systems are imperfect, several were proposed to guide rationing and triage during the COVID-19 crisis.
- Predictive analytics in industry and power generation: The ability to improve safety and quality should always be improved. Yet in domains largely shaped by industry and empirical standards, there are few widely agreed upon standards to guide the appropriate use of traditional and machine learning software in industries essential for civil society.
- Law enforcement: Though crime reduction is a worthy goal, the ability to make accurate projections about individual and group behavior has led to information technologies that can infringe on civil liberties or be used for political purposes. The adoption and use of such technologies has outpaced ethical analyses of software standards and of what constitutes appropriate uses and users.
This effort, a joint initiative of the University of Miami’s Institute for Bioethics and Health Policy and Institute for Data Science and Computing, will undertake analyses of these three applications of predictive analytics with special regard to
- Software engineering: What is needed to identify best practices related to documentation and annotation, including provenance; version control; and data curation, including the statistical problem of missingness?
- Bias: Clearly documented as an ethical challenge for artificial intelligence in health and other domains, it remains to identify the roles and responsibilities of developers, vendors and users in minimizing bias in machine learning and other analytics. Note that transparency and explainability, for instance, are close cousins of bias reduction and mitigation.
- Governance: Development of governance mechanisms for predictive analytics and other software has become a growth industry. This is a fertile opportunity for a kind of meta-analysis of existing approaches and to attempt to identify a suite of ethically optimized approaches to oversight and regulation.