The Good, the Bad and the Ugly of Predictive AI

Governments have long used prediction to make decisions about where to deploy resources, whom to prioritize for services, which risks require attention, and where problems are likely to emerge. Done well, increasingly sophisticated predictive AI systems promise to make these decisions faster, cheaper, and more accurate. Predictive systems are already shaping decisions about inspections, benefits, healthcare, public safety, fraud detection, workforce planning, and service delivery. Some have helped governments allocate resources more effectively and anticipate needs before problems occur. Others have produced inaccurate forecasts, reinforced existing inequities, or encouraged agencies to place too much confidence in uncertain predictions.

Offered in collaboration with the Center for Information Technology Policy at Princeton University, The Good, the Bad, and the Ugly of Predictive AI explores when predictive systems genuinely improve public-sector decision-making and when they create new risks, blind spots, or harms.

Through practical examples and discussion, participants will examine how predictive systems work, where they create value, where they fall short, and what safeguards are needed to ensure that predictive tools augment rather than replace human judgment. The series also explores a fundamental question: not every public-sector problem should be approached as a prediction problem. How can public professionals distinguish between situations where prediction can improve outcomes and those where it may create new risks or unintended consequences?

By the end of this series, participants will be able to:

• Explain how predictive systems work and what they can—and cannot—tell us.
• Distinguish between appropriate and inappropriate uses of predictive AI in public-sector settings.
• Assess the quality, limitations, and risks of predictive models, including data quality issues, bias, uncertainty, and feedback effects.
• Evaluate predictive systems and identify the forms of monitoring and oversight needed to ensure that prediction supports rather than replaces human judgment.
• Design governance and oversight practices that preserve transparency, accountability, fairness, and public trust.