AI for Impact 
Prediction Isn’t Intelligence: How Predictive Models Really Work in Government
Prediction Isn’t Intelligence: How Predictive Models Really Work in Government
AI is not one thing. Government agencies are being offered a rapidly expanding ecosystem of tools from workflow automation to document classification to chat-based assistants, enterprise search, recommendation engines, and predictive models. But these tools differ in capabilities, requirements, risks, and appropriate uses. This workshop offers a clear conceptual map to help public servants understand the differences and distinguish real functionality from marketing hype. Participants will learn why prediction is not intelligence, why predictive systems have hard limits that matter for government, and how confusing automation with prediction could lead to costly or harmful deployments. Through case studies in public safety, benefits administration, and inspections, the session highlights how misidentifying the problem—rather than technical failure—often explains why systems underperform.
-
Distinguish predictive models from automation, retrieval systems, chat-based agents, and rule-based tools.
-
Understand why prediction has inherent limits and why those limits matter in government settings.
-
Recognize common points of failure such as biased training data, feedback loops, and drift.
-
Build a skeptical, evidence-first mindset toward claims that “AI can predict X.”