New Taxonomy Reveals The Five Different Faces Of AI In The Public Sector

Most public administration studies still lump artificial intelligence into a single catchall category—a dangerous oversimplification as governments race to deploy AI systems. A newly published taxonomy now identifies five distinct 'faces' of AI in the public sector, offering a framework that could reshape how agencies evaluate, procure, and govern these tools.

Dr. Eleanor Vance, a research fellow at the Kennedy School's Ash Center for Democratic Governance, unveiled the taxonomy in a paper released July 24, 2026. The framework classifies public-sector AI into five categories: Efficiency AI (automating back-office tasks), Engagement AI (citizen-facing chatbots and portals), Enforcement AI (predictive policing and compliance monitoring), Exploration AI (data analysis for policy research), and Ethics AI (systems designed to audit and govern other AI). Each face carries distinct risks, benefits, and regulatory needs, says Vance.

Why now? Governments worldwide are pouring billions into AI—the U.S. federal government alone spent $4.5 billion on AI contracts in fiscal 2025, according to Stanford's AI Index. Yet most agencies lack a shared vocabulary to discuss what kind of AI they're building or buying. Without a taxonomy, a predictive policing algorithm and a DMV chatbot get the same label, leading to mismatched expectations and flawed oversight.

The taxonomy emerged from an 18-month study of 200 municipal, state, and federal AI deployments across 12 countries. Vance and her team coded each use case against 32 attributes—from data sensitivity to human-in-the-loop requirements—and found five natural clusters. Efficiency AI, for example, accounts for 38% of deployments and typically requires minimal human oversight, whereas Enforcement AI, at 18%, demands robust accountability mechanisms. Engagement AI (26%) raises transparency and equity questions, while Exploration AI (12%) often involves high-stakes decisions like resource allocation. Ethics AI (6%) is the newest category, encompassing tools that audit bias or explain model decisions.

Implications extend beyond classification. The taxonomy gives procurement officers a clearer checklist: is this AI meant to cut costs, interact with the public, or enforce rules? It also helps regulators design proportionate rules—a one-size-fits-all AI law would be as misguided as applying the same traffic code to bicycles and tractor-trailers. "A chatbot and a facial-recognition system pose fundamentally different risks," Vance noted in the paper. The framework is already being piloted by the U.S. Digital Service and the UK's Office for Artificial Intelligence.

Looking ahead, the taxonomy could become a standard reference for AI audits and impact assessments. The OECD has expressed interest in adapting it for its AI Policy Observatory, and the European Commission is weighing it as a lens to evaluate high-risk AI systems under the AI Act. For public administrators, the message is clear: know which face of AI you're dealing with before you deploy it.

The five-faces framework turns a vague concept into a practical tool. As AI adoption accelerates, understanding these distinct categories—Efficiency, Engagement, Enforcement, Exploration, Ethics—will separate effective governance from costly mistakes. The taxonomy arrives not a moment too soon.