Artificial Intelligence (AI)

Software that can perform tasks typically requiring human judgment—recognizing patterns, generating text, making predictions, translating languages. In this book, AI primarily refers to generative AI and predictive tools used by local government.

Automation Bias

The tendency of people to defer to computer-generated recommendations even when those recommendations are wrong. Research shows government workers disagree with algorithmic outputs only 3.2 percent of the time.

Bias (in AI)

When an AI system produces systematically unfair results for certain groups. AI bias usually comes from biased training data—if the data reflects historical patterns of inequality, the AI will reproduce and amplify those patterns.

Chatbot

A software application that simulates conversation with users. Older chatbots selected from pre-written responses. Generative AI chatbots construct novel responses, which can be helpful but also unpredictably wrong.

Copilot

Microsoft’s brand name for AI features embedded in Office 365 and other products. Not a separate product you buy—it’s built into software you may already have.

Data Portability

The ability to export your data from one system in formats that other systems can use. Important for avoiding vendor lock-in.

Enterprise AI Tools

AI products licensed and configured specifically for organizational use, typically with stronger data protections than free consumer tools. Example: ChatGPT Enterprise vs. free ChatGPT.

Generative AI

AI that creates new content—text, images, translations, summaries—rather than simply analyzing existing data. ChatGPT, Google Gemini, and Microsoft Copilot are generative AI tools.

GovAI Coalition

A coalition of 700+ government agencies, launched by San José, California, that shares AI governance resources including policy templates, vendor evaluation tools, and contract language. Free to join.

Hallucination

When an AI tool generates information that sounds plausible but is factually wrong—including fabricated references, statistics, or events. All generative AI tools can hallucinate. This is why human review of AI output is essential.

ICMA

The International City/County Management Association—the professional association for local government managers. Conducted the 2024 AI survey that is the primary data source for this book.

Jagged Technological Frontier

A concept from research showing that AI’s competence varies sharply across tasks that appear similar. AI may be excellent at drafting a standard letter but terrible at drafting one requiring political nuance. The frontier is “jagged”—you can’t predict where AI will succeed or fail based on task similarity.

Lock-In (Vendor Lock-In)

A situation where switching away from a vendor is difficult or expensive because your data, processes, or workflows depend on their proprietary system. Negotiate data portability and exit provisions to reduce lock-in risk.

Model

In AI, a model is the trained system that generates outputs. When people say “ChatGPT uses the GPT-4 model,” they mean the underlying AI system that powers the tool. Different models have different capabilities and limitations.

NACo

The National Association of Counties. Published the AI County Compass toolkit.

NIST AI Risk Management Framework

A framework developed by the National Institute of Standards and Technology for identifying and managing AI risks. The GovAI Coalition’s governance templates are aligned with this framework.

PEARS Framework

Predictable, Explainable, Accountable, Reversible, Sensitive to equity. A quick-reference checklist used in this book. Run through these five questions before deploying any AI tool.

Perception-Reality Gap

The documented tendency for AI users to overestimate productivity gains. Research found that experienced workers believed AI made them 20 percent faster when it actually made them 19 percent slower—a gap of nearly 40 percentage points.

Predictive AI

AI that analyzes data to forecast outcomes, classify information, or assess risk. Examples: crime prediction, risk scoring, demand forecasting. Distinguished from generative AI, which creates new content.

Prompt

The text you type into a generative AI tool to tell it what you want. The quality of the prompt significantly affects the quality of the output.

Three-Tier Framework

The Urban Institute’s classification of AI applications by risk level, used throughout this book. Tier 1: Internal productivity tools (low risk). Tier 2: Resident-facing applications (medium risk). Tier 3: Decision support affecting individual rights (high risk).

Training Data

The information an AI system learns from. If training data is biased, incomplete, or unrepresentative, the AI’s outputs will reflect those limitations.

Transparency (in AI)

The ability to explain how an AI system works, what data it uses, and how it reaches its outputs. Essential for public accountability and trust.

Vendor FactSheet

A standardized document (pioneered by San José) that AI vendors complete to describe their product’s data use, testing, performance metrics, and limitations. A tool for consistent vendor evaluation.

© 2026 Alton Henley. Ready or Not.
altonhenley.com · LinkedIn