Learning Objectives

After completing this lecture, you will be able to:

• Understand the key concepts of what ai is and is not in a government context

• Apply knowledge of ml, llm, nlp, computer vision, generative ai, agentic ai

• Apply knowledge of separating hype from reality

• Identify next steps for applying these concepts in your role

Key Topics Covered

• Demystifying core terminology: ML, LLM, NLP, computer vision, generative AI, agentic AI
• Separating hype from reality
• Government context for what ai is and is not
• Practical applications and next steps

Why This Matters for Government

Government agencies face unique challenges when it comes to AI adoption. This lecture addresses these challenges head-on by providing all government employees with the knowledge and frameworks needed to navigate AI in the public sector responsibly and effectively.

As part of the L1 (AI Aware) curriculum, this lecture builds on the foundational principle that every AI system in government ultimately serves citizens. Whether you are working with AI tools daily or setting strategy for your agency, understanding what ai is and is not is essential for responsible, effective government AI adoption.

======================================================================

TRANSCRIPT: What AI Is and Is Not

======================================================================

What you will learn: Core terminology demystified (ML, LLM, NLP, computer vision, generative AI, agentic AI). Separating hype from reality. Understanding what these terms actually mean.

Welcome. If you're sitting down to learn about AI because your agency is implementing new systems, or because you've heard the term so many times you want to actually understand it, you're in the right place. This is the very first lecture in the Government AI CLUB curriculum, which means there are no prerequisites and no assumptions about your technical background.

Here's what we'll do today: we're going to untangle the terminology. You've probably heard people throw around terms like "machine learning," "LLM," "neural networks," and "AI" as if everyone knows what these mean. Spoiler alert: they don't always mean the same thing, and precision matters when you're making decisions about government systems that affect real citizens.

By the end of this lecture, you'll be able to explain these concepts to a colleague in plain English, understand the difference between hype and actual capability, and know what you're dealing with when your agency is considering an AI tool.

Purpose

Let's start with the most basic question: what even is artificial intelligence? Here's the honest answer: "AI" is an umbrella term that's been stretched so far it covers everything from simple automation to systems that make decisions that affect your life. In government, precision matters. When we adopt a system affecting citizens' benefits, loan applications, or hiring decisions, we need to know exactly what we're dealing with—not marketing speak.

Why This Matters for Government

Government agencies operate under public scrutiny and accountability requirements that private companies don't face. When a private tech company's algorithm shows you ads you might like, that's one thing. When a government AI system determines whether someone gets unemployment benefits, denies a permit, or flags them for fraud investigation—that's something that affects people's constitutional rights, economic security, and civil liberties.

Understanding what AI actually is (and what it isn't) is foundational to your responsibility as a government employee. You need to know:

• What can we actually ask this system to do?

• What are its real limitations?

• What happens when it fails?

• Who is accountable when something goes wrong?

These questions have no answer if you don't understand the underlying technology, even at a basic level.

Core Concepts

1. Artificial Intelligence (the umbrella term)

"AI" is remarkably vague. Technically, it refers to any computer system that performs tasks that we might normally associate with human intelligence. But that's so broad it includes everything from chess-playing computers from the 1980s to systems that can write essays.

In government context, when someone says "we're implementing AI," you should immediately ask: "What specifically?" Because AI can mean:

• Automation (executing pre-defined rules)

• Machine learning (finding patterns in data)

• Language models (predicting text, having conversations)

• Computer vision (analyzing images)

• Decision systems (making or supporting decisions)

• Predictive systems (forecasting outcomes)

Each of these works differently, has different failure modes, and requires different governance.

1. Machine Learning (ML)

Machine learning is a subset of AI. Instead of programming a computer with explicit rules ("if temperature > 32 degreesF, label as 'freezing'"), ML systems learn patterns from data.

Here's how it works in three steps:

First, you feed the system examples. Lots of them. For a fraud detection system, you might feed it thousands of past transactions, labeled as either "fraud" or "legitimate."

Second, the system finds patterns. It doesn't understand fraud the way a detective does. It's finding statistical correlations: "transactions at 3 AM with amounts over $10,000 from unknown IP addresses tend to be labeled 'fraud' in our training data."

Third, when you show it a new transaction, it applies what it learned: "This looks 71% similar to fraudulent transactions we've seen before, so I'll flag it."

The critical point: the system is doing pattern matching on historical data. It's not reasoning. It's not truly understanding fraud. It's finding correlations. And those correlations are only as good as the data it learned from.

1. Deep Learning (DL)

Deep learning is a more sophisticated type of machine learning that uses something called "neural networks"—inspired by (but very different from) how human brains work.

Deep learning is what powers the most impressive AI systems today: image recognition that can identify objects in photos, systems that can play Go at superhuman levels, and language models that can write coherent text.

Why is it called "deep"? Because these neural networks have many layers that progressively transform data. The first layer might detect edges in an image. The next layer might detect simple shapes. The next might detect parts of objects. And eventually, deep in the network, high-level patterns emerge—like "this is a face."

What makes deep learning powerful also makes it mysterious. Even engineers can't always explain exactly what each layer is doing. It's powerful. It's often accurate. But it's also a "black box" in many ways—and that's genuinely problematic for government.

1. Large Language Models (LLM)

LLMs are the technology behind ChatGPT, Gemini, Claude, and similar systems. They're trained on enormous amounts of text from the internet.

Here's what they actually do: they predict the next word. That's not poetic license. If you give an LLM the text "The capital of France is," it calculates probabilities for what word comes next. Paris gets 99% probability. Then it "says" Paris and does it all again: given "The capital of France is Paris," what comes next? Probably a period or a comma.

This sounds trivial, but when you chain millions of these predictions together, you get something that can hold a conversation, answer questions, write essays, and code. The magic is that next-word prediction, done at scale on massive amounts of diverse text, produces something that seems to understand and reason.

But here's the catch: it's still just doing what it was statistically trained to do. It's not consulting a knowledge base. It's not looking things up. It's using patterns from its training data. And when those patterns lead it wrong, it still sounds confident while being completely incorrect. We call these errors "hallucinations," and we'll talk about them more in a later lecture.

1. Natural Language Processing (NLP)

NLP is the field that deals with human language. It includes everything from identifying the main topic of a document, to sentiment analysis (is this email angry or satisfied?), to translation, to conversation.

LLMs are one type of NLP technology, but NLP is broader. You might use NLP to:

• Extract key information from unstructured documents (automatic resume screening)

• Classify incoming requests (which department should handle this citizen question?)

• Detect sentiment in constituent feedback

• Translate documents between languages

1. Computer Vision (CV)

Computer vision is AI's ability to "see"—to analyze images and video.

It can identify objects ("this is a dog"), detect people and their activities, read text in images, and analyze patterns. In government, this might mean:

• Identifying documents (is this a driver's license, passport, or something else?)

• Analyzing satellite imagery for border security or environmental monitoring

• Detecting equipment failure from machinery images

• Analyzing video from surveillance systems

Like other AI systems, computer vision has limitations. It can be fooled by things it wasn't trained on. A system trained to detect faces might fail on people with certain skin tones if the training data wasn't diverse. It's powerful but not infallible.

1. Generative AI

Generative AI is systems that create new content: text, images, code, audio, video. ChatGPT generating an essay, DALL-E creating an image, Jukebox generating music—these are all generative.

The distinction matters because generative systems have different risks than classification systems. A system that categorizes documents as "urgent" or "non-urgent" has one set of risks. A system that generates a letter to a citizen has different risks—around accuracy, bias, and appropriateness of tone.

1. Agentic AI (Autonomous Systems)

This is emerging technology: AI systems that don't just respond to requests but take actions autonomously. They set sub-goals, execute steps, and report back.

An agentic system might: "I've been asked to streamline the vendor approval process. I'll identify bottlenecks, research best practices, draft new procedures, and flag them for human review." It's taking initiative, making decisions about what to do next, and operating with some degree of autonomy.

This is where government needs to be especially careful. Autonomy is powerful for efficiency, but government decisions require human judgment, accountability, and due process. An agentic system that autonomously approves permits or denies benefits without human review is a governance nightmare.

Practical Use Cases

Case 1: Social Security Administration (Disability Claims)

The SSA processes hundreds of thousands of disability benefit claims. Many claims require careful review of medical evidence, work history, and current job market conditions. An AI system could help by:

• Scanning medical records and extracting relevant information

• Flagging claims that are very likely to be approved or denied

• Organizing similar cases for batch processing

What it should NOT do: make the actual approval/denial decision without human review. Why? Because disability claims involve discretion, contextual judgment, and sometimes life-changing decisions. The law requires meaningful human review.

Case 2: EPA Environmental Monitoring

The EPA monitors air and water quality across thousands of locations. Computer vision could help by:

• Analyzing satellite imagery to detect pollution sources

• Identifying changes in waterways over time

• Detecting equipment problems in monitoring stations

The system is analyzing images and finding patterns. This is within AI's wheelhouse. The EPA still needs to validate findings and take action, but AI makes the monitoring process more efficient.

Case 3: VA (Veterans Benefits)

The VA receives applications from veterans seeking healthcare. NLP could help by:

• Extracting relevant information from complex application documents

• Classifying applications by type and urgency

• Identifying missing information

An LLM might draft a follow-up letter to a veteran requesting additional information. But should it? The letter affects a veteran's experience with the VA. If it's confusing or incorrect, that matters. So the VA would likely generate a draft with AI but have a human review and send it.

Anti-Patterns / Misuse Risks

Anti-Pattern 1: Assuming AI Understands

Risk: Treating an AI system as if it actually comprehends what it's doing. Asking it for advice on sensitive matters as if it has judgment.

Why It Happens: AI systems are conversational and articulate. They respond to questions thoughtfully. It feels like talking to a knowledgeable colleague. But they're not reasoning—they're doing sophisticated pattern matching.

What Goes Wrong: You ask an LLM for advice on policy interpretation. It gives a confident, well-articulated answer. You follow it. Later you discover it was completely wrong, but it sounded so sure. Now your agency is out of compliance.

Example: A FEMA employee asks ChatGPT about disaster relief eligibility rules. ChatGPT confidently explains a rule that doesn't actually exist in the statute. The employee relies on it in guidance to states. Disaster.

How to Avoid: Treat AI outputs as drafts and starting points, never as authoritative guidance on matters of policy, law, or procedure. Always verify with subject matter experts and official sources.

Anti-Pattern 2: Using AI on Sensitive Data Without Authorization

Risk: Copy-pasting classified or PII-laden documents into consumer AI tools (ChatGPT, Google Bard) to help summarize them.

Why It Happens: The tools are convenient. The employee doesn't realize the data is being sent to external servers where it could be retained, used for training, or accessed by the vendor.

What Goes Wrong: A Department of Defense contractor pastes classified specs into ChatGPT to help write a summary. The data is transmitted to OpenAI's servers. The employee thinks the conversation is deleted, but the vendor's terms allow data use for model improvement. Classified information has leaked.

Example: A VA employee pastes a veteran's full medical record (including SSN, address, health conditions) into ChatGPT to help organize it. Now that data is on external servers.

How to Avoid: Never use unapproved, consumer-facing AI tools with sensitive data. Only use AI systems your agency has vetted and approved. Know where the data goes and whether it's encrypted in transit and at rest.

Anti-Pattern 3: Trusting AI to Be Factual Without Verification

Risk: Sharing information that an AI system generated without fact-checking it first.

Why It Happens: AI systems are trained to generate plausible-sounding text. That confidence is baked in. And verifying everything takes time.

What Goes Wrong: A housing agency uses an LLM to generate talking points for a press release. The LLM includes a statistic about homelessness that sounds right but is actually off by half. The agency distributes the talking points. Media pick it up. The agency loses credibility.

Example: An employee asks ChatGPT about recent court decisions on immigration. ChatGPT cites a case that doesn't exist or misrepresents a real case. The employee puts it in a briefing. The briefing is wrong.

How to Avoid: Make verification a standard step. Use the VERIFY method we'll discuss in Chapter 3: check facts against authoritative sources, look up citations, cross-reference with official records.

Anti-Pattern 4: Deploying AI to High-Stakes Decisions Without Governance

Risk: Rolling out an AI system for decisions that significantly affect citizens without proper testing, transparency, or override mechanisms.

Why It Happens: Budget pressure. Technical enthusiasm. Assumption that if it works in testing, it'll work in the real world.

What Goes Wrong: A state welfare agency deploys an AI system to identify fraud. The system has a 92% accuracy in testing. In production, it works on a different data distribution—the live population is different from the test set. It starts flagging benefits as fraud that shouldn't be. Thousands of people have benefits cut. Lawsuits follow.

Example: The Michigan Unemployment Agency's AI system (MDHHS) flagged thousands of people as fraud perpetrators during COVID relief. The algorithm was flawed. People lost unemployment benefits they were entitled to. Lawsuits cost the state millions.

How to Avoid: Require human review for high-stakes decisions. Test on data that matches production conditions. Monitor outcomes after deployment, especially for disparate impact on protected groups. Have clear appeals processes.

Practice / Reflection Prompts

• Terminology Recognition: Find an article from your field about "AI implementation." Can you identify which of today's categories (ML, LLM, NLP, CV, generative, agentic) it's actually describing? Write a sentence clarifying what the AI system specifically does.

• Limitation Exercise: Take a task in your agency that people have suggested automating with AI. What would the AI system need to do? What contextual judgment or human expertise would it still need? Is there room for error—and if so, what would be the consequences?

• Hype Calibration: Find a quote about AI from a vendor pitch or industry article. Rewrite it in more precise language about what the technology actually does (not what it seems to do).

• Use Case Assessment: Identify one repetitive document-processing task in your unit. Would NLP/LLM help? What would need to be true for it to be worth the effort and risk?

• Governance Reflection: If your agency deployed a new AI system next month, who should know? Who should approve it? Who should check if it's working fairly? What would "working fairly" even mean?

Key Takeaways

• AI is an umbrella term covering many different technologies—from simple automation to large language models. When someone says "we're using AI," ask specifically what technology and what it does.

• Machine learning finds patterns in historical data—it doesn't reason, understand, or consult external sources. It is fundamentally a statistical system making correlations.

• Large language models predict text based on training data—they sound confident even when wrong, and they can't look anything up. They're powerful for writing and analysis, but not for factual accuracy without verification.

• Computer vision, NLP, and other specialties all have different capabilities and limitations—understanding which you're actually using prevents misapplying the tool.

• Generative AI creates new content, which introduces different risks than classification systems—hallucinations, tone, appropriateness all matter in government communications.

• Agentic AI (autonomous systems) raises governance questions unique from other AI types—because autonomy requires clear rules about what decisions require human approval.

• No AI system should be deployed to high-stakes government decisions without human oversight, testing on realistic data, and appeals mechanisms.

Terms / Glossary Items

Artificial Intelligence (AI): A broad term for computer systems that perform tasks associated with human intelligence. Includes machine learning, deep learning, language models, computer vision, and autonomous systems.

Machine Learning (ML): A type of AI where systems learn patterns from data rather than being explicitly programmed with rules. Requires training data and doesn't consult external sources.

Deep Learning (DL): A sophisticated form of machine learning using neural networks with many layers. Powers impressive systems like image recognition and language models, but is often opaque in how it reaches decisions.

Large Language Model (LLM): A system trained on enormous amounts of text that predicts the next word in a sequence. Examples: ChatGPT, Gemini, Claude. Can write, summarize, translate, and code, but is not factual without verification.

Natural Language Processing (NLP): The field addressing human language understanding and generation. Includes classification, extraction, summarization, translation, and conversation.

Computer Vision (CV): AI's ability to analyze images and video. Can identify objects, detect patterns, read text in images, and analyze visual information.

Generative AI: Systems that create new content (text, images, audio, video) rather than just classifying or analyzing existing content.

Hallucination: When an AI system (especially language models) generates false information while sounding confident. It's not lying—it's extrapolating beyond its training data.

Today we've built a common vocabulary. You now know what people mean (and don't mean) when they say AI. This foundation matters because it lets you ask smart questions:

• What specifically is this system doing?

• What data did it learn from?

• How accurate is it in testing versus real-world conditions?

• What happens when it fails?

• Who decides if it's deployed?

The next five lectures in Chapter 1 zoom in on different aspects: how AI systems actually learn, what types of systems exist and where government uses them, what AI does well versus where it fails, real government deployments you might not know about, and the timeline from early AI research to today.

As you progress through the Government AI CLUB curriculum, you'll build on today's vocabulary. Each lecture assumes you know these terms. And you'll move from just understanding what these systems are to understanding how to evaluate them, use them responsibly, and oversee them in your agency.

Take two minutes. Think of an AI system you've heard about in your work context or news. Maybe your agency mentioned adopting one. Maybe a contractor pitched one. In your head or on paper, answer three questions:

• What does it actually do? (Be specific—not "it helps," but "it predicts X or generates Y or classifies Z")

• Which category from today does it fit? (ML? LLM? Computer vision? Something else?)

• What would I need to verify before trusting it with something important?

These questions will guide your critical thinking throughout this program.

That's lecture 1.1.1 of the Government AI CLUB. You've just learned the foundational vocabulary that everyone else will use for the next 214 lectures.

Next up: How AI Actually Works. We'll go deeper into the mechanics—how systems learn, what training data means, and why the data you feed a system determines what it becomes.

Start applying this immediately. When you hear "AI" in a meeting, ask clarifying questions. Use the vocabulary. Help your agency communicate more precisely about what it's actually doing. That precision is where good governance starts.

You're building competence in a critical domain. The next step is yours. See you in the next lecture.

Government AI CLUB Certification Program

Level 1: AI Aware | What AI Is and Is Not | Lecture 1.1.1

A GOVT.CLUB initiative.

1.1.2 How AI Actually Works ->

Start Your CLUB Certification

This lecture is part of L1: AI Aware—8 hours of comprehensive government AI training.

Explore CLUB Certification

Related Lectures

L1 1.1.2—How AI Actually Works 20 min - Video + Interactive

L1 1.1.3—Types of AI Systems 20 min - Video + Reading

L1 1.1.4—What AI Does Well and Where It Fails 20 min - Video + Case Studies

Frequently Asked Questions

What will I learn in What AI Is and Is Not?

In this 20 min video + reading lecture, you will Demystifying core terminology: ML, LLM, NLP, computer vision, generative AI, agentic AI. Separating hype from reality

What level is What AI Is and Is Not?

This is a Level 1 (AI Aware) lecture, part of Chapter 1.1 \u2014 AI Foundations. It is designed for all government employees.

How long is lecture 1.1.1?

Lecture 1.1.1 (What AI Is and Is Not) takes 20 min. It is delivered as a video + reading format.

Do I need prerequisites for What AI Is and Is Not?

This lecture is part of L1 (AI Aware). Prerequisites: None.

What is the CLUB Certification?

CLUB (Community Leading Unified Benchmarks) is a maturity-based AI certification for government professionals with 5 levels (L1-L5), 215 lectures, and 25 chapters aligned with NIST AI RMF, OMB, and GAO frameworks.