Chapter 2: Engaging with AI – Accessing Information and Expanding Horizons

In Active Review, Subject to Change

This content is currently in active review and is subject to change. Check back for updates!

Engaging with AI 🔍

Accessing Information and Expanding Horizons

In Chapter 1, we established that you are already interacting with Artificial Intelligence every single day. The question is no longer whether you will use AI, but how you will use it. The first domain of the AILit Framework is Engaging with AI. This domain is all about being a smart, intentional consumer of AI-driven tools — whether you are asking a chatbot to explain a difficult physics concept or swiping through a customized video feed.

When you simply accept whatever an algorithm hands you, you are a passive consumer. To become an active participant in the Algorithmic Age, you need to understand how these systems process information, why they show you specific content, and how to tell the difference between digital truth and algorithmic fiction. Engaging with AI effectively means treating these systems not as all-knowing oracles, but as incredibly fast, highly opinionated digital librarians that need careful instruction and constant supervision.

---

2.1 — AI as a Search Engine and Tutor 📚

We are used to using traditional search engines like Google. You type in a few keywords, hit enter, and receive a list of web pages to read. But modern AI chatbots — like ChatGPT, Gemini, or Claude — represent a massive leap forward. These systems are powered by Large Language Models (LLMs) Large Language Models (LLMs): Powerful AI systems trained on enormous amounts of text data. They generate human-like text by predicting the most probable next word in a sequence, one word at a time. .

To understand an LLM, you have to understand that it does not actually “know” facts the way a human does. Instead, it is essentially a highly advanced version of the autocomplete feature on your phone. Programmers feed these models massive amounts of text from the internet — books, articles, websites, and forums. The AI analyzes all this data to learn the statistical patterns of human language. When you ask it a question, it isn’t looking up the answer in a database; it is predicting the most likely next word, and then the next, and then the next, based on the patterns it learned during its training.

Because they understand conversational language, LLMs can be incredible tutors. Instead of reading a static article, you can have a dynamic back-and-forth conversation. If you don’t understand how cellular respiration works, you can ask the AI to explain it. If it’s still too complicated, you can ask it to explain it again — but this time using an analogy about a power plant, or to summarize it for a 5th grader.

The Art of the Prompt ✍️

The secret to engaging with LLMs effectively is mastering the prompt prompt: The instruction or question you give to an AI system. Because AI predicts text based on your input, a vague prompt produces a vague, generic answer. . A prompt is the instruction or question you give the AI. Because the AI is predicting text based on your input, a vague prompt will get a vague, generic answer. Prompt engineering Prompt engineering: The skill of crafting clear, specific, and structured instructions to get high-quality, useful outputs from an AI model. is the skill of crafting clear, specific, and structured instructions to get exactly what you want from the machine.

Think of the AI as a brilliant but literal-minded actor. If you just say “talk about space,” it won’t know what character to play or what script to read. But if you say:

“Act as an expert astrophysicist and explain the concept of black holes in three simple bullet points, using a conversational tone.”

…you give the model the context, the role, the format, and the tone it needs to generate a high-quality response.

Prompt Element	What It Does	Example
Role	Tells the AI what persona to adopt	”Act as a high school chemistry teacher…”
Context	Provides background information	”I am a 10th grader who just learned about…”
Format	Specifies how to structure the output	”…in three bullet points”
Tone	Sets the style and voice	”…using a conversational tone”
Constraints	Sets boundaries on the response	”…in no more than 200 words”

---

2.2 — Navigating AI Recommendations 📱

You don’t just engage with AI when you type into a chatbot; you also engage with it every time you open a streaming service or a social media app. Recommendation algorithms Recommendation algorithms: AI systems that analyze your past behavior to suggest what you might want to see, hear, or buy next. Their primary goal is to maximize the time you spend on a platform. are the hidden engines driving platforms like Spotify, TikTok, YouTube, and Netflix. Their primary goal is to keep your attention on the screen for as long as possible — to do this, they have to predict exactly what you want to see next.

These algorithms generally use two main strategies:

• Content-based filtering Content-based filtering: A recommendation strategy that looks at the characteristics of things you already like and finds similar content to show you next. : Looks at the characteristics of the things you already like. If you watch a lot of fast-paced superhero movie trailers, the algorithm will find other fast-paced superhero content to show you.

• Collaborative filtering Collaborative filtering: A recommendation strategy that finds other users with similar tastes to yours and recommends content that those users enjoyed but you haven't seen yet. : Looks at other users who have similar tastes to you. If you and a million other teenagers consistently listen to the same three pop artists, the algorithm assumes you share a similar profile. If those users start listening to a brand new artist, the algorithm will automatically recommend them to you.

Filter Bubbles and Echo Chambers 🫧

While personalized feeds can be highly entertaining, they come with a significant risk. Because the algorithm’s goal is to keep you comfortable and engaged, it will rarely show you content that challenges your beliefs or upsets you. Over time, this creates a filter bubble filter bubble: An invisible digital boundary created by personalization algorithms where you are only exposed to information, news, and opinions that align with what you already believe. — an invisible digital boundary where you are only exposed to information and opinions that align with what you already think.

When a filter bubble involves politics or social issues, it can harden into an echo chamber echo chamber: A situation where a person only encounters information or opinions that reflect and reinforce their own, making it seem like the whole world agrees with them. . If you click on an article expressing a specific political viewpoint, the algorithm will feed you more of the same. Eventually, your entire feed reflects that single viewpoint, making it seem like the whole world agrees with you — and anyone who disagrees must be wrong.

Key Takeaway

Engaging with AI responsibly requires actively breaking out of filter bubbles by deliberately seeking out different perspectives and intentionally searching for topics outside your usual habits.

---

2.3 — Evaluating AI Outputs: Truth, Hallucinations, and Bias ⚖️

Because AI systems can write so eloquently and instantly generate authoritative-sounding answers, it is incredibly easy to fall into the trap of believing everything they say. However, critical thinking is more vital now than ever.

AI Hallucinations

The most common trap is the AI hallucination AI hallucination: When an AI system generates information that sounds confident and plausible but is actually incorrect, fabricated, or not grounded in real facts. . Remember that an LLM is a prediction engine trying to guess the next logical word. Sometimes — especially when asked about obscure facts or highly specific details — the AI will predict a sequence of words that sounds perfectly natural but is entirely factually incorrect. It isn’t lying to you maliciously; it is simply generating a grammatically correct fabrication.

It will confidently:

• 📅 Invent historical dates
• 📖 Summarize books that don’t exist
• 🔬 Cite fake scientific studies

You cannot trust an AI to be your sole source of truth — you must always verify important claims using reliable, human-authored sources.

Bias in AI Outputs

Furthermore, we have to consider algorithmic bias algorithmic bias: When an AI system produces results that are systematically skewed or prejudiced because of flaws, gaps, or historical prejudices embedded in its training data. . An AI is only as good as the data it was trained on. Because these models are trained on texts written by humans across the vast expanse of the internet, they absorb human prejudices, stereotypes, and historical biases.

If an AI is asked to generate a story about a doctor and a nurse, it might default to making the doctor male and the nurse female — reflecting historical gender biases present in its training data.

Engaging with AI means constantly questioning the output:

• ❓ Who created the data?
• ❓ What perspectives might be missing?
• ❓ Does this reflect a specific cultural or historical viewpoint?

By keeping a skeptical eye on AI outputs, you protect yourself from being misled by both honest mistakes and deeply ingrained societal biases.

---

2.4 — Advanced Prompt Engineering Techniques 🔬

Now that you understand the basics of prompting, it’s time to go deeper. Expert prompt engineers don’t just add a few adjectives to their questions — they apply a systematic set of techniques that dramatically improve the quality, precision, and usefulness of AI outputs. These techniques are the difference between a mediocre AI interaction and a genuinely powerful one.

Technique 1: Chain-of-Thought Prompting

One of the most powerful and well-documented techniques is chain-of-thought prompting chain-of-thought prompting: A prompting technique where you instruct an AI to reason through a problem step-by-step before arriving at a final answer. Adding phrases like 'Think step by step' dramatically improves performance on math, logic, and complex reasoning tasks. . Research from Google Brain published in 2022 demonstrated that simply adding the phrase “Think step by step” to prompts requiring reasoning dramatically improved AI performance on math, logic, and multi-step problems.

Why does it work? Because LLMs generate tokens (words) sequentially, asking the model to articulate intermediate reasoning steps forces it to build a logical scaffold before committing to a final answer — catching errors it would otherwise make by “jumping” to a conclusion.

Without Chain-of-Thought	With Chain-of-Thought
”If a train travels at 60mph for 2.5 hours, how far does it go?” → May produce an incorrect number	”If a train travels at 60mph for 2.5 hours, how far does it go? Think step by step.” → Explicitly calculates 60 × 2.5 = 150 miles

Technique 2: Few-Shot Examples

Instead of just describing what you want, show the AI examples of what you want. This technique — called few-shot prompting — is extraordinarily effective because it communicates format, tone, and quality through demonstration rather than description.

Example:

“I’m going to give you two examples of the kind of writing I want. Then I’ll give you a new topic, and I want you to match the style exactly.

Example 1: ‘The Industrial Revolution didn’t just change factories — it rewired the human concept of time itself, turning sunrise and sunset into irrelevant relics.’

Example 2: ‘Social media didn’t invent loneliness; it just gave it a faster delivery system.’

Now write a sentence about: artificial intelligence.”

Technique 3: The Full 5-Element Framework

Section 2.1 introduced you to the five prompt elements. Here is what a fully assembled prompt looks like in practice:

Role: “Act as a veteran science journalist writing for The Atlantic.” Context: “The audience is educated adults with no computer science background.” Task: “Explain why AI systems sometimes ‘hallucinate’ false information.” Format: “Write three paragraphs. Open with a compelling real-world example. End with a practical takeaway.” Constraints: “Do not use technical jargon. Do not use bullet points. Keep total length under 300 words.”

Every element earns its place. Remove any one and the output degrades.

Technique 4: Iterative Refinement

The most powerful prompting technique is also the most underused: treat your interaction as a conversation, not a transaction. Start broad, then narrow.

1. “Explain how the immune system works.” (Get an overview)
2. “Now focus specifically on how T-cells identify infected cells.” (Zoom in)
3. “Explain the same thing using an analogy about airport security.” (Change the frame)
4. “That analogy is great — now use it to explain the difference between innate and adaptive immunity.” (Build on success)

Each step refines and sharpens the output. Professional AI researchers call this process “steering” — you’re not just asking questions, you’re guiding the model toward exactly what you need.

Technique 5: Negative Constraints

Finally, learn the power of telling the AI what not to do. LLMs have strong default tendencies (bullet points everywhere, excessive hedging, hollow phrases like “Certainly!” and “Great question!”). Negative constraints short-circuit these defaults:

• “Do not use bullet points.”
• “Do not include caveats or disclaimers.”
• “Do not start your response with ‘Certainly’ or ‘Of course’.”
• “Do not repeat the question back to me.”

Practice Challenge

Take a prompt you’ve used before and apply all five techniques to it. Compare the output to what you got originally. Which technique made the biggest single difference for your use case?

---

2.5 — Fact-Checking AI Outputs: The SIFT Method 🔎

Given that AI systems hallucinate confidently and absorb the biases of their training data, how do you protect yourself from being misled? The answer comes from the world of media literacy: a framework called SIFT method SIFT method: A four-step media literacy framework (Stop, Investigate the source, Find better coverage, Trace claims) developed by Mike Caulfield to help people evaluate information before sharing or accepting it. Originally developed for social media, it applies powerfully to AI-generated content. , developed by digital literacy researcher Mike Caulfield. Originally designed for evaluating social media posts, SIFT applies with even greater urgency to AI outputs.

S — Stop

Before you accept or act on surprising, emotionally charged, or highly specific information from an AI, stop. The AI’s confident, authoritative tone is designed to feel trustworthy — but that tone is simply a product of its training data, not a guarantee of accuracy. The emotional response you feel (relief, excitement, alarm) is precisely the moment to pause.

I — Investigate the Source

AI outputs don’t have sources in the traditional sense — the model synthesizes patterns from its training data and rarely tells you exactly where a specific claim came from. So instead, ask: Who would know if this were true? Then go find those sources independently. What institutions, journals, or experts have established credibility in this domain?

F — Find Better Coverage

Do not rely solely on the AI’s answer. Find multiple independent, human-authored sources that confirm or deny the claim. If three reputable sources agree with the AI, you can have higher confidence. If no source can be found to confirm the claim, treat it with significant skepticism — it may be a hallucination.

T — Trace Claims to Their Origin

When an AI cites a statistic or a study, it is often correct that such research exists — but the details are frequently garbled. A real study may have found something slightly different, may have been retracted, or may be far less certain than the AI implies. Follow the claim back to the primary source. Is the original study actually real? Does it actually say what the AI claims it says?

Applying SIFT: A Worked Example

Imagine an AI confidently tells you: “Studies show that humans only use 10% of their brains.”

• Stop: This is a well-known claim. It feels familiar and authoritative. But familiarity isn’t the same as truth.
• Investigate the source: Who are the domain experts? Neuroscientists. Search for what neuroscientists say about this claim.
• Find better coverage: Multiple peer-reviewed sources from institutions like Johns Hopkins and MIT immediately debunk this as a persistent myth. Brain imaging studies show activity throughout the entire brain.
• Trace to origin: The “10% myth” appears to have originated from misquotations of William James and was popularized by self-help books. There is no peer-reviewed study that supports it.

Verdict: The AI confidently stated a myth. SIFT caught it in under three minutes.

Critical Rule

Never cite an AI as a source. Cite the human-authored sources you found while verifying the AI’s claim. The AI is a starting point for inquiry, not the end of it.

---

2.6 — The Attention Economy and Your Brain 🧠

To truly engage with AI wisely, you need to understand the business model you are operating inside of. Every free platform you use — TikTok, YouTube, Instagram, Snapchat — runs on the same economic engine: the attention economy attention economy: An economic system in which human attention is the primary scarce resource. Companies profit by capturing user attention and selling it to advertisers. The longer you watch, scroll, or engage, the more money the platform earns. .

Human attention is finite. There are only 24 hours in your day, and advertisers want to reach you during as many of those hours as possible. Platforms profit by capturing your attention and selling it. This means that every design decision — every autoplay, every infinite scroll, every notification — is engineered to claim as much of your conscious awareness as possible.

In 2017, Facebook’s founding president Sean Parker gave a candid public admission: “The thought process was: ‘How do we consume as much of your time and conscious attention as possible?’” He went on to describe Facebook as designed to exploit “a vulnerability in human psychology” — and admitted that he and other early executives knew what they were building and built it anyway.

The Slot Machine Principle

The core psychological mechanism platforms exploit is the variable reward schedule — a concept discovered by behavioral psychologist B.F. Skinner. Skinner found that if you give a rat food every time it presses a lever, the rat will only press when hungry. But if you give the rat food sometimes — unpredictably — the rat will press the lever compulsively, over and over, even when not hungry.

A variable reward schedule variable reward schedule: A behavioral psychology principle where unpredictable rewards are more addictive than predictable ones. Social media platforms are deliberately designed around variable reward schedules — sometimes you get likes, sometimes you don't — to create compulsive checking behavior. is precisely what social media platforms deploy. Sometimes you pull down to refresh and see something amazing. Sometimes you get nothing. That unpredictability is not a bug — it is the core design feature. It is the reason you check your phone first thing in the morning before you even get out of bed.

Social Validation and Dopamine

When you post something and receive likes, comments, or shares, your brain releases dopamine — the same neurotransmitter released when you eat something delicious, win a game, or receive a genuine compliment from a friend. The neurological response to a like notification is real, measurable, and powerful.

Platforms have tuned their notification systems specifically to maximize this dopamine hit. The red notification badge on your phone is not red by accident — red is the color associated with urgency and danger, triggering a more powerful neurological response than, say, beige.

FOMO Amplification

Finally, algorithms specifically surface content designed to make you feel like you are missing out on something. Party photos when you’re at home. Exciting experiences when you’re studying. Heated debates when you’re trying to relax. FOMO — the Fear of Missing Out — drives you back to the platform to make sure you haven’t missed anything crucial.

Key Takeaway

Knowing these mechanisms does not make you immune to them — your brain is your brain. But it does give you the awareness to catch yourself, step back, and ask: Am I here because I chose to be, or because the algorithm pulled me back?

---

2.7 — Who Controls the Information? 📡

The question of who controls information has always been central to human civilization. The printing press, the telegraph, radio, television — each technology reshuffled the power to shape public consciousness. AI is reshuffling it again, at unprecedented speed.

Platform Monopolization

Today, a remarkably small number of companies control the information diet of most of the world. Google handles approximately 92% of all internet searches globally. Meta (Facebook and Instagram) reaches roughly 3 billion people. TikTok has over a billion active users. These companies do not just distribute information — their algorithms determine what information is visible, effectively acting as the most powerful editorial decisions in human history, made by machines optimizing for engagement rather than accuracy, depth, or public good.

News Deserts

Local journalism has been collapsing for over a decade, accelerated by AI-generated content flooding digital platforms. Between 2005 and 2024, more than 3,000 local newspapers in the United States closed. “News deserts” — communities with no local reporting of any kind — have spread across the country. Without local journalism, no one is watching city council meetings, school board decisions, or local government contracts. AI cannot replace this accountability function — it can only optimize what already exists.

The Death of the Editorial Gatekeeper

For most of the 20th century, professional editors at newspapers and broadcasters served as quality gatekeepers. Their job was to evaluate: Is this true? Is it fair? Is it important? They were imperfect and biased in their own ways, but they applied human judgment to the question of what the public should know.

Algorithms have replaced editorial judgment with engagement metrics. Content gets distributed not because it is true, fair, or important — but because it is likely to make you react. Outrage, fear, and tribal satisfaction reliably generate more clicks than nuance. The algorithm amplifies whatever gets the reaction, regardless of its accuracy.

AI-Generated Disinformation at Scale

Perhaps most alarming: what once required an entire PR firm, a film crew, and a professional writer can now be produced by one person with a laptop and a free API key. AI can generate:

• Realistic fake news articles indistinguishable from legitimate journalism
• Deepfake videos of real people saying things they never said
• Hundreds of fake social media accounts with plausible backstories
• Personalized emotional manipulation targeting individuals based on their data profiles

This is not a future threat. It is happening right now, in every major democracy. AI literacy — specifically, the skills in sections 2.3 through 2.5 — is your primary defense.

---

Chapter 2 Project: The Prompt Engineering Challenge 🧪

The Challenge: You are going to test how the quality of a prompt changes the quality of an AI’s output by acting as an interrogator trying to get the best possible explanation of a complex historical event.

Step 1: Choose a Topic

Pick a complex historical event you are currently studying (e.g., the causes of World War I, the Industrial Revolution, the Civil Rights Movement).

Step 2: The Weak Prompt

Open a free AI chatbot (like ChatGPT, Gemini, or Claude). Type a very basic, vague prompt:

“Tell me about [Your Topic].”

Save the response.

Step 3: The Engineered Prompt

Now, craft a highly specific prompt. Give the AI a persona, a target audience, a format, and specific constraints. For example:

“Act as a high school history teacher. Explain the top three causes of [Your Topic] to a 9th-grade student. Use bullet points and include one analogy related to modern high school life to make it easy to understand. Do not exceed 250 words.”

Step 4: Analyze the Difference

Compare the two responses. In a short reflection paragraph, explain exactly how the engineered prompt improved the usefulness, tone, and clarity of the information provided.

Bonus Challenge

Try a third prompt that adds one more element of specificity — for example, asking the AI to include a counterargument or write from a specific historical perspective. How much does each additional layer of detail improve the output?

Step 5: Class Showcase

Present your best prompt and your worst prompt to the class. Share the AI output from each and explain what made one dramatically better than the other.

Class Vote: After all presentations, the class votes on:

• 🥇 Most Effective Prompt — which student got the highest-quality AI output?
• 🤔 Most Interesting Failure — which weak prompt produced the most amusingly bad result?

Peer-Review Rubric: When evaluating a classmate’s prompt presentation, use these four criteria (score each 1–4):

Criterion	1 (Needs Work)	4 (Excellent)
Specificity	Prompt is vague and generic	Prompt uses all five elements precisely
Quality Difference	Weak and strong outputs are similar	Strong output is dramatically superior
Analysis Depth	Student says “it was better”	Student explains exactly why each element helped
Creativity	Topic is obvious and simple	Topic is complex and reveals interesting AI behavior

---

Key Concepts Checklist

• I can explain how Large Language Models (LLMs) generate responses
• I understand the five key elements of an effective prompt
• I can describe the difference between content-based filtering and collaborative filtering
• I can define “filter bubble” and “echo chamber” and explain how they form
• I know what an AI hallucination is and why it happens
• I can identify signs of algorithmic bias in AI-generated content
• I know how to critically evaluate an AI output using external sources
• I can apply all five advanced prompt engineering techniques
• I can use the SIFT method to fact-check an AI-generated claim
• I can explain how the attention economy and variable reward schedules affect user behavior
• I understand the role of platform monopolization in shaping the information ecosystem

---

📖 Love what you're reading?

Get AI Literacy: Navigating the Algorithmic Age with activities, worksheets, and bonus content — perfect for classrooms and homeschool families.

Get the Full Textbook

☐ Mark Chapter Complete

🍎 For Educators

Teaching this chapter? Get the full grading rubrics, slide decks, and lab asset bundles in the Educator Portal.

Visit the Educator Portal →