Week 1: Building an AI Mindset

Starting the course

This is Week 1, so there is no earlier week to build on yet. This week is the foundation for everything that follows. Nothing here requires you to have used any AI tool before. By the end of this week, you will understand what AI actually is, what it is not, where it already shows up in daily life and work, what it does well, what it gets wrong, and how to think about it in a way that keeps you in control.

Learning objectives

By the end of this week, you will be able to:

1. Explain, in plain language, what artificial intelligence is and what it is not, including large language models, using the terms prompt, token, parameter, and context window.
2. Give at least three examples of how AI acts as an equalizer for blind and low vision professionals.
3. Tell the difference between AI fear, AI hype, and the healthy "pilot and co-pilot" middle path.
4. List at least three ethical practices for honest, responsible AI use, including verification, disclosure, and privacy.
5. Describe at least four real-world places AI already shows up: in the workplace, in accessibility tools such as image description and OCR, in customer service, and in daily life.
6. Explain why AI is particularly strong at drafting, summarizing, coding, and describing images, and connect that strength to how AI generates text.
7. Define hallucination, bias, privacy risk, inaccessible output, and over-reliance, and give an example of each.
8. Describe at least three independent techniques for verifying AI output using a screen reader.
9. Explain what can make AI-generated output inaccessible to a screen reader, and how to ask for accessible output instead.
10. Identify at least three areas of your own life or work where AI could realistically help you.
11. Complete a first hands-on interaction with an AI chatbot using only the keyboard, and read its response with your screen reader.

Lesson

What artificial intelligence really is

Artificial intelligence, usually shortened to AI, is a general name for computer systems that can perform tasks that normally need human intelligence, such as recognizing patterns, understanding language, answering questions, and making predictions. AI is not one invention; it is an umbrella term covering many technologies built for many jobs. Some AI systems recognize speech, some recognize images, and some generate new text, like the chatbots you will learn about in this course, such as Claude, ChatGPT, and Gemini, which belong to a specific kind of AI called a large language model, or LLM, explained in plain language shortly.

The most important idea to hold onto is this: AI systems learn patterns from very large amounts of data, instead of following one fixed list of rules written by a programmer for every possible situation. A traditional computer program might follow a rule like "if the user types the word refund, show the refund policy page." An AI system, by contrast, is trained on huge amounts of example text and learns general patterns from those examples. This is why AI can respond to questions it has never seen before in that exact wording, and it is also why AI can get things wrong in surprising ways, covered later in this lesson.

Think of AI as a very well-read assistant who has skimmed an enormous number of books, articles, and websites, but who sometimes misremembers a detail or fills in a gap with a guess that sounds confident. You likely already use several kinds of narrow AI without calling it that: a weather app, a spam filter, or a phone keyboard that predicts your next word. A chatbot is a far more general kind of AI, trained to hold open-ended conversations about almost any topic, and it is the kind this course focuses on most, because it has the broadest impact on employment and daily independence.

What AI is not

Popular movies and news stories have created confusion about what AI actually is. Today's AI systems do not have consciousness, feelings, or self-awareness. They do not experience curiosity or satisfaction, and cannot truly want anything, even though their responses can sound warm or apologetic; that tone is a pattern learned from human writing, not a sign of an inner emotional life.

AI does not understand language the way a person does. When you ask a question, it is not grasping the meaning the way a human listener would; it is predicting which words are most likely to come next, based on patterns learned during training. The result usually reads as if the AI understood you perfectly, but the underlying process is prediction, not human-style understanding. Most AI chatbots also do not remember your previous, separate conversations unless a memory feature has been turned on; each new conversation typically starts fresh.

Most importantly for how you will use AI in this course: AI is not always right. When an AI system states false information with complete confidence, this is called a hallucination, such as a made-up statistic, a fake quote, an incorrect date, or a citation to a source that does not exist. Every claim AI makes should be treated as a draft to check, not a fact to trust blindly, a skill you will practice directly in this week's exercises.

Large language models in plain language

An LLM is trained on an enormous amount of text, often a large portion of the publicly available internet, plus books and articles. During training, the model gradually adjusts millions or billions of internal numbers, called parameters; you do not need the math, just know these are the internal settings tuned during training that let the model produce useful responses afterward.

Once trained, an LLM generates its response one small piece of text at a time. These pieces are called tokens: a token might be a whole short word, part of a longer word, or a punctuation mark. The model looks at everything typed so far, including your question, predicts the single most likely next token, adds it to the text, and repeats the process extremely quickly until it has built a complete response, often called next-token prediction.

The text you type to an AI system is called a prompt; the text it generates back is its response. Every AI chatbot also has a context window: the amount of text the model can consider at one time, including your prompt, its previous responses, and any files you have shared. Once a conversation grows longer than the context window, the oldest parts may be dropped or summarized. Think of the context window as the model's short-term working memory for the current conversation.

One more term worth knowing: multimodal AI. Some AI systems only read and write text; multimodal systems can also work with images and sometimes audio. A multimodal AI can look at a photograph you share and describe what is in it, one of the most useful features for blind and low vision learners, covered next. What matters here is having a plain-language map of the vocabulary: prompt, token, parameter, context window, and multimodal.

Why AI is a powerful equalizer for blind professionals

Screen readers already opened up computers, phones, and the internet to blind and low vision users long before AI chatbots existed. AI adds an entirely new layer of independence on top of that foundation, and this is one of the central reasons this course exists. Three examples show why. AI image description can look at almost any photograph, whether a family photo or a diagram in a work document, and generate a detailed description on demand, without waiting for a sighted person to write good alt text, the short description a screen reader needs to announce an image. AI-powered optical character recognition, or OCR, can read printed or handwritten text out of a photo, such as mail, a menu, or a medication label, something older OCR tools struggled with unless the page was perfectly scanned. And AI tools can explain, in plain spoken language, how a complex document or webpage is organized, helping you build a mental map before you navigate it with your screen reader. Both of these tools are covered in far more depth later in this lesson.

Historically, many visual tasks required asking a sighted person for help: reading a printed letter, checking a photo before sending it, or reviewing a document's formatting. AI reduces how often that kind of help is necessary. That reduction in day-to-day reliance on sighted assistance for routine visual tasks is exactly why this lesson calls AI a powerful equalizer for blind professionals. It does not eliminate every barrier, and it is not perfect, but it shifts real independence back into your own hands, and it matters for employment too: employers increasingly value workers who can use AI tools to draft documents, handle data, and communicate clearly, and a blind professional with strong AI skills, built on the screen-reader fluency you already have, is a highly competitive candidate.

Picture a common workplace moment: a coworker emails you a scanned PDF with no accessible text layer, just a picture of a page. A few years ago, your only realistic options were to ask someone to read it aloud or wait for an accessible version. Today, you can open the scanned file, ask an AI tool to read the text and describe any charts or layout, and have a working understanding of the document in less time than it would take to track down a coworker. You still decide what to do with that information, but AI removed a barrier that used to depend entirely on someone else's schedule.

Overcoming fear and hype

Two opposite reactions to AI are worth noticing and avoiding. The first is AI fear: the belief that AI is dangerous, will replace you, or cannot be trusted at all, often coming from unfamiliarity or frightening headlines rather than firsthand experience. Practice is the best cure: the more you use AI tools in small, low-stakes ways, the more accurately you will judge what they can and cannot do. The second is AI hype: the belief that AI is magical, always correct, and needs no oversight. Hype ignores everything you just learned about hallucination and AI's lack of true understanding.

The healthy middle path treats AI as a capable but imperfect assistant, similar in spirit to a calculator: a genuinely useful tool, not a person and not an authority. A calculator multiplies large numbers far faster than you can by hand, but you still decide what problem to solve and still notice if the answer looks obviously wrong. AI works best under that same confident, engaged supervision; it does not replace your critical thinking, it supports it, if you let it.

One image will serve you well throughout this course: think of yourself as the pilot, and AI as your co-pilot. A co-pilot handles a great deal of the workload and speeds things up, but the pilot always makes the final call and is accountable for the outcome. That is the relationship to build with AI: confident, engaged, and always in charge of the final decision. Curiosity will take you further than fear, so this course has you try small, low-stakes AI tasks starting this very week, so comfort builds through direct experience rather than worry.

Ethics and honesty when using AI

Using AI well is not only a technical skill; it is an ethical one. Always verify facts an AI gives you, especially names, dates, numbers, and medical, legal, or financial information, since a wrong answer can sound exactly as certain as a correct one. Be honest about when and how you used AI, especially when a school or workplace policy asks you to disclose it; submitting AI-written work as entirely your own can violate a policy even if the writing is good. Respect other people's privacy: do not paste someone else's private medical, financial, or personal information into an AI tool without permission, since many AI systems process what you type on remote servers, so treat that text the way you would treat an email sent to a stranger.

Stay alert to bias. Because AI systems learn patterns from huge amounts of existing text, they can reproduce unfair or one-sided patterns present in that data; if a response about a group, a place, or a topic feels one-sided, slow down and think critically, the same way you would with any other source. Never use AI to deceive or impersonate someone without their knowledge and consent.

Suppose an AI tool drafts a cover letter that includes a specific accomplishment, such as a number of projects completed. Before sending it, you would check that number against your own records, since AI can hallucinate details; decide whether disclosing your AI use is appropriate; and make sure the letter still sounds like you rather than generic AI output. That combination, verifying facts, being honest about your process, and keeping your own voice, is what ethical AI use looks like in practice.

AI in the workplace today

AI has moved from an experimental technology into an everyday part of many workplaces, sitting alongside email, spreadsheets, and word processors, without replacing human judgment. Workers commonly ask AI to draft first versions of emails, agendas, and reports, turning rough notes into an organized paragraph they then edit into a final version. AI also summarizes long documents quickly, sorts and prioritizes email, and analyzes spreadsheet data, spotting patterns or generating a plain-language summary of a set of numbers.

AI shows up in scheduling and meeting tools too, suggesting meeting times, generating meeting notes from a recorded conversation, and summarizing decisions made during a call. For a blind professional, an AI-generated meeting summary is especially useful, turning an hour of audio into a few paragraphs you can read quickly with your screen reader instead of re-listening to a long recording. None of this replaces a human worker's judgment: a good AI-assisted summary still needs a human check for accuracy, and a good AI-drafted email still needs a human check for tone, before it is sent. This is the pilot-and-co-pilot mindset applied directly to office work. Two more workplace uses are worth knowing: data handling, such as cleaning up a messy spreadsheet, and automation, setting up a repeated task, such as sorting new emails into folders, so it happens on its own instead of requiring a person to redo the same steps every time.

AI in accessibility tools: image description and OCR

A good AI image description does more than a short, generic caption: it can describe who appears to be in a family photo and what they seem to be doing, help you find the right product size or flavor on a shelf, or translate a diagram in a work document into a spoken explanation. Because apps and menus change over time, the reliable approach is to use your screen reader's own exploration tools to find the description feature inside whatever app you are using, and to treat the description as a helpful draft rather than a guaranteed, perfectly accurate account. It can occasionally misread a scene, miscount people, or misidentify an object, so when details matter a great deal, double-check with another method or another person.

AI-powered OCR reads printed or handwritten text out of a photo or scan and turns it into text your screen reader can read aloud, useful for mail, printed handouts, menus, medication labels, and signs. Unlike older OCR technology, which often needed a flat, well-lit, perfectly aligned scan, AI-powered OCR is generally far more forgiving of a photo taken at an angle or in uneven lighting. It is not perfect; it can misread unusual fonts, very small print, or difficult handwriting. When exact wording truly matters, such as a medication dosage, treat the result the way you treat any AI output: as something to double-check, for example by asking a pharmacist to confirm a label.

Image description and OCR often live inside the same app, and sometimes respond to the same photo: a tool might first describe a scene, then read the actual printed text within it. Recognizing which capability you are using helps you judge how carefully to double-check the result, since a wrong guess about a scene is usually low stakes, while a misread word in a printed letter can matter a great deal.

AI in navigation, customer service, education, and government

AI has also changed how many blind and low vision people move through unfamiliar places. Traditional GPS apps announce turn-by-turn directions; AI-enhanced navigation tools go further, adding spoken descriptions of what is around you, such as nearby points of interest or a room's layout from a photo, pointing out doorways or obstacles. These tools should support your own orientation and mobility skills and your cane or guide dog, never replace them, since AI description can be wrong, delayed, or unclear about a genuinely important safety detail. Think of it as layered information: your cane or guide dog gives the most immediate information right around you; your training gives you a route strategy; GPS gives turn-by-turn directions; and AI-enhanced description adds a fourth layer of wider context. Each layer supports the others; none replaces the rest.

Many companies now use AI-driven chat and voice systems for customer service. An AI-driven text chat can often understand a typed question in plain language, such as "why was I charged twice," without forcing you through a numbered phone menu, and it works well with a screen reader since it is simply text you can read and respond to by typing. On the drawback side, AI voice systems can misunderstand accents or background noise, and some AI chat systems are built with poor accessibility, using custom-coded windows a screen reader cannot easily read. When an AI tool is not accessible or not working for you, the most reliable fallback is usually to ask for a human directly, such as by typing or saying "talk to a person."

Many schools now use AI-driven help systems to answer common questions, which can mean getting a clear answer at ten at night instead of waiting for office hours. Government agencies and benefits offices are also beginning to use AI to explain dense official letters and forms in plain language. This is exactly the kind of high-stakes situation where verifying medical, legal, and financial information matters most: an AI's plain-language summary of an official letter is a helpful starting point, never a replacement for reading the actual letter carefully before acting on it.

AI in writing, coding, and daily life

People use AI to draft personal letters, brainstorm ideas for a story, or rewrite a paragraph in a different tone. A particularly useful pattern is asking an AI writing tool for output in a specific, accessible format, such as plain paragraphs instead of a complex table. AI-assisted writing still requires you to protect your own voice and verify any facts included in the draft.

AI is also widely used to help write and explain computer code. Tools built into code editors can suggest the next line of code, explain what an unfamiliar piece of code does in plain language, or help track down the cause of an error message. For a blind or low vision programmer, an AI explanation in plain language can be especially valuable, since reading dense code symbol by symbol with a screen reader is often slower than reading a clear explanation of what a block of code is doing. Coding is one more field where AI has become a common assistant rather than a replacement for the human programmer's judgment, and a blind professional comfortable with both a screen reader and AI-assisted coding tools can compete for roles that used to feel out of reach.

Outside of work, AI shows up in voice assistants that understand spoken requests, streaming services that recommend shows or music, shopping apps that suggest products, and cooking apps that suggest a recipe from ingredients you already have. Many smart home systems, such as connected lights, thermostats, and door locks, use AI-driven voice assistants as their main control method, a genuine accessibility win since speaking a request removes the need to locate a small physical panel or an inaccessible screen. Across every example in this lesson, the same mindset applies: treat AI output as a helpful draft from a capable but imperfect co-pilot, and keep your own judgment as the pilot in charge of the final decision.

Why AI genuinely excels at drafting, summarizing, coding, and describing

The same design that lets an LLM generate text by predicting the most likely next token, based on patterns learned from an enormous amount of training text, is exactly why AI is so strong at these four tasks. Drafting rewards exactly the skill an LLM has practiced the most: producing plausible, well-formed text quickly, based on countless similar examples, which is why first drafts tend to be well-organized even when details need correcting; this is also why drafting is one of the fastest wins for a new AI user, replacing the intimidating blank page with the easier task of editing something that already exists. Summarizing rewards a related skill: identifying which pieces of a text are most central. AI tools process a long document far more quickly than a human can read it, which is why a fifty-page report can become a three-paragraph summary in seconds; the speed is a genuine strength, but the accuracy of that summary is something you must still verify.

Computer code follows fairly strict, learnable patterns, so an AI tool trained on huge amounts of code can often predict the next line, or explain an unfamiliar block of code, with real accuracy. Describing images draws on a different kind of pattern recognition, built from huge numbers of images paired with written descriptions during training, which is why AI image description can produce a detailed, mostly accurate description of a photo it has never seen before. Coding and image description are two areas where blind and low vision people have historically faced the steepest barriers, so AI's genuine strength here is a large part of why this lesson calls AI a powerful equalizer. Understanding that one design choice, recognizing patterns from huge amounts of training data and generating a plausible response quickly, produces all of these strengths also explains why that same design choice is the direct source of AI's most serious weaknesses, covered next.

The hallucination problem, revisited

An AI system does not have a built-in way to check whether a specific fact is true. It generates the next token based on which words are statistically likely to follow, given everything written so far. Most of the time, this produces accurate information, because accurate patterns are common in the training data. But when an AI system is asked about something obscure, something outside its training data, or something with very specific numbers or names, it can still generate confident, well-formed, completely wrong text, because the system is optimized to produce plausible-sounding language, not to only state verified facts.

Hallucinations can be small, such as a wrong date, or serious, such as a fabricated quotation attributed to a real person, a made-up statistic presented as if it were from a real study, or a citation to a research paper or article that does not exist at all. The confidence of the AI's tone gives no reliable clue about whether a specific claim is accurate; a hallucinated fact and a correct fact can sound exactly the same. Imagine asking an AI tool for the exact publication date of a specific news article, and receiving a confident, specific answer. That date might be exactly right, or the AI might have generated a plausible-sounding date that is completely invented, because a specific date is a small, precise detail that is easy for a language model to get wrong even while sounding certain. This is exactly why names, dates, and numbers deserve extra verification: they are precisely the kind of detail most vulnerable to hallucination, because there are many plausible-sounding wrong answers and only one right one.

Bias, privacy risk, and inaccessible output

Bias means an unfair or one-sided pattern in AI output. It happens because an AI system learns its patterns from existing text and data, and that material can itself contain unfair or one-sided patterns, from historical inequality or gaps in what topics were well documented online. No single conversation can fully fix it; the practical response is awareness: when a response about a group, a place, or a contested topic feels one-sided, slow down and think critically. Bias can also affect blind and low vision users specifically: an AI tool's default assumptions about disability can sometimes lean on outdated or narrow ideas, such as assuming blindness always means total vision loss. Noticing this and gently correcting the AI tool with more accurate context in your prompt is a practical skill worth building.

Using AI tools involves real privacy considerations. Most AI chatbots process what you type on remote servers operated by the company that runs the tool, so your conversation text, and anything you upload, typically leaves your device. This creates two risks: sensitive information becomes data held by another company, subject to its own privacy policy; and some tools use conversations to improve their systems unless you turn that setting off. Avoid pasting truly sensitive personal information into an AI tool unless you understand its privacy practices. A useful rule of thumb: before pasting anything, ask whether you would be comfortable if a stranger at that company read it. If not, remove the sensitive details first or use a different method.

AI can also generate output that is technically correct but inaccessible to a screen reader user, a weakness this course takes especially seriously. An AI writing tool might generate a complex table when a simple list would work better, or webpage code missing proper headings, labels, or alt text, because most AI systems were trained mainly on how sighted people format content; an AI does not automatically know a table needs scoped headers unless specifically asked. The practical response is asking for accessible output directly, such as plain paragraphs instead of a table. An AI tool can also misjudge its own accessibility, confidently claiming a document is accessible without truly being able to inspect the file the way your screen reader can; this is another form of hallucination, and it is exactly why you should check the structure yourself rather than only asking the AI to confirm its own work.

Over-reliance on AI

The final weakness this lesson covers is over-reliance: leaning on AI so heavily that your own skills, judgment, or independent verification habits weaken over time. The pilot-and-co-pilot mindset guards directly against this. Over-reliance looks like accepting an AI's first answer without question, letting AI handle every writing task without ever practicing your own drafting skills, or trusting an AI-generated summary of an important document without ever checking the original. Over-reliance is a real risk precisely because AI is genuinely useful and often correct, which can make it easy to stop questioning it. The antidote is not avoiding AI. It is intentional practice: keeping your own skills active, spot-checking AI output regularly rather than only when something feels obviously wrong, and treating every one of AI's real strengths as reasons to use AI thoughtfully rather than reasons to stop thinking for yourself. Think of it as cross-training rather than replacement: the goal is a broader set of tools you can reach for, not a single tool you depend on completely.

Verifying AI output independently with your screen reader

This section turns everything above into concrete techniques you can use every time you work with AI. Each works entirely with a screen reader and does not require sight, and each works with any major screen reader, such as JAWS, VoiceOver, or TalkBack; check your own screen reader's documentation or shortcut list for exact key combinations.

First, ask the AI tool to cite its source or explain its reasoning; "how do you know that" often reveals whether a claim is well-supported. Second, cross-check specific facts, especially names, dates, numbers, and anything medical, legal, or financial, against a separate, reliable source before relying on them. Third, read AI output slowly using Say All, or your screen reader's continuous reading command, rather than skimming, so you notice a claim that sounds slightly off. Fourth, ask the AI tool to check its own work; a follow-up like "are you sure that is correct" sometimes catches an error, though this is not foolproof and does not replace independent verification. Fifth, check accessibility structure directly, using your screen reader's heading navigation, landmark navigation, and its list of links, tables, or interactive elements, to confirm real headings, labeled fields, and table headers, rather than trusting the AI's own claim. Sixth, get a second opinion from a different AI tool; comparing answers to the same question can reveal a hallucination, since two tools are less likely to make the same specific error.

You do not need all six techniques every time. A quick, low-stakes question might only need careful reading. A high-stakes task, such as a medical, legal, or financial matter, deserves several techniques together. Judging how much verification a task deserves is itself part of the AI mindset this lesson is building, and it gets easier with practice, the same way fear fades through small, low-stakes experience.

Building your AI mindset going forward

Put together, this lesson gives you a balanced picture: real strengths worth using confidently, real weaknesses worth watching for deliberately, and concrete techniques that let you keep the pilot-and-co-pilot mindset in practice, not just in theory. Finally, know that AI tools and best practices change quickly; a specific button, menu, or keystroke described in this course could shift after an app update. When that happens, lean on your screen reader's own exploration tools, such as heading or landmark navigation, rather than memorizing something that might already be out of date. That skill, exploring confidently instead of memorizing rigidly, is itself part of the AI mindset this week is building.

Key terms from this week

Use this list to review the vocabulary introduced in this lesson before you start the exercises and the test.

Artificial intelligence (AI)

A general term for computer systems that can perform tasks that normally need human intelligence.

Large language model (LLM)

A kind of AI, trained on huge amounts of text, that generates responses by predicting the most likely next piece of text over and over.

Hallucination

When an AI system confidently states false information as if it were true.

Prompt

The text you type or say to instruct an AI system.

Context window

The amount of text an AI system can consider at one time during a conversation.

Multimodal AI

An AI system that can work with images or audio, in addition to text.

Bias

An unfair or one-sided pattern in AI output, often from unfair or one-sided patterns in the data the AI was trained on.

Privacy risk

A danger to your personal information created by how an AI tool collects, stores, or uses the text and files you share with it.

Over-reliance

Depending on AI so heavily that your own skills, judgment, or independent verification habits weaken over time.

Pilot and co-pilot mindset

A way of thinking about AI use in which you, the human, remain the pilot who makes the final decisions, while AI acts as a helpful but not fully trusted co-pilot.

Keyboard-only exercises

These exercises use only your keyboard. They are written to work with any screen reader. Where a specific command is mentioned, check your own screen reader's shortcut list or documentation for the exact keys, since JAWS, VoiceOver, and TalkBack, among others, each use different key combinations for the same feature.

Exercise 1: Explore this page's structure

Before working with any AI tool, practice exploring the structure of this very lesson page. This builds a habit you will use in every AI tool's interface throughout the course.

1. Open this Week 1 file in your web browser, if it is not already open.
2. Open your screen reader's list of headings, or use its heading navigation command, to review every heading on this page, starting with the Heading 1, "Week 1: Building an AI Mindset."
3. Close that list, or activate a heading to jump straight to it.
4. Press Tab one time from the very top of the page, before doing anything else. Confirm that your screen reader announces a link called "Skip to main content." This link should always be the very first stop when you press Tab on any file in this course.
5. Use your screen reader's heading-by-heading navigation command to move from heading to heading down the page, without opening the full list. Notice how this lets you skim the whole lesson quickly.

Exercise 2: Your first AI conversation

In this exercise, you will send one simple message to an AI chatbot and read its reply using only your keyboard. You do not need an account for this exercise.

1. Open your web browser.
2. Move focus to the browser's address bar. In most Windows browsers, press Alt plus D, or Ctrl plus L. On Mac, press Command plus L.
3. Type the web address for an AI chatbot your instructor or this course has approved for practice, and press Enter. Wait until your screen reader announces that the page has finished loading.
4. Find the message entry box. Use your screen reader's browse-mode navigation for edit fields or form controls, or press Tab repeatedly from the top of the page until you hear an edit field described with words like "message" or "how can I help."
5. With the message box focused, type this exact question: "In plain language, what is artificial intelligence?"
6. Press Enter to send your message.
7. Wait a few seconds for the response to begin appearing. Then use Say All, or your screen reader's continuous reading command, to have your screen reader read the new response aloud from top to bottom.
8. Compare the response you received to what this lesson taught about what AI is. Notice any similarities or differences in wording.

Exercise 3: Practice fact-checking AI output

This exercise practices the verification habit from this week's lesson, using a short passage provided here so that every learner works with the same text.

1. Open a plain text editor. On Windows, press Windows key plus R to open the Run dialog, type notepad, and press Enter.
2. Type or paste the practice passage above into the editor.
3. Read the passage back using Say All, or your screen reader's continuous reading command.
4. Identify every claim in the passage that sounds like it needs checking. There is more than one error.
5. Using your browser, search for reliable information to check each claim you flagged.
6. Below the original passage, type a corrected version of the passage in your own words.
7. Save the file. Press Ctrl plus S, type the file name week1-factcheck, and press Enter.

Exercise 4: Check an AI-generated table for accessibility

1. Ask an AI chatbot to generate a short document that includes a data table, for example, "Create a short table comparing three fruits by color, taste, and price."
2. If the AI tool can produce a webpage or downloadable file, open the result. If it can only produce a chat response, treat the table it generated in the chat as your object to check.
3. Open your screen reader's list of tables, or use its table navigation commands, to confirm whether the table has proper column and row headers.
4. Move through the table cell by cell and confirm whether your screen reader announces the correct header for each cell.
5. If the table is missing proper headers, ask the AI tool to redo it as a simple accessible list instead, and compare how much easier the list is to read with your screen reader.

Portfolio project: My AI Foundations Portfolio

This week's portfolio piece has three short parts: a personal statement about where you are starting from, a written inventory connecting this week's lesson to your own life, and a reusable checklist you can use every time you rely on AI output going forward. Answer honestly rather than trying to sound impressive. Keep this file in a dedicated portfolio folder, since you will draw on work from across the whole course later on.

1. Open a word processor or plain text editor.
2. Write a heading at the top: "My AI Foundations Portfolio."
3. Under a sub-heading called "My AI Mindset Statement," answer the following three prompts in your own words, using a few sentences for each:
   1. What is one thing you hope AI will help you accomplish, either for employment or for personal independence?
   2. What is one fear or concern you have about using AI, and what is one specific step you will take to manage it, based on this week's lesson?
   3. What is one ethical rule from this week's lesson that you personally commit to following every time you use AI?
4. Under a sub-heading called "AI in My Life and Work," list three to five specific places from this week's lesson, such as the workplace, image description, OCR, navigation, customer service, writing, coding, or daily life, where AI could realistically help you personally. For each one, write one or two sentences describing exactly how you would use it, using an example from your own life rather than a generic description.
5. Under a sub-heading called "My AI Verification Checklist," list at least five specific steps you personally commit to taking before relying on AI output for anything important, drawing on the verification techniques from this week's lesson, written in your own words. For at least two of the steps, add one sentence explaining when you would use that step, for example, always, or only for medical and financial information.
6. Save the file with the name my-ai-foundations-portfolio, in a format you can open again later, such as .docx or .txt, in a dedicated portfolio folder.

Weekly test

Answer key

Each answer below includes a one-sentence explanation drawn from this week's lesson.

Looking ahead

The next part of this course moves from foundational understanding into hands-on practice with specific AI tools. Keep the pilot-and-co-pilot mindset, the verification habits, and the accessible-output requests from this lesson close at hand, since every tool and technique ahead builds on the foundation you built this week.