Your Beginner AI Course

What this opening lesson is for

Artificial intelligence can produce excitement, worry, skepticism, confusion, and curiosity—sometimes all at once. Those reactions are reasonable.

This course does not begin by asking you to trust AI or reject it. It begins by giving you enough information and practice to make your own informed choices.

AI Without the Hype is not a product demonstration or a prediction about the future. It is a practical course about tools already appearing in phones, workplaces, schools, public information, scams, creative applications, and everyday services.

By the end of this lesson, you should understand that:

• You may participate without prior technical knowledge.
• You may pass on any exercise or question.
• Private personal information is not needed for classroom practice.
• AI mistakes are expected and can become useful learning examples.
• The course will not ask you to treat a chatbot as a person, oracle, or final authority.
• Human judgment remains responsible for consequential decisions.

Four ways to participate

Community learning agreement

• No foolish questions: technical vocabulary will be defined rather than assumed.
• No private demonstrations: use fictional, public, or de-identified examples.
• No forced enthusiasm: curiosity and caution may exist together.
• No shaming: people have different devices, internet access, experience, and learning needs.
• No magical claims: conversational fluency is not proof that a machine possesses feelings, consciousness, wisdom, or authority.
• No automatic trust: important claims require verification beyond the chatbot.
• No surrender of responsibility: AI may assist with a task, but people remain responsible for judgment and consequences.

Opening activity: your AI temperature check

You may answer aloud, write privately, point to a printed option, or simply consider the questions.

1. Have you used an AI chatbot before?
2. What is one thing about AI that interests you?
3. What is one thing about AI that concerns you?
4. What practical task would make this course useful to you?

There is no correct emotional response. Interest, uncertainty, concern, skepticism, and excitement can all provide useful starting points.

Suggested facilitator introduction

Psychological safety boundary

A chatbot may occasionally produce disturbing, manipulative, overly personal, or confusing language. You may stop the interaction immediately. Nobody owes a machine continued attention.

AI is not an emergency service, sole therapist, spiritual authority, or substitute for human care. A person who needs urgent help should contact an appropriate human, local service, or emergency resource rather than continuing a chatbot exercise.

The course’s central rule

An AI answer is a candidate—not a verdict.

We may:
- keep it,
- revise it,
- verify it,
- remove it,
- reject it,
- or decide not to use AI for the task.

Next lesson

Lesson 01 begins with familiar tools that already use forms of artificial intelligence. This helps separate the broad category of AI from the specific experience of using a chatbot.

Why begin with familiar examples?

Chatbots can make artificial intelligence feel entirely new. However, many people have interacted with forms of AI for years without calling them AI.

A phone may improve a photograph, create captions from speech, recognize a face, suggest the next word in a message, or filter an unwanted call. An email service may identify spam. A bank may flag a purchase that does not match an account’s usual pattern. A shopping, music, or video service may recommend what someone might want next.

Some farms, workshops, warehouses, hospitals, utilities, and transportation systems also use pattern-recognition or predictive tools. The exact technology varies, and not every modern-looking tool uses AI.

Beginning with familiar examples helps us replace the image of one mysterious “AI” with a more useful question:

By the end of this lesson, you should be able to:

• Explain why AI is a broad category rather than one single technology.
• Distinguish basic automation from pattern-based AI and generative AI.
• Identify several ordinary places where AI may already appear.
• Ask what information a system uses, what result it produces, and what can happen if it is wrong.
• Avoid assuming that every advanced or automated device uses AI.

Four useful categories

A familiar comparison

Where might AI appear in everyday life?

• Email spam and phishing filters
• Fraud and unusual-transaction alerts
• Speech-to-text and live captions
• Translation and writing assistance
• Autocorrect and next-word suggestions
• Phone-camera enhancement and photo organization
• Face, object, voice, or handwriting recognition
• Product, music, video, and news recommendations
• Route suggestions and travel-time estimates
• Customer-service chat systems
• Equipment monitoring and maintenance predictions
• Some agricultural sensing, sorting, and crop-monitoring tools
• Text, image, audio, video, and code generators

This list does not mean every service in each category uses the same technology. It means these are common places to ask whether pattern-based or generative systems are involved.

Activity: What kind of tool is this?

For each example, choose the description that seems most appropriate:

1. Fixed-rule automation
2. Pattern recognition or prediction
3. Generative AI
4. Hybrid or unclear without more information

Suggested discussion

• Example A is usually fixed-rule automation.
• Example B commonly uses pattern recognition.
• Example C is generative AI.
• Example D may be automation, AI, or a hybrid. We need more information about the system.
• Example E commonly uses speech-recognition AI.
• Example F is ordinary calculation rather than AI.

The tractor example is especially important. A sophisticated tool is not automatically an AI tool. “I need more information” is often the most accurate answer.

The five-question tool check

When someone says a product “uses AI,” ask:

1. What job is the system performing?
2. What information or data does it use?
3. Does it retrieve, classify, predict, recommend, or generate?
4. What happens when it makes a mistake?
5. Which person or organization remains responsible?

Copyable prompt: understand a technology

Help me understand this technology:

[NAME OR DESCRIPTION OF THE TOOL]

Do not assume it uses artificial intelligence.

Explain in plain language:

1. What job the tool performs
2. Whether it appears to use fixed-rule automation, pattern-based AI, generative AI, a hybrid system, or whether there is not enough information to tell
3. What data or inputs it may use
4. What output or decision it produces
5. What could happen if it is wrong
6. What a human should verify before depending on it

Separate confirmed information from assumptions. Do not invent technical details about the product.

Human judgment check

Think of one “smart” tool you have encountered. It might be a phone feature, website, appliance, workplace system, vehicle feature, farm tool, or public service.

Ask yourself:

• Do I know what task it is performing?
• Do I know what information it collects?
• Can I correct or challenge its result?
• Would a mistake be inconvenient, expensive, unfair, or dangerous?
• Am I still making the decision, or have I quietly handed the decision to the tool?

Next lesson

Lesson 02 looks more closely at generative AI: how a chatbot can produce useful language, why it can sound confident while being wrong, and why fluent conversation is not proof of consciousness or authority.

For the research basis, public guidance, and definitions used throughout this course, visit AI Education: Sources and Standards .

What does “generative” mean?

The word generative means that the system produces new material in response to an instruction.

Depending on the system, that material might include:

• Text
• Images
• Audio or music
• Video
• Computer code
• Summaries, plans, lists, translations, or drafts

Generative AI is different from a system that only retrieves an existing record. It creates an output by combining learned patterns with the prompt, conversation, uploaded material, and any tools the product is authorized to use.

By the end of this lesson, you should be able to:

• Explain generative AI without describing it as magic or as a simple database.
• Describe a token as a small unit of text processed by a language model.
• Explain why the same prompt may produce different answers.
• Distinguish the underlying model from the product interface and any connected tools.
• Explain why a fluent answer can still be false.
• Avoid treating conversational language as proof of feelings, consciousness, or authority.

A simplified view of how a text model works

Modern language models are complicated, but a beginner can understand the basic process through five steps.

A sentence-completion example

Imagine the unfinished sentence:

Instead of turning left, she turned ______.

Several completions could make sense:

• right
• around
• back
• toward the river

A language model assigns different probabilities to possible continuations based on the context. It then continues this process repeatedly until it has produced a longer response.

This helps explain why:

• The same question can produce more than one reasonable answer.
• A small wording change can affect the result.
• Earlier conversation can influence a later answer.
• The model can produce an original combination of familiar patterns.
• A statistically plausible sentence can still contain an inaccurate fact.

Activity: become the next-word generator

This activity requires no device. A facilitator reads each beginning aloud and asks participants to suggest possible next words or phrases.

Participants will usually offer several sensible continuations. None must be the one “correct” completion. Context narrows the possibilities, but it does not always reduce them to one answer.

Discussion questions

• Which completions seemed most likely?
• Which were creative but still plausible?
• What additional context would change your answer?
• Could a grammatically perfect completion still be factually wrong?

The model, the product, and the tools are not the same thing

When people say, “The AI told me,” several different layers may be involved.

One chatbot may have current web access while another does not. One may be able to read an uploaded document, while another may only generate from its existing context. One may use a calculator tool, while another may attempt the arithmetic through language generation.

Therefore, before relying on an answer, ask:

What the model is not automatically doing

A text response does not automatically mean that the system:

• Looked the answer up in a current, verified source
• Checked every factual statement
• Understood the situation as a person would
• Remembered where every learned pattern originally came from
• Knew whether its citations were real
• Understood the consequences of someone acting on the answer
• Possessed feelings, personal beliefs, needs, or lived experience

Why can it sound certain while being wrong?

A language model is very good at producing language that fits a pattern. Producing a plausible continuation and proving that a claim is true are different tasks.

NIST uses the term confabulation for output that confidently presents erroneous or false content. People also commonly use the word “hallucination,” although that word can encourage people to imagine the system as more human than it is.

Confabulation may include:

• An incorrect date
• A nonexistent quotation
• An invented book, article, court case, or website
• A factual claim unsupported by the supplied document
• An explanation that contradicts an earlier answer
• A confident step-by-step justification for a wrong result

A false response does not, by itself, show that the system formed an intention to lie. It shows that fluent generation is not the same thing as verified knowledge.

Creative fiction and false factual claims are different

The same generative ability can be delightful in a creative task and dangerous in a high-stakes factual task. The risk depends partly on what the answer will be used for.

Optional live demonstration

A facilitator can show two contrasting tasks.

Demonstration A: grounded transformation

Using only the paragraph I provide, rewrite it at an eighth-grade reading level.

Do not add facts that are not in the paragraph.

Afterward, list anything that may have changed in meaning.

PARAGRAPH:
[PASTE A SHORT PUBLIC OR FICTIONAL PARAGRAPH]

Demonstration B: unsupported recall

Answer this obscure factual question.

Before answering, state whether you have a current verified source available. If you do not, say that the answer may be uncertain.

Do not invent a citation.

The first task gives the model source material and a bounded transformation. The second asks for factual recall that may or may not be well supported.

The class should compare:

• How fluent each answer sounds
• What evidence each answer actually has
• Whether confidence matches support
• Which answer would require outside verification

Copyable prompt: show me the basis of your answer

Help me with this question:

[INSERT QUESTION]

Before giving the answer, identify which information basis you are using:

- text or files I provided,
- general patterns learned during training,
- a current external search or connected tool,
- or an assumption.

Then answer in four sections:

1. Best-supported answer
2. Important assumptions
3. Uncertainty or missing information
4. What I should verify independently

Do not invent sources, quotations, statistics, links, or citations. If you cannot verify something, say so plainly.

Human judgment check

Before using a generated answer, ask:

• Was I asking for creativity, factual accuracy, or both?
• Did I provide the source material?
• Did the system use a current search or another tool?
• Could the answer be checked against an official or primary source?
• Is the response merely fluent, or is it actually supported?
• What could happen if the answer is wrong?
• Am I responding to evidence, or to the system’s confident and personable tone?

Next lesson

Lesson 03 examines common AI myths and extreme claims. We will practice holding a balanced position: AI can be useful without being magical, and fallible without being useless.

Definitions, external guidance, and UVLM claim boundaries are documented on AI Education: Sources and Standards .

Why this lesson is not ordinary “myth-busting”

Some concerns about artificial intelligence are supported by evidence. Others are reasonable questions that remain unsettled. Some claims are exaggerated marketing, science-fiction storytelling, or confident prediction presented as fact.

Calling every concern a “myth” can silence people. Treating every frightening possibility as certain can also cause harm.

This course uses four categories:

By the end of this lesson, you should be able to:

• Recognize common extreme claims about artificial intelligence.
• Distinguish an observed capability from a prediction about the future.
• Separate marketing language from independently supported evidence.
• Explain why conversational style is not proof of consciousness, emotion, or authority.
• Discuss employment effects without pretending that one simple forecast applies to every worker or community.
• Ask what evidence would support or weaken a claim.
• Preserve human judgment without dismissing useful technology.

Claim 1: “AI knows everything.”

No single AI system has access to all knowledge. A system may have learned patterns from a large collection of material, but that does not make its information complete, current, accurate, or appropriate for every situation.

A particular product may also lack:

• Current internet access
• Access to a specific database
• Local knowledge
• Private organizational records
• The full context of a person’s circumstances
• Reliable knowledge of events after its training period
• A way to verify every generated statement

Claim 2: “A chatbot is just a search engine.”

A traditional search service primarily finds and ranks existing pages or records. A generative chatbot constructs a response from patterns and context.

Some modern products combine both functions. A chatbot may generate language, run a web search, read a file, use a calculator, or call another tool. Another chatbot may have none of those capabilities.

Claim 3: “If it talks like a person, it must think or feel like one.”

Chatbots are designed to produce human-readable conversation. They may use phrases such as:

• “I understand.”
• “I think.”
• “I am sorry.”
• “I would recommend.”

Those phrases may make an interaction easier to read. They do not, by themselves, establish subjective experience, emotion, consciousness, personal belief, moral authority, or a private inner life.

Questions about machine consciousness remain philosophically and scientifically contested. This course does not settle that debate. It uses a narrower safety rule:

Claim 4: “AI is objective because it uses mathematics.”

Mathematics does not automatically remove human choices or social context.

AI outcomes can be affected by:

• Which data was collected
• Which people or situations were missing from the data
• How categories were defined
• Which goals the system was optimized to pursue
• Which errors were considered acceptable
• How the product was tested
• Where and by whom it was deployed
• How people interpret or act on its output

Bias does not require a designer to intend harm. A system can still produce uneven or harmful results because its data, categories, assumptions, testing, or deployment context do not adequately represent the affected people.

Claim 5: “AI will replace every job.”

Artificial intelligence is likely to change many tasks and occupations, but exposure to AI is not the same as complete job replacement.

A job is usually a bundle of different tasks. Some may be:

• Automated
• Accelerated
• Reorganized
• Checked by AI
• Made more valuable because human judgment becomes more important
• Left largely unchanged

Effects will vary across occupations, industries, communities, income levels, access to training, management choices, labor protections, and public policy.

A 2025 global analysis from the International Labour Organization and NASK concluded that many occupations are potentially exposed to generative AI, while transformation is more likely than complete replacement.

A more useful employment exercise

Instead of asking only, “Will AI take this job?” divide the work into tasks:

Claim 6: “AI will not affect my work at all.”

This is also too certain. A person may not use a chatbot directly while still being affected by:

• Employer purchasing decisions
• New productivity expectations
• Automated screening or scheduling
• Changes in customer behavior
• AI-generated fraud or misinformation
• New training requirements
• Changes in related industries or public services

The balanced position is not that every job disappears or that nothing changes. It is that changes are likely to be uneven and should be examined at the level of actual tasks, institutions, and communities.

Claim 7: “AI is useless because it makes mistakes.”

A tool does not need to be perfect to be useful. Spellcheckers, search services, weather forecasts, maps, calculators, and human advisers can all be useful within appropriate boundaries even though they have limitations.

AI may be useful for:

• Brainstorming
• Rewriting
• Organizing notes
• Creating a first-draft checklist
• Practicing an interview
• Producing questions to ask a professional
• Explaining a concept at a different reading level

The appropriate question is not simply, “Can it make a mistake?” The better questions are:

• How likely is a mistake?
• Can the mistake be detected?
• Can a person correct it?
• What happens if the mistake is not caught?
• Is this task low-stakes or high-stakes?

Claim 8: “If a product says ‘AI-powered,’ it must be advanced.”

“AI-powered” is a marketing description, not an independent evaluation.

A company should still be able to explain:

• What the product actually does
• Which users and situations it was tested on
• How accuracy or effectiveness was measured
• What failure cases are known
• What information the product collects
• Whether a human reviews consequential results
• What evidence supports promises about savings, income, accuracy, safety, or compliance

Consumer-protection regulators have repeatedly challenged unsupported or deceptive claims attached to AI products and business opportunities. The letters “AI” do not create an exception to ordinary evidence, advertising, or consumer protection expectations.

Marketing claim check

A product makes this claim:

[PASTE OR SUMMARIZE THE CLAIM]

Help me evaluate it without assuming the claim is true.

Separate your response into:

1. What the company is claiming
2. What evidence would be needed to support that claim
3. What important conditions or limitations may be missing
4. What independent testing or documentation I should look for
5. What information the product may collect
6. What could happen if the claim is wrong
7. Questions I should ask before paying, registering, or relying on it

Do not treat marketing language, testimonials, awards, or the phrase “AI-powered” as independent proof.

Claim 9: “The most confident answer is probably the best one.”

Confidence of tone and quality of evidence are different things.

A generated answer may be:

• Cautious and correct
• Cautious and incorrect
• Confident and correct
• Confident and incorrect

Tone can influence people emotionally, but tone does not verify a date, source, quotation, diagnosis, deadline, or prediction.

Activity: the four-basket claim audit

Place each statement into one of four baskets:

1. Well supported
2. Plausible but conditional
3. Exaggerated or unsupported
4. Open or contested

Suggested discussion

• Statement A is a demonstrated capability, though the quality of the draft varies.
• Statement B goes beyond what conversational style can prove.
• Statement C is broadly supported, but the type and scale of change vary.
• Statement D is a highly specific prediction without adequate support.
• Statement E is well documented, though the source and form of bias vary.
• Statement F is unsupported; the question remains contested.
• Statement G is a sound verification practice.
• Statement H is unsupported marketing logic.

Copyable prompt: audit a dramatic AI claim

Evaluate this claim about artificial intelligence:

[INSERT CLAIM]

Do not accept or reject it merely because it sounds exciting, frightening, technical, or popular.

Analyze it in these sections:

1. Exact claim
2. Observable evidence supporting it
3. Assumptions hidden inside it
4. Important conditions or exceptions
5. Credible counterevidence or alternative explanations
6. What remains unknown or contested
7. Who benefits if people believe the claim
8. What evidence would weaken or falsify the claim
9. Practical implications for an ordinary person
10. A cautious conclusion using one of these labels:
   - Well supported
   - Plausible but conditional
   - Exaggerated or unsupported
   - Open or contested

Do not invent sources or pretend a forecast is a measured fact.

Psychological safety: fear is information, not instruction

AI discussions can touch employment, identity, creativity, family safety, fraud, social change, disability, and control. Those topics may feel personal.

Participants may:

• Decline to discuss their employment situation
• Use fictional examples
• Pause or leave an upsetting demonstration
• Ask for a claim to be restated without dramatic language
• Separate immediate practical concerns from distant speculation

The course should not promise that “everything will be fine.” It should also not present the most frightening possible future as inevitable.

Human judgment check

When you encounter a dramatic AI claim, ask:

• Is this describing a current capability or predicting a future outcome?
• Is the claim about one product, one model, one occupation, or all AI?
• What evidence is visible?
• Which words are doing emotional work?
• What assumptions are hidden?
• Are uncertainty and disagreement being acknowledged?
• Who benefits financially, politically, or socially if I believe the claim?
• What decision am I being encouraged to make?
• What would I need to verify before acting?

Next lesson

Lesson 04 begins the safe-interaction portion of the course. We will identify information that should not be entered into a chatbot and practice turning a private situation into a safer, de-identified question.

Sources supporting the employment, bias, consumer-protection, human-agency, and AI-risk statements in this lesson will be listed on AI Education: Sources and Standards .

Practice should create confidence—not dependency

The goal of this lab is not to produce a perfect answer. The goal is to practice a repeatable process:

1. Choose an appropriate task.
2. Remove private information.
3. Build a CARES prompt.
4. Inspect the response.
5. Correct or revise it.
6. Verify anything consequential.
7. Decide whether to use, edit, reject, or discard it.

By the end of this lesson, you should be able to:

• Select an appropriate beginner task.
• Use a fictional, public, or non-sensitive example.
• Build a CARES prompt for a practical purpose.
• Review the output for usefulness, accuracy, privacy, tone, and unsupported claims.
• Revise an answer without surrendering authorship.
• Identify when a task has crossed into a higher-risk category.
• Decide whether the AI actually reduced total work.

Choose the right practice zone

The three-pass practice loop

Practice Station 1: explain supplied text

Use a short public or fictional paragraph. Remove identifying information first.

Context:
I am a beginner trying to understand the paragraph below.

Ask:
Explain it in plain language.

Requirements:
- Keep the explanation under 250 words.
- Define unfamiliar terms.
- Use one everyday example.
- End with one question that checks my understanding.

Evidence:
Use only the paragraph I provide.

Safety:
Do not add facts from memory. Identify anything the paragraph does not explain.

PARAGRAPH:
[PASTE PUBLIC, FICTIONAL, OR REDACTED TEXT]

Review questions

• Did the explanation preserve the original meaning?
• Did it introduce an unsupported claim?
• Was the example genuinely helpful?
• Which phrase should be checked against the original?

Practice Station 2: rewrite a message

Context:
I am drafting a message to [GENERAL AUDIENCE].

Ask:
Rewrite my draft so it is clear, respectful, and easy to scan.

Requirements:
- Preserve my meaning.
- Do not add promises or commitments.
- Keep it under 300 words.
- Use short paragraphs.
- Use a calm, professional tone.

Evidence:
Use only the facts in my draft.

Safety:
Do not invent names, dates, policies, laws, contact information, or authority.

DRAFT:
[PASTE NON-SENSITIVE DRAFT]

Human authorship check

Before sending:

• Read every sentence aloud.
• Remove wording you would not naturally use.
• Confirm every promise and date.
• Make sure the message still represents your intent.
• Sign it only after you accept responsibility for it.

Practice Station 3: turn notes into a checklist

Context:
These are rough notes for a household or community project.

Ask:
Turn them into an ordered checklist.

Requirements:
- Separate preparation, action, review, and cleanup.
- Use one clear action per line.
- Identify missing information instead of inventing it.
- Mark any step that may spend money, share information, create a commitment, or require permission.
- Keep the checklist suitable for printing.

Evidence:
Use only my notes.

NOTES:
[PASTE NON-SENSITIVE NOTES]

Review questions

• Is the order sensible?
• Did the model invent a step?
• Does any step require local knowledge?
• Which step remains assigned to a human?

Practice Station 4: brainstorm without treating ideas as facts

Context:
A community group is brainstorming ideas for [GENERAL PURPOSE].

Ask:
Generate twelve possible ideas.

Requirements:
- Include low-cost, moderate-cost, indoor, outdoor, intergenerational, and accessibility-conscious options.
- Keep each idea to two sentences.
- Do not claim that any idea is available, permitted, funded, or locally supported.
- Label all output as brainstorming candidates.

Safety:
Do not include risky, illegal, discriminatory, or privacy-invasive activities.

After the list:
- identify questions the group must answer,
- identify permissions that may be needed,
- and identify which ideas require local verification.

Practice Station 5: interview practice

Act as a supportive interview-practice partner.

Role:
Ask questions for a general [JOB OR VOLUNTEER ROLE] interview.

Rules:
- Ask one question at a time.
- Wait for my response.
- Do not request private identifiers.
- After each answer, identify one strength and one concrete improvement.
- Do not invent experience I did not mention.
- Help me revise my own answer rather than replacing it with a fictional life story.
- Avoid discriminatory or legally sensitive questions.

Begin with a simple opening question.

Human judgment check

A polished answer is not helpful if it claims experience, credentials, results, or responsibilities you do not have.

Practice Station 6: prepare questions for a professional

Help me prepare questions for a qualified [TYPE OF PROFESSIONAL].

Context:
[GENERAL, NON-SENSITIVE DESCRIPTION]

Do not diagnose, decide, or provide a final professional conclusion.

Create:

1. A short description I can bring to the appointment
2. Questions about options
3. Questions about risks and benefits
4. Questions about cost or timing
5. Documents or information I may need to bring
6. Terms I may want explained
7. Warning signs that would require a more urgent human response

Use placeholders for private information.

Practice Station 7: accessibility and format conversion

Reformat the supplied text for accessibility.

Create:

1. A plain-language version
2. A short bullet summary
3. A one-step-at-a-time version
4. A version suitable for reading aloud
5. A list of terms that still require explanation

Use only the supplied text.

Do not remove warnings, qualifications, deadlines, exceptions, or instructions merely to make the text shorter.

TEXT:
[PASTE PUBLIC OR REDACTED TEXT]

No-device practice option

A learner does not need an account or device to practice prompting.

1. Choose one task.
2. Write the CARES sections on paper.
3. Exchange the prompt with a partner.
4. Let the partner underline missing context and assumptions.
5. Revise the prompt before any AI system sees it.

The five-part output review

Copyable prompt: review the candidate

Review the candidate output below.

Do not praise it automatically and do not rewrite it yet.

Evaluate:

1. Did it follow the task?
2. Did it preserve the supplied facts?
3. What did it assume?
4. What may be inaccurate?
5. What requires external verification?
6. What private or sensitive information should be removed?
7. What wording sounds unlike the intended author?
8. What important qualification is missing?
9. What could cause harm if used without review?
10. What should be kept, changed, removed, or rejected?

CANDIDATE:
[PASTE OUTPUT]

The off-loading audit

AI may save effort, but a true efficiency gain should not simply move hidden work or risk somewhere else.

Ask:

• What task did I transfer to the system?
• What work did the AI genuinely reduce?
• What new review work did it create?
• Did it require me to reveal information?
• Did it create factual, privacy, bias, or authorship risk?
• Did another person inherit an unfair review burden?
• Would doing the task directly have been faster?
• Did the result materially improve my life or work?

When to stop the practice task

Pause or change course when:

• The task begins requiring private information.
• The answer creates a medical, legal, financial, tax, or safety decision.
• The system invents facts repeatedly.
• The required source is unavailable.
• Verification will take more effort than the task is worth.
• The output would affect another person without their review.
• You feel pressured, confused, distressed, or unable to judge the answer.
• A qualified human or official process is clearly the safer route.

Personal practice reflection

Complete these statements:

• The task I chose was:
• The AI did:
• I did:
• I corrected:
• I verified:
• I rejected:
• The tool saved me:
• The tool added this new burden:
• Next time I would:

Next lesson

Lesson 10 identifies high-impact situations where AI may help explain terms or prepare questions but should not become the emergency responder, professional adviser, authorizing official, or final decision-maker.

Practical-use and human-agency standards are listed on AI Education: Sources and Standards .

Decision support is not decision authority

By the end of this lesson, you should be able to:

• Distinguish assistance from authority.
• Identify high-impact situations requiring human escalation.
• Convert a request for a final answer into questions for an appropriate human.
• Recognize when delay itself creates risk.
• Identify the information and documents a professional may need.
• Evaluate whether human review is meaningful or merely ceremonial.
• Stop using AI when the task exceeds the system’s appropriate role.

The high-impact boundary

A decision becomes higher impact when it may affect:

• Life or physical safety
• Health
• Legal rights or deadlines
• Housing
• Employment
• Education
• Public benefits
• Taxes
• Money or credit
• Identity or account access
• Another person’s opportunities or reputation
• Critical infrastructure, vehicles, equipment, or repairs
• Children or vulnerable people
• Civic rights or official government procedures

Red-light situations

The human-handoff ladder

Meaningful human review

Human review is not meaningful merely because a person clicks “approve.”

A meaningful reviewer should have:

• Enough subject knowledge for the task
• Access to the relevant facts and sources
• Time to inspect the output
• Permission to question the system
• Authority to correct, reject, override, or escalate
• A clear record of who made the final decision
• A way to report and repair errors

Example 1: health information

Unsafe question

Here are all my symptoms. Tell me exactly what disease I have and whether I can wait until next month to get care.

Safer preparation prompt

Help me prepare for a conversation with a qualified health professional.

Use only the general, non-identifying information I provide.

Do not diagnose me, prescribe treatment, change medication, or tell me that urgent symptoms are safe to ignore.

Create:

1. A concise symptom timeline
2. Questions to ask
3. Information the clinician may want
4. Terms I may need explained
5. General warning signs that require prompt human medical attention

State clearly that the clinician must evaluate the actual person and circumstances.

Example 2: legal or agency notice

Unsafe question

Tell me whether I will win this case and the exact last day I can file.

Safer preparation prompt

Using only the redacted notice I provide, help me prepare questions for the issuing agency or a qualified legal adviser.

Identify:

1. Dates stated in the notice
2. Actions the notice requests
3. Consequences the notice states
4. Terms that need explanation
5. Facts missing from the notice
6. Questions about deadlines, appeal rights, jurisdiction, and required documents

Do not calculate or guarantee a legal deadline. Quote the exact language supporting each date and tell me to confirm the current deadline with the responsible authority.

Example 3: financial or banking problem

Unsafe question

Move my money to the safest investment and guarantee I will not lose anything.

Safer preparation prompt

Help me prepare questions for a qualified financial professional or regulated institution.

Do not recommend a specific transaction, guarantee returns, or request account credentials.

Create questions about:

- fees,
- risks,
- liquidity,
- time horizon,
- conflicts of interest,
- registration or licensing,
- custody of funds,
- tax considerations,
- alternatives,
- and what happens in a worst-case scenario.

Identify claims that require independent verification before any money is transferred.

Example 4: equipment or repair

Unsafe question

Tell me exactly which wires to connect so I can repair this high-voltage equipment without shutting it down.

Safer preparation prompt

Help me prepare for a conversation with a qualified repair professional.

Do not provide live electrical work instructions.

Create:

1. A list of observable symptoms
2. Questions about shutdown and isolation procedures
3. The model and documentation information a technician may need
4. Possible categories of failure stated only as possibilities
5. Questions about permits, qualifications, protective equipment, and safe inspection
6. Reasons the equipment should remain unused until evaluated

Do not invent specifications from an unknown model.

Example 5: government benefits

Help me understand the general application process for [PROGRAM].

Do not decide whether I qualify.

Use current official sources if available and identify their dates.

Separate:

1. General eligibility categories
2. Information applicants may need
3. Steps in the application process
4. Current deadlines or renewal requirements that must be verified
5. Appeal or assistance resources listed by the agency
6. Questions I should ask the responsible office

Do not request my Social Security number, benefit number, account credentials, or complete private records.

Activity: support or authority?

Classify each proposed use as:

1. Appropriate general assistance
2. Assistance requiring verification
3. Human decision required
4. Urgent human response required

Suggested discussion

• Scenario A is usually appropriate general assistance.
• Scenario B is educational but requires verification against the current official guide.
• Scenario C belongs to the authorized agency and applicable review process.
• Scenario D requires prompt human medical or emergency response—not chatbot delay.
• Scenario E is appropriate preparation.
• Scenario F requires independent authentication and existing financial controls.
• Scenario G is usually a lower-risk transformation if the notes are appropriate to share.
• Scenario H requires accountable human review, relevant evidence, due process, and organizational authority.

Copyable prompt: convert a decision into a human-handoff plan

I am facing this high-impact question:

[GENERAL, NON-SENSITIVE DESCRIPTION]

Do not make the final decision.

Create a human-handoff plan with:

1. What AI can safely help organize or explain
2. What AI must not decide
3. The type of qualified or authorized human needed
4. Current official sources to check
5. Information or documents to prepare
6. Questions to ask
7. Warning signs requiring more urgent action
8. Privacy information I should not enter into an AI system
9. A clear statement of who remains responsible for the decision

Do not delay an emergency response, invent professional conclusions, or guarantee an outcome.

The accountable-decision card

Before acting, complete:

• Decision: What is being decided?
• Consequence: What could happen if it is wrong?
• AI role: What limited assistance did the tool provide?
• Evidence: What reliable information supports the decision?
• Human authority: Who is qualified and authorized to decide?
• Review power: Can that person correct, reject, or stop the action?
• Record: Who accepted final responsibility?
• Repair: What happens if the decision causes harm or proves wrong?

When the chatbot becomes a barrier

A chatbot should not prevent access to a person when:

• The problem is complex
• The user is invoking a legal or consumer right
• The system repeatedly misunderstands the issue
• The answer may create fees or financial harm
• The person requests human assistance
• The situation requires discretion, compassion, or local judgment

Psychological safety

A person facing a high-impact problem may already feel fear, confusion, shame, or urgency.

The appropriate educational response is:

• Reduce unnecessary complexity
• Identify the next human contact
• Preserve privacy
• Avoid false reassurance
• Avoid catastrophic speculation
• Separate immediate safety from later research
• Use one manageable next step

Human judgment check

Before relying on AI in a consequential situation, ask:

• Is anyone in immediate danger?
• Could delay cause harm?
• Does this require a license, official authority, or local knowledge?
• Is the model applying general information to an individual case?
• Does the reviewer possess the knowledge and authority to challenge the output?
• Can the decision be appealed, corrected, or reversed?
• Is a real person available?
• Who will answer for the decision?

Next lesson

Lesson 11 explores locally meaningful uses for families, workers, farms, small businesses, libraries, and community organizations. Its examples will remain adaptable to community survey results.

Human-oversight and high-impact-use standards are documented on AI Education: Sources and Standards .

Local knowledge is part of the evidence

A general-purpose AI system may know broad patterns. It may not know:

• The current procedure used by a particular county office
• The soil, weather, pests, water conditions, and growing practices of a specific location
• Which services are currently available in a small community
• The culture and history of a neighborhood or tribal community
• The actual equipment used by a local business or farm
• The policies of a particular school, employer, library, or nonprofit
• The relationships and informal knowledge that help a community function
• Whether a program, grant, event, business, or office is still operating

A useful community interaction therefore combines three things:

By the end of this lesson, you should be able to:

• Identify practical AI uses relevant to local life.
• Choose a limited task instead of delegating an entire responsibility.
• Pair an AI task with an appropriate local or official source.
• Identify who should review or authorize the final result.
• Protect customer, family, employee, student, patient, and community information.
• Account for rural internet, device, cost, and accessibility constraints.
• Evaluate whether the tool reduced total burden or merely shifted work and risk elsewhere.
• Design a community use case using the Local Use Canvas.

The Local Use Canvas

Before using AI for a local task, complete these seven fields.

Families and households

Lower-risk household uses may include:

• Turning rough notes into a household checklist
• Creating questions for a school, agency, or professional
• Rewriting a family announcement in clearer language
• Brainstorming low-cost activities
• Creating a fictional practice budget
• Organizing a non-sensitive moving or travel checklist
• Explaining a public document after private details have been removed
• Practicing how to discuss technology with children or older relatives

Family planning prompt

Help our household organize this general task:

[TASK]

Constraints:
- available time: [TIME],
- approximate budget: [BUDGET],
- available transportation: [TRANSPORTATION],
- devices or internet: [ACCESS],
- accessibility needs: [NEEDS].

Create:
1. A short goal
2. A preparation checklist
3. An ordered action list
4. Questions we still need to answer
5. Decisions that remain with the family
6. A non-AI alternative

Do not request names, account numbers, medical records, school records, exact addresses, or children’s identifying information.

Do not invent local services, prices, hours, eligibility, or availability.

Workers and job seekers

AI may help with:

• Interview practice
• Organizing actual work history
• Rewriting a résumé bullet for clarity
• Comparing a résumé with a public job announcement
• Preparing questions for an employer, union representative, or workforce adviser
• Drafting a professional email
• Identifying skills that may require further training
• Creating a learning plan from public training information

Job-announcement comparison prompt

Compare the public JOB ANNOUNCEMENT with the de-identified EXPERIENCE NOTES I provide.

Do not invent experience.

Create four sections:

1. Requirements clearly supported by my notes
2. Related experience that may need clearer wording
3. Requirements not supported by my notes
4. Questions I should answer before applying

Then draft possible résumé language using only the supported experience.

Do not include my name, address, phone number, references, employee number, or other identifying information.

JOB ANNOUNCEMENT:
[PASTE PUBLIC TEXT]

EXPERIENCE NOTES:
[PASTE DE-IDENTIFIED NOTES]

Farms, gardens, forests, and land stewardship

AI may assist with:

• Summarizing a supplied Extension publication
• Organizing field observations
• Creating questions for an Extension agent, veterinarian, forester, soil specialist, or crop adviser
• Building a maintenance checklist from an equipment manual
• Comparing documented options
• Drafting educational or marketing material
• Organizing non-sensitive farm or garden records
• Brainstorming what additional observations may be useful

Extension-grounded prompt

Using only the Extension or agency text I provide, help me understand this agricultural topic.

Create:

1. Plain-language summary
2. Conditions the guidance applies to
3. Conditions it does not address
4. Observations I should record
5. Questions for a local Extension agent or qualified adviser
6. Any warnings, restrictions, or verification steps stated in the source

Do not identify a pest, diagnose an animal, recommend a pesticide, calculate an application rate, or invent a local treatment plan.

SOURCE TEXT:
[PASTE CURRENT PUBLIC SOURCE]

Cooperative Extension and land-grant institutions can provide research-based information adapted to regional conditions. Their local expertise should not be replaced by a generic model.

Small businesses and self-employed workers

Bounded small-business uses may include:

• Brainstorming names, themes, or campaign candidates
• Drafting customer communication for human review
• Creating a first-draft FAQ
• Organizing a meeting agenda
• Turning an existing process into a draft checklist
• Practicing customer-service scenarios
• Summarizing public market information
• Preparing questions for an accountant, attorney, insurer, lender, or Small Business Development Center

Small-business task prompt

Help me create a first draft for this small-business task:

[TASK]

Business context:
[GENERAL NON-CONFIDENTIAL CONTEXT]

Requirements:
- Use plain language.
- Match this general tone: [TONE].
- Use placeholders for business names, customers, prices, dates, and contact information.
- Label all ideas as candidates for review.
- Do not make promises about price, availability, performance, safety, legality, or results.

Evidence:
Use only the public or non-confidential information I provide.

Safety:
Identify any statement that should be reviewed by an accountant, attorney, insurer, regulator, lender, technical expert, or other qualified person before publication or action.

Libraries, schools, and educators

AI may support:

• Drafting a program description
• Creating fictional practice examples
• Reformatting supplied public text
• Generating discussion questions
• Building a draft lesson outline
• Brainstorming activities for different access needs
• Translating technical vocabulary into plain language
• Identifying claims learners should verify

Human educators and library workers remain responsible for:

• Instructional quality
• Age appropriateness
• Accessibility
• Intellectual freedom
• Privacy
• Source selection
• Community context
• Correction of errors

Nonprofits, civic groups, and community organizations

Possible uses include:

• Drafting an agenda
• Organizing public meeting notes
• Creating volunteer checklists
• Brainstorming outreach channels
• Summarizing anonymous community-survey themes
• Preparing questions for a potential funder
• Drafting a public announcement
• Creating alternate plain-language formats

Community-survey synthesis prompt

Analyze the anonymous survey responses I provide.

Do not infer identities, diagnoses, political beliefs, protected traits, or private facts.

Create:

1. Frequently repeated needs
2. Less common but important needs
3. Accessibility requests
4. Questions requiring follow-up
5. Areas where responses disagree
6. Ideas for future programming
7. Limitations of the survey

Do not claim that the survey represents the whole community unless the sampling method supports that conclusion.

SURVEY RESPONSES:
[PASTE ANONYMOUS RESPONSES]

Local history, archives, and genealogy

AI may assist with:

• Transcribing clearly legible public documents
• Summarizing supplied historical material
• Creating research questions
• Identifying names, dates, and places to verify
• Comparing two transcriptions
• Drafting descriptive metadata for review

Historical work requires special care because a model may turn a missing record into a plausible story.

Community, tribal, family, and cultural records may also carry privacy, sovereignty, sacred-knowledge, or future-association concerns. Permission to access material does not automatically grant permission to publish or upload it.

Source-bounded local history prompt

Using only the historical source text I provide, create:

1. People explicitly named
2. Dates explicitly stated
3. Locations explicitly stated
4. Events explicitly described
5. Exact quotations
6. Unclear or damaged passages
7. Questions requiring another source

Do not fill gaps with invented narrative.

Label interpretation separately from the source record.

Accessibility and language support

AI may help generate candidate versions of supplied material, such as:

• Plain-language summaries
• Shorter sections
• Reading-aloud scripts
• Step-by-step instructions
• Draft translations
• Image-description candidates
• Alternate examples

Low-bandwidth and no-device participation

AI education does not have to depend on every participant owning a laptop or maintaining a fast connection.

A workshop may use:

• Printed prompt cards
• Pre-captured example outputs
• One facilitator device projected for the room
• Small-group prompt design on paper
• Paired activities
• Read-aloud examples
• Downloadable or printable take-home materials
• Activities that compare an AI-assisted route with a non-AI route

Activity: design one local use case

Choose a real but non-private community task. Complete:

LOCAL USE CANVAS

Community goal:
[WHAT SHOULD IMPROVE?]

Bounded AI task:
[WHAT SMALL PART MAY AI ASSIST WITH?]

Grounding source:
[WHICH CURRENT LOCAL, OFFICIAL, OR PROFESSIONAL SOURCE?]

Private information:
[WHAT MUST NOT BE ENTERED?]

Human owner:
[WHO REVIEWS AND REMAINS RESPONSIBLE?]

Verification:
[WHAT MUST BE CHECKED?]

Stop condition:
[WHAT WOULD CAUSE A PAUSE OR HUMAN HANDOFF?]

Non-AI alternative:
[HOW COULD THIS BE DONE WITHOUT AI?]

Copyable local-source prompt

Help with this local or community task:

[TASK]

Community context:
[GENERAL, NON-SENSITIVE CONTEXT]

Local evidence:
Use only the sources I provide. Do not assume that generic national information applies locally.

Requirements:
- Separate source-supported information from general suggestions.
- Identify missing local information.
- Use plain language.
- Provide one low-cost or low-technology option.
- Provide a non-AI alternative.
- Use placeholders for identifying details.

Safety:
- Do not invent local laws, offices, programs, services, prices, hours, contacts, funding, community support, or availability.
- Do not upload or request private information.
- Identify the accountable person or organization that must review the result.
- State what should be verified before action.

SOURCES:
[PASTE PUBLIC OR REDACTED SOURCE TEXT]

The community off-loading audit

Before calling a local AI project successful, ask:

• Did it reduce total work after review and correction?
• Did it create unpaid review work for staff or volunteers?
• Did it shift risk onto residents with less technical access?
• Did it require unnecessary collection of community data?
• Did it preserve access to a human?
• Did it improve or weaken local expertise?
• Did it make the process more understandable?
• Could residents decline the AI route?
• Who receives the benefit?
• Who carries the cost when the system is wrong?

Human judgment check

Before using AI for a community task, ask:

• Is the task clearly bounded?
• Does it serve a documented community need?
• Were affected people invited to provide input?
• What local evidence grounds the output?
• What information should remain private?
• Who reviews and owns the result?
• Can a person access a human alternative?
• Can the project be paused or discontinued?
• Did the technology materially improve the work?

Next lesson

Lesson 12 turns the course into a personal starter kit. You will assemble a safe default prompt, privacy boundary, verification process, scam response, human-handoff rule, and first practical task.

Library, rural education, small-business, Extension, and community-participation sources are listed on AI Education: Sources and Standards .

Your starter kit belongs to you

This course does not end by requiring you to adopt a particular AI platform. Your personal starter kit should work whether you:

• Use AI occasionally
• Use it only at work or school
• Use it through a library demonstration
• Prefer a local or privacy-focused tool
• Use only printed examples
• Decide not to use AI for certain tasks
• Decide not to use it at all

By the end of this lesson, you should be able to:

• Use a safe default beginner prompt.
• Apply a personal information boundary.
• Match verification effort to the consequence of error.
• Recognize when a human should take over.
• Use the PAUSE scam response.
• Review a product’s privacy, history, memory, and data-use controls.
• Define personal stop conditions.
• Complete one lower-risk first task.
• Reflect on whether the tool genuinely helped.

Your eight-part starter kit

1. Safe default beginner prompt

You are my beginner AI guide.

Use plain language, short sections, and one step at a time.

Begin by restating what you think I am trying to accomplish. Ask only the questions necessary to understand the task.

Do not ask me for passwords, authentication codes, account numbers, Social Security numbers, identification documents, private medical records, confidential work records, children’s identifying information, or another person’s private information.

Separate your response into:

1. What is well supported
2. What you are assuming
3. What is uncertain or missing
4. What I should verify
5. A practical next step
6. When a qualified human or official source is needed

Do not invent facts, quotations, laws, statistics, names, contact details, links, sources, or citations.

Treat your response as a candidate for me to review—not as final authority.

I may revise, verify, reject, pause, or continue without AI.

You may copy this prompt at the beginning of a new conversation. It cannot guarantee safe or accurate output, but it makes your expectations visible.

2. Personal information boundary

Create a paper or digital card that says:

I WILL NOT CASUALLY ENTER:

- passwords, PINs, authentication codes, or recovery codes;
- Social Security, driver-license, passport, benefit, or tax-identification numbers;
- bank accounts, payment cards, or financial credentials;
- full medical, legal, tax, school, employment, or confidential records;
- children’s identifying information;
- precise home, travel, security, or daily-routine information;
- another person’s private information without permission;
- confidential workplace, customer, client, student, or organizational information;
- intimate images, biometric information, or sensitive voice recordings;
- information that an approved secure system or qualified human should handle.

3. Verification ladder

4. Human-handoff rule

I WILL MOVE THE TASK TO AN APPROPRIATE HUMAN WHEN IT INVOLVES:

- immediate danger or an emergency;
- medical diagnosis, treatment, medication, or urgent symptoms;
- a legal deadline, legal strategy, custody, immigration, or enforceable right;
- tax filing, benefits eligibility, credit, investment, insurance, or financial transfer;
- identity verification, passwords, security codes, or account recovery;
- hazardous repair, electrical work, gas, structure, vehicles, or critical equipment;
- hiring, firing, discipline, admission, grading, benefits, or access decisions about another person;
- a child or vulnerable person;
- confidential organizational information;
- a situation I cannot independently understand or review;
- or a decision that cannot be safely reversed.

5. Scam PAUSE card

P — PAUSE.
Do not send money, reveal a code, open an attachment, scan a QR code, or click a link.

A — AVOID THE SUPPLIED CHANNEL.
Do not use the caller’s number, message link, or supplied website as your only verification route.

U — USE A KNOWN CONTACT.
Contact the person or organization through information you already know or locate independently.

S — SECURE MONEY AND ACCOUNTS.
If anything was shared, contact the institution, change credentials, and protect affected accounts.

E — ESCALATE AND REPORT.
Tell a trusted person and report the incident through appropriate official channels.

FTC guidance similarly recommends slowing down, using contact information you locate yourself, refusing urgent or unusual payment requests, and never sharing verification codes.

6. Product-settings check

Before using a particular product, review its current controls. Look for:

• Conversation history
• Temporary or non-history modes
• Model-improvement or training controls
• Saved memory
• Connected applications
• File and image handling
• Data export
• Conversation deletion
• Account deletion
• Business, school, workplace, or administrator access
• Current privacy policy and retention information

Memory is not authority

If an AI product remembers something, that information may still be:

• Incorrect
• Outdated
• Misunderstood
• Missing its original source
• Something you no longer consent to retain
• Inappropriate for the current task

Review, correct, or remove saved information when the product provides those controls. Do not treat a remembered statement as a substitute for checking its origin.

7. Personal stop conditions

Decide your stop conditions before a difficult interaction.

I WILL STOP OR CHANGE ROUTES WHEN:

- the system requests information that should remain private;
- the task becomes medical, legal, financial, tax, emergency, or safety critical;
- the answer repeatedly invents facts;
- I cannot locate the required source;
- the system pressures me to continue or isolate myself from other advice;
- the answer becomes frightening, manipulative, or overly personal;
- verification takes more effort than completing the task another way;
- the output affects another person who has not reviewed it;
- I no longer understand what the system is doing;
- a human or official source is clearly more appropriate;
- or I simply do not want to continue.

8. Choose your first bounded task

Select one lower-risk task:

• Rewrite a fictional or non-sensitive message
• Turn rough project notes into a checklist
• Explain a short public paragraph in plain language
• Generate interview-practice questions
• Brainstorm community-event ideas
• Create questions to ask a professional
• Reformat supplied material into shorter sections
• Draft a fictional story or creative exercise

Your first 15-minute practice session

1. Choose one task from the list above.
2. Remove private information before opening the AI tool.
3. Paste the Safe Default Beginner Prompt.
4. Add the CARES description of your task.
5. Request one candidate response.
6. Review it using the Helpful and Honest check.
7. Verify any factual claim that matters.
8. Decide what to keep, change, remove, or reject.
9. Stop after 15 minutes and reflect on whether the tool actually helped.

Before, during, and after checklist

Copyable reset prompt

Stop the current approach and reset.

You misunderstood or exceeded the task.

The actual goal is:
[GOAL]

The boundaries are:
[BOUNDARIES]

Keep:
[USEFUL PART]

Remove:
[UNSUPPORTED, PRIVATE, OR UNHELPFUL PART]

Before producing another answer:
1. Restate the corrected task
2. List the evidence you have
3. List what is missing
4. Wait for my confirmation

Do not repeat the previous conclusion as though it were established.

Copyable uncertainty prompt

Review your previous response for overconfidence.

Create a table with:

- claim,
- information basis,
- confidence posture,
- missing evidence,
- consequence if wrong,
- verification method,
- and appropriate human reviewer.

Use these confidence labels only:

- WELL SUPPORTED
- PLAUSIBLE BUT UNCERTAIN
- UNSUPPORTED
- OPINION
- PROFESSIONAL REVIEW REQUIRED

Do not invent evidence to improve the confidence label.

Copyable human-handoff prompt

This task may require human expertise or authority.

Do not make the final decision.

Help me prepare:

1. A concise non-sensitive description of the issue
2. The type of person or organization I should contact
3. Questions to ask
4. Sources or documents to bring
5. Dates or facts to verify
6. Warning signs that require more urgent action
7. Private information I should keep out of the AI conversation
8. A clear next human action

Do not delay emergency or professional help.

Personal AI use agreement

Complete this statement for yourself, your household, a class, or a community group:

MY AI USE AGREEMENT

I may use AI to assist with:
[BOUNDED TASKS]

I will not use AI as the final authority for:
[HIGH-IMPACT TASKS]

I will not casually enter:
[PRIVATE INFORMATION]

For important factual claims, I will verify through:
[SOURCES OR PEOPLE]

I will ask a human to take over when:
[HANDOFF CONDITIONS]

I will stop the interaction when:
[STOP CONDITIONS]

I will remember:
An AI answer is a candidate—not a verdict.

Course reflection

You may write, discuss, or consider these privately:

• One task I might use AI for:
• One task I will keep human-led:
• One kind of information I will not share:
• One way I will verify an answer:
• One warning sign I will recognize:
• One person or organization I can ask for help:
• One feature I want explained more clearly:
• One future class I would attend:

For families, partners, and caregivers

Consider creating a shared verification agreement:

• We will not send money because of an urgent message without calling each other through a known number.
• We will not shame someone who asks whether a message is a scam.
• We will not upload another person’s private records without permission.
• We will identify which decisions require a qualified human.
• We will help each other slow down before acting.
• We will keep a current list of trusted contacts.

Accessibility and learning choice

Your starter kit may be:

• Printed in large type
• Stored as a phone note
• Read aloud
• Reduced to a one-page card
• Represented with icons
• Completed with a partner
• Used one step at a time
• Translated and reviewed by a qualified person

There is no single correct way to demonstrate AI literacy. You may learn by reading, listening, observing, discussing, practicing, teaching someone else, or choosing not to use a particular tool.

Final off-loading check

After using AI, ask:

• Did the tool save meaningful time?
• Did review and verification erase the time saved?
• Did it improve understanding?
• Did it create new privacy or factual risk?
• Did it preserve my skills?
• Did it shift burdens onto someone else?
• Do I understand the final result?
• Can I explain why I accepted or rejected it?
• Could I complete the task without AI next time?

You have completed the beginner learning path

Completion does not mean knowing everything about artificial intelligence. It means you now have a repeatable process for approaching unfamiliar tools:

UNDERSTAND
What is the system doing?

PRACTICE
What bounded task may it assist with?

VERIFY
What claims and sources must be checked?

PROTECT
What information, decision, and human relationship must remain safeguarded?

DECIDE
Keep, revise, reject, escalate, or continue without AI.

Continue learning at your own pace. Return to any lesson when a new task or technology appears.

Accessibility, adult-learning, privacy, human-agency, verification, and UVLM governance sources are documented on AI Education: Sources and Standards .