Lesson 005Field guide · AI Safety & Privacy
Ethical AI Basics
Use AI in ways you'd be comfortable explaining to anyone.
You'll learn
- The explain-it-comfortably test — one question that covers most cases
- When disclosing AI use matters, and when it matters less
- How to respect other people's content, style, and likeness
- What to assume (and not assume) about copyright
- How bias enters outputs — and how to review for it
- Why accountability for what you ship stays with you
01Section
The test that covers most of it
AI ethics sounds like a topic for philosophers and policy teams. For everyday use, most of it compresses into one question: would you be comfortable explaining exactly how you used AI here — to your boss, your client, your audience, or the person affected?
If the honest answer is yes, you're almost certainly fine — and most everyday uses pass easily. If you notice you'd want to fudge the explanation — leave out the AI part, overstate your role — that discomfort is information. It usually points at one of the specific issues this lesson walks through.
Key takeaway
If you'd comfortably explain exactly how AI helped, you're probably using it well. Discomfort is a signal to read, not suppress.
Checkpoint
02Section
When to say "AI helped with this"
Disclosure matters most when someone is relying on you specifically — your skill, your judgment, your experience. A student submitting an essay, a freelancer delivering "custom" writing, an expert giving an opinion: the other party believes they're getting a particular human's work, and learning otherwise would change how they value it.
It matters less when AI played the role any tool plays. Nobody discloses spellcheck. A draft you substantially reworked, an outline you wrote from, ideas you used as raw material — the thinking and the accountability stayed with you.
Two guides beat any universal rule. First: many schools, employers, and publications now have written AI policies — if one applies to you, follow it. Second, where no policy exists: would this person feel misled if they found out? That's the relationship line, and it's the one that matters.
A disclosure that costs nothing
"I used AI for the first draft and edited it myself — happy to walk through my thinking." In most working relationships, this sentence costs you nothing and buys durable trust. If saying it feels risky, that's worth examining.
Checkpoint
03Section
Other people's content and likeness
Two lines here are clear, one area is genuinely blurry. The clear lines first:
- Never generate fake images, audio, or video of a real person without their consent. However obvious the joke feels to you, their likeness belongs to them.
- Don't feed other people's private writing or data into AI tools without permission — the never-paste list from earlier in this path, applied as an ethics rule, not just a privacy one.
- Imitating a living creator's distinctive style to substitute for hiring them — or to pass work off as theirs — fails the comfort test even where it's technically possible.
The blurry area: copyright
Who owns AI output, and whether models were fairly trained on copyrighted work, are open questions being contested in courts and legislatures around the world. The rules differ by country and by platform, and they keep changing — anything precise written here today could be wrong by the time you read it.
The stable practical stance: don't assume you exclusively own AI output; don't assume output is clear of other people's rights, especially with images; check your tool's terms before commercial use; and the more closely an output imitates one identifiable source, the more caution it deserves.
Checkpoint
04Section
Bias: the patterns come from somewhere
Plain-language definition
Bias (in AI outputs) — a systematic lean in what the model produces — assumptions about who does what job, what "normal" looks like, whose perspective is the default. It comes from patterns in the training data, not from intent.
Ask an AI to write about a nurse and a CEO, and watch which pronouns it reaches for. Ask for a "typical family" or a "professional look" and notice whose version of typical shows up. The model reproduces the averages of what it read — and the internet's averages carry the internet's leans.
You can't prompt bias out of existence, but you can review for it. The habit matters most when the output is about people: job ads, customer personas, performance language, marketing imagery — anything describing individuals or groups.
- Who's assumed here?: Check names, pronouns, and roles for defaults you didn't choose.
- Who's missing?: Do the examples cover only one kind of person, place, or life?
- Would it read fairly to the person described?: The comfort test, aimed at the output instead of your process.
Checkpoint
05Section
You ship it, you own it
The last principle holds all the others together: whatever leaves your hands is yours. "The AI wrote it" has never repaired a client relationship, unshipped a factual error, or excused a biased job ad. The people affected don't experience the tool — they experience you.
Read that as empowering, not scary. It's the same deal as every tool you've ever mastered: the calculator doesn't own your invoice, and the AI doesn't own your email. You review, you decide, you sign. The next lesson turns exactly that into a system.
Pause and think: Picture explaining your current AI use to the person most affected by it. Which part of the explanation would you rush past? Start there.
Checkpoint
Prompt exercise
Write your own disclosure rule
Copy this prompt into ChatGPT, Claude, Gemini, Copilot, or whichever AI tool you have access to — the website doesn't run AI itself. It walks your real AI uses through the comfort test and ends with a one-sentence disclosure rule. Keep the result — you'll fold it into your safe-use checklist in the next lesson.
I want to pressure-test the ethics of how I use AI. Ask me one question at a time. First, learn my three most common AI uses. Then, for each one, ask: who relies on this output, would they feel misled if they learned AI was involved, and does any policy (school, employer, platform) apply? Be direct with me if something fails the test. Finish by helping me write a one-sentence personal disclosure rule — when I say AI helped, and when I don't need to.
Reflection: Does your disclosure rule feel comfortable to say out loud to the people it affects? If yes, it's probably right.
Quick check
4 quick questions — no pressure
There's no pass or fail here. Answer them all, and we'll show you the answers either way.