Using AI Responsibly

Disclosure: when to say AI helped

Here's the plainest way to think about it: disclosure means telling people AI was involved when they'd reasonably want to know. Nobody expects you to footnote a grocery list. But three situations clearly cross the line into "say it":

• School and university assignments. Whatever the institution's policy says, that's the rule — full stop. If the policy allows AI for brainstorming but not writing, that's the line. If there's no policy, ask before assuming. Submitting AI-written work where the rules forbid it isn't a grey area or a clever hack; it's plain cheating, the same as copying a friend's essay.
• Published writing. If your name sits on an article, a blog post, or a report presented as your analysis, and AI wrote the substance of it, readers are being misled about whose thinking they're getting. Many publications now require an AI note; even where they don't, "AI-assisted" costs you one line and buys you a clear conscience.
• Work deliverables where it matters. Using AI to polish your report? Normal, like using spellcheck. Presenting an AI-generated market analysis as weeks of your own research? That's misrepresenting effort and — worse — misrepresenting how much verification stands behind it. When someone will make a decision based on how trustworthy they believe your process was, tell them what the process was.

And there's a purely selfish reason to keep this honest, especially for students: skills you fake are skills you don't build. Outsource every essay and you graduate with a certificate and without the ability it certifies — into a job market that, as the previous chapter showed, pays precisely for the abilities AI can't replace. The smart move is the opposite: use AI as a tutor that explains, quizzes you, and critiques your drafts — so the learning lands in your head. Learning with AI shows exactly how.

Accountability: the sender owns the message

Burn this one sentence into memory: AI's mistake in your email is your mistake.

When you hit send, sign, or publish, you become the author — legally, professionally, and socially. "The AI wrote that part" has never once rescued anyone's reputation, and it never will, for a simple reason: nobody forced you to ship it unread. The lawyer who filed AI-invented court citations wasn't punished for using AI; he was punished for not checking. The same logic applies at every scale — a wrong price quoted to a customer, a garbled fact in a school newsletter, a mistranslated condolence message.

The practical habit is small: anything that leaves your hands gets one careful human read first — and for facts, numbers, names, and dates, an actual check (chapter on hallucinations tells you how). Scale the care to the stakes: a dinner-plan prompt needs no audit; a message to your child's school, your boss, or a customer does.

Fairness: the bias that comes built in

AI models learned from oceans of human writing — and human writing carries human prejudice. The models inherit it, in subtler forms than you'd expect: a hiring-ad draft that describes the "ideal candidate" with quietly gendered wording, an evaluation that reads warmer for one name than another, an example list where doctors are all "he" and nurses all "she". The model isn't malicious; it's a mirror with a memory.

The rule of thumb: any AI output that describes, ranks, or decides about people deserves a second look with bias in mind. That includes job descriptions, CV screening notes, performance-review drafts, tenant or loan assessments, even a "who should we invite to speak" brainstorm. Two cheap countermeasures work surprisingly well: ask the AI itself to critique its output for bias ("review this job ad for wording that might discourage any group from applying"), and swap the names or genders in a draft to see if the tone changes. If it does, you've caught something worth fixing.

Other people's rights

Your AI use touches more people than you. Three lines worth drawing for yourself:

• Likeness and voice. Don't generate images of a real person, or clone a voice, without that person's consent — not as a joke, not "just for the group chat". You already know from the scams and deepfakes chapter how these fakes do harm; don't be the neighbourhood's small-scale version of the problem. (Public figures in obvious satire are a legal and ethical debate of their own — but private individuals are simply off-limits.)
• Other people's private information. Pasting your own details into a chatbot is your call. Pasting your friend's medical situation, your client's contract, or your colleague's salary is a different act: that information was given to you, not to a third-party service. The privacy chapter covers what happens to pasted text — the short version is, when in doubt, strip names and details that aren't yours to share.
• Artists and writers. Image and text models learned partly from creative work whose makers never agreed to it, and "generate this in the style of [living artist]" sits at the centre of a genuine, unresolved dispute — lawsuits, licensing deals, and honest disagreement among reasonable people. You don't have to settle that debate to act decently within it: prefer styles described by attribute ("watercolour, muted colours") over a living artist's name, and if you'd feel awkward showing the artist your prompt, that's information.

Two more honest notes: the planet, and the kids

The environment. Training a large AI model consumes serious energy and water — that's real, and worth knowing. Your individual chats, though, are tiny: one query costs roughly the energy of a few seconds of video streaming. So keep perspective in both directions. Don't lose sleep over asking AI for a recipe; equally, know that AI at global scale has a real footprint, which is a fair thing to weigh in public debates about data centres — just not a reason to feel guilty about your homework help.

Children and AI. Most chatbots set a minimum age (typically 13, with parental consent below 18) — those rules exist for good reasons, so check them for whichever app enters your house. Beyond the rules, two habits matter more than any parental-control setting: supervision — for younger kids, AI use happens with an adult nearby, same as the early internet years — and teaching verification early. A ten-year-old who hears "the AI sometimes makes things up — let's double-check it together" is learning the single most important digital habit of their generation. Kids who grow up treating AI as a clever-but-fallible assistant will be safer than kids who grow up treating it as an oracle.

The five-question check

Almost every "is this okay?" situation resolves in under ten seconds with five questions:

When a situation is genuinely fuzzy — a new client, an unclear school policy, a grey-zone disclosure — AI itself is a surprisingly good sounding board, as long as you ask it to challenge you rather than reassure you:

Here's my situation: [e.g. "I used AI heavily to draft a report for a client who doesn't know that. I plan to add this line: 'Prepared with AI assistance; all figures verified by me.'"]

Act as a tough but fair ethics reviewer:
1. Does my disclosure line honestly match what I actually did? Where might a reader feel misled?
2. What would full honesty look like here, in one sentence?
3. Play the other side: what's the strongest argument that I'm overthinking this — and do you buy it?
Don't reassure me. Point at the weak spot.

A personal code, in five lines

Policies differ by school and employer, but a personal standard travels with you everywhere. Here's one that fits on a sticky note:

Five lines, and notice what they add up to: the same standards you already hold for every other tool — honesty about your work, care with other people's trust, and ownership of what you put into the world. AI didn't change the ethics. It just raised the speed limit, which makes steering matter more.

Key takeaways

• Disclose AI's role where people would reasonably want to know: school assignments (per policy), published writing, and work where trust in your process matters. Where rules forbid it, AI-written work is plain cheating.
• The sender owns the message. AI's mistake in your email is your mistake — one careful read before anything ships.
• Models inherit human bias. Give a second, bias-aware look to anything AI writes about people — hiring, evaluations, rankings.
• Respect others' rights: no generated likenesses or cloned voices without consent, no pasting information that isn't yours to share, and be thoughtful about artists' styles.
• Keep perspective on the environment (training is costly; one chat is tiny), and teach kids verification habits early, with supervision.
• Five-line code: disclose when it matters, verify, own the output, don't fake skills, don't harm people.