We’re avid users of AI: to learn, to critique our essays, and to point out how many absurd mistakes we’ve inserted. Somehow, it’s less painful coming from a bot.
But we will never (knowingly!) publish AI writing.
Below are our rules for AI writing. Add your own in comments below. And check out this essay on why human writing matters.
—Tom Rachman & Conor Griffin, AI Policy Perspectives
1: The Embarrassment Rule
If you’d feel ashamed by exposure, don’t use AI to write. (Or credit it)
Much as DNA testing is identifying the guilty years later, tomorrow’s AI-detection apps will expose today’s clandestine usage.
This is happening already, with suspicious readers uploading post-2022 fabrications that once seemed undetectable.
Rule-of-thumb: Assume your AI usage will be discovered.
2: The Intern Rule
Use AI in the process of writing. But treat its contribution as you’d treat that of a person.
It’s fine for AI to explain a subject when you’re finding your way, much as you might consult a knowledgeable colleague. But if your final text consisted of nothing but that colleague’s points, you’d be remiss to take full credit. Likewise with AI.
It’s fine for AI (or a human research assistant) to ferret out articles, and to summarize them, especially technical material and turgid writing. It’s not fine for you to read nothing yourself.
It’s fine for AI to help brainstorm your ideas. It’s not fine for AI to supply all the ideas without credit.
It’s fine to have AI proofread. It’s not fine for AI to rewrite.
Rule-of-thumb: Test whether you’re going too far by asking: Would it be wrong to accept this help from a human without crediting them?
3: The Ventriloquist’s Rule
If you want to learn a subject, put it into words. When AI writes, it deletes your future wisdom.
Sometimes you’ll just need an artifact. But remember that, if a machine is generating all of your words, you’ll end up as knowledgeable as the ventriloquist’s dummy.
Rule-of-thumb: To speak more fluently on a topic, write about it.
4: The No-Dumping Rule
Nobody wants to read outsourced AI replies.
If someone asks your view, and you ask a chatbot, and send them a 10-page document that you identify as AI, this is burden-dumping.
Nobody should read your burden-dumps. They could’ve created those without you.
Rule-of-thumb: If someone respects you enough to seek an opinion that you cannot offer, respect them enough to say no.
5: The Byline Rule
Your byline is an oath.
Readers assume that the named author wrote the text. If this is only partly true, your byline can be an act of deception.
Many types of writing are valuable without personal authorship: weather forecasts; instruction manuals; explanations about the world. Generate as much as you want. Don’t take the byline.
In other types of writing, authorship is the whole point—for instance, a school essay or a condolence letter. Never generate this.
News reporting and scientific research occupy an ambiguous midpoint, where opinions may differ. In their idealized form, each aspires to share knowledge above all else. If AI can write these findings better than the people doing the finding, there’s arguably no problem.
But a byline also establishes accountability for claims, and builds networks of thinkers. Not to mention the motivation of an author’s career ambitions and vanity.
I really like how, without any judgment or prejudice, these rules are described. More about transparency and good advice. "Remember that, if a machine is generating all of your words, you’ll end up as knowledgeable as the ventriloquist’s dummy." I love it!