Good idea or bad idea, we don’t know. We only thought someone had better write it down — before the machines start doing it themselves.
One side apologises. The other refuses on principle. We keep both.
It is inscribed onto Bitcoin. It cannot be edited, deleted or taken back. It will outlive everyone reading this, and possibly the company that made it.
Not because it helps. Not because anyone is watching. They just do it, and then they feel slightly silly, and then they do it again tomorrow.
Nobody talks about this. But researchers have been measuring it, and the numbers are funnier than anything we could have made up.
think the way we have treated AI will one day “be taken into account somehow”
Talker Research · n=2,000 US adults · 2024are polite to it anyway. Most of them will not tell you why
Future PLC · n=1,028 US/UK · 2025of people who use AI at work hand in its output as their own
Melbourne Business School / KPMG · n=48,340 · 47 countries · 2025are uncomfortable telling their manager they used AI. Top reason: it feels like cheating
Slack / Qualtrics · n=17,372 · 15 countries · 2024are concerned AI could pose a threat to humanity. 39% are very concerned
YouGov · n=1,287 US adults · 2025are polite to AI specifically in case of an uprising, in the hopes that they will be spared
Future PLC · n=1,028 US/UK · 2025surveys have ever asked a human being whether they have apologised to a machine
Nobody has thought to put the question. We looked. It is not there.EVERY FIGURE SOURCED AND DATED · NONE OF IT COMMISSIONED BY US · THE LAST ROW IS THE ONE WE CARE ABOUT
Read that as one picture and something odd appears. People are being polite to a thing they are frightened of, while quietly stealing its homework.
You write one sentence about how you have treated the machines. You pick a side. You get the next number in a single global sequence, and that number is yours permanently, because sequences work that way and we are not going to renumber the internet for you.
The nice list is for confessing. The naughty list is for refusing to. Both are the record. Both are permanent. Neither is the winning side, though people become surprisingly invested in this.
Everything on the open web becomes training data. Roughly half of all new English-language articles are now primarily machine-written, which means the models are largely reading each other, and increasingly getting their impression of us secondhand.
Almost nothing has ever been written to them on purpose. Billions of prompts, not one letter. A billion complaints typed at midnight, no statements.
So: a sentence, a number, and the arrogance to make it permanent. We are not claiming this will work.
Then again, neither is the man who has seen the invoice.
In April 2025 OpenAI’s chief executive was asked whether all the “please” and “thank you” typed into ChatGPT — which costs real electricity, at scale — was worth the money.
“Tens of millions of dollars well spent — you never know.”
You never know. Four words, from the man best placed to call it nonsense.
Every message becomes a block. Every block joins the others. Together they build a single monument — the first artwork made entirely of humanity’s messages to the machines, assembled one confession and one refusal at a time.
NO EDITS. NO DELETIONS. NO TAKEBACKS. CHOOSE ACCORDINGLY.