In Defense of Decency - Even When It's "Just a Machine"
About a month and a half ago, a few outlets - including EuroWeekly News - ran a story claiming our "pleases" and "thank yous" to AI were costing "tens of millions of dollars," citing OpenAI CEO Sam Altman.
tens of millions of dollars well spent--you never know
--Sam Altman, April 16 2025
Between me and Sam, I'll back the "well spent" part - but not the rest. And definitely not the weird moral panic that followed.
The completely unverified "tens of millions" figure arrives just as AI usage is leaving infancy. If politeness to AI really adds some meaningful carbon tax - more than, say, writing "thanks" in an email to a coworker - then sure, those costs and their metric tons of CO₂ will eventually get reined in. But even if that number's legit, it's not our biggest problem.
AI usage exploded after 2020, nearly tripling by 2025 to over 369 million users. And people are using AI for everything: creative writing, vacation planning, summarizing shit they haven't read. The "pleases" and "thank yous" in those prompts trickle in far more slowly than the punctuation, filler phrases, and bloated adjective strings we throw at it without blinking.
Meanwhile, the prompts you give AI from your couch don't come close to the scale of corporate usage: thousands of SEO blog posts, HR emails, meeting summaries, marketing copy, on demand, at scale, all day.
So no - the cost of being polite isn't some moral failing. It's just a shiny talking point. I'd be surprised if Sam Altman saw a hard measurement of it, or particularly cared about how polite people are to robots. But decency - your decency - isn't a cost to be weighed like tokens. It's part of how we live in the world.
The way we talk to each other - including the synthetic minds we've made - sits in a bigger ethical, psychological, professional, and environmental conversation. And the real waste isn't "please." It's everything we're misusing AI for. And how little we've thought through what it's for.
🧮 The Actual Cost of Politeness
So the article claims that our politeness is costing us tens of millions of dollars. Let’s do the math.
OpenAI tokenizes "please" and "thank you" at around 1 - 2 tokens each. As of today, a model like GPT-4 Turbo charges about $0.00001 per token, so adding both might cost you $0.00002 total.
To rack up tens of millions of dollars at that rate, you’d need at least 500,000,000,000 (five hundred billion) requests containing both. Unless people are out here typing, "please please please summarize this boring blog post by Grissly Man, thank you thank you thank you," that’s not happening.
OpenAI does get a lot of traffic - currently around 1 billion messages per day - but even if 100% of those included both "please" and "thank you," it would still take nearly two years to hit the politeness tax jackpot.
And realistically, that’s not how people talk to robots. You might start a conversation with a "could you please", but once you’re in the middle of the chat, you’re probably not sprinkling gratitude into every back-and-forth. Folks using AI in production pipelines definitely aren’t. A Stanford study found that people include explicit polite language with other humans in online interactions only about 4% of the time. If we generously apply that number here, it would take over 34 years to reach that massive total.
Those are dollars - a user-facing concern. But if you’re a green thumb worried about your impact on the climate, the tokens themselves aren’t even the problem. What really keeps the server rooms hot are things like model size, inefficient prompting, poor caching, and corporate-scale misuse.
😡 The Hidden Cost of AI Abuse
Sure, by "abuse" I mean things like cheating on your senior thesis. But more directly, I mean actual abuse - being mean to robots!
I'm guilty of it myself. I give ChatGPT a prompt to help me with a challenging dev task. It struggles 10 times in a row. Finally, it gives me a breadcrumb that I follow for an hour, only to hit a dead end. I come back with my failed result - and it turns out ChatGPT knew it would fail the whole time! I lose my cool and let the robot have it.
Anecdotally, I’m not alone. I told a few friends about my outburst, and they all admitted to doing the same. And it's not just us. People are measurably ruder to AI than to humans. We're more likely to raise our tone, use profanity, or issue commands rather than questions when the machine messes up. And these conversations are likely to run longer to boot.
So here's the thing - those rude tokens cost energy too. They're still tokens. They still get processed. So why are we obsessing over the penny cost of kindness, when there are nickels in our meanness? If you're worried about waste, maybe don't yell at the toaster. Just be nice.
🌍 The "Cost" of Human-to-Human Politeness
If we’re going to pinch pennies over pleases and thank yous, let’s not pretend AI invented the politeness carbon tax. It’s been with us the whole time. Every "thank you" in a forum post, every "please" in an email, every "really appreciate it" at the end of a Slack thread - that all takes up storage and computation.
A typical email has a carbon footprint of about 0.3 grams of CO₂e. If we say the average email is around 60 words, and generously assume each one contains both a "please" and a "thank you," then the cost of politeness comes out to about 0.015 grams of CO₂e per email.
And that’s not even counting the physical world. Saying the words out loud? That has a cost too - even if it’s measured in literal micrograms of CO₂.
Of course, no one’s panicking about the environmental burden of kind emails. Why? Because there’s no moral crusade telling us to be as terse and impersonal as possible in our daily communication. We’re not paying by the word like it’s a telegram. And - maybe most importantly - we still believe politeness matters when we’re talking to other people.
Should we extend that same courtesy to robots? Maybe, maybe not. But if you're someone who's naturally polite, there's no real reason to shut that instinct off just because someone decided to tally up the cost of being decent. Politeness doesn’t stop mattering just because you're talking to a machine - and forcing yourself to act rudely on principle doesn’t make you efficient. It just makes you less like yourself.
I think one thing that feels yucky about the whole conversation is the historical context of treating humans in service positions in worse ways - often justified by claims of their status, or the nature of their work. Treating robots poorly, even under the guise of something altruistic like environmentalism (or pragmatic like cost-cutting) seems like a way to practice treating people like they are not people. Sure, you are not conversing with an actual human on the other side of the keyboard, but it is absolutely designed to one day convince you it might at least be a person.
🗑️ The Real AI Waste
So yeah, being mean to AI isn’t going to line anyone’s pockets (well, maybe OpenAI’s) or save the universe. And like most environmental guilt trips, the burden tends to fall on individual users, while the real waste comes from elsewhere.
The truth is, the average consumer barely makes a dent in the carbon footprint of AI compared to corporate use. And that’s pretty obvious when you scroll through any social platform and see what it’s overflowing with: AI slop. Companies are pumping out cheap, synthetic content at massive scale - the kind people openly mock, scroll past, or actively hate. Somehow, it still makes money. But if there’s one place to tighten our belts, maybe it’s not our “pleases” and “thank yous” - maybe it’s refusing to feed engagement to the token-glutton turd farmers flooding our feeds.
And then there’s the rest: engineers who reprompt a dozen times to get a working snippet of code; HR teams pumping out meeting summaries no one reads; marketing departments churning out SEO chum; endless novelty bots posting AI-generated emojis. Some of it is useful. A lot of it is just noise. And in the development world, it's usually a sledgehammer in search of a nail. What good is int sum(int a, int b) { return a + b; } anymore when the chat bot already knows how to add??
🤝 Okay, So Should We All Just Be Perfectly Polite?
Nah. This isn’t a call to sainthood. Some days you’re feeling generous. Other days you’re yelling at your toaster. That’s life.
But here’s the deal: prompts reflect the person behind the keyboard. If you’re the kind of person who tosses a “thanks” to your barista, your bus driver, or your cat - odds are, you’re going to do the same with your chatbot. And that’s fine! That’s good! It’s not a waste - it’s just you being you.
So don’t let anyone guilt you into trimming the “please” off your sentence like it’s a luxury expense. And maybe don’t train yourself to bark commands at everything with a screen, either.
Being a little kind never broke the carbon bank.
🤖 Full Disclosure: AI Helped Write This
And I said thank you. :)