To be or not to be courteous with AI, that is the question

I've always been courteous with artificial intelligence, using words such as "please" and "thank you". Not because I think it has feelings, or that politeness makes it perform better, but simply out of habit.

That habit wavered after a remark I heard at a training session. An expert argued that polite words are wasted on AI and, at best, merely add to the computing load. It made me wonder whether being polite to AI is irrational in an age when energy and resources matter.

Technically, the expert made sense. Politeness is unnecessary when giving instructions to large language models. As long as the request is clear, the tone makes no difference. Whether we say "Summarize this" or "Please summarize this", the output is the same. Adding polite words simply means more tokens to process and higher computation costs. In that narrow sense, calling it "waste" isn't entirely wrong.

But what unsettled me was the assumption behind the critique. If a form of expression doesn't improve output, we should drop it. Does that mean we should do that in our daily interactions as well?

Being polite, after all, is part of social protocol. It may not be a requisite for transmitting information, but it reduces friction, tempers the use of power, and may keep some interactions from sliding into aggression. We use polite language in emails, customer service scripts and automated replies not because the recipient needs emotional care, but because this mode of speaking makes cooperation smoother.

Should that logic be dispensed with when interacting with AI? Not necessarily.

AI is always available, never offended, never hurt, never uncooperative. In front of something that won't push back, humans can exert frustrated authority at will. Commands can be abrupt, even hostile, but the software will obey the commands. In such a setting, the question isn't about what AI needs, it's about what we choose to become when restraint is no longer needed.

More importantly, politeness can be practically useful — to me. When I begin with "please", I tend to write full sentences and specify what I want. When I end with "thank you", it reminds me this is collaboration, not an outlet for frustration. Maybe politeness doesn't matter to AI, but it does to me as it keeps me grounded.

Being polite to AI isn't inherently risky unless one starts assuming that AI "understands me", "it cares", "it keeps me company". For some users, mistaking an unemotional tool as having empathy can cause real harm.

So should I stop saying "please" and "thank you" to AI?

I don't think I should be indifferent just to save computing time and energy. If I truly care about resource use, there are far more effective steps: ask for shorter answers, avoid repetitive follow-ups, and stop the model from over-explaining what I already know. That would save more computing costs than deleting a single "thank you".

So, I have decided, I'll stick to a language that comes naturally to me. Politeness isn't something I owe to AI. It's a way of managing my own expression. In an era where efficiency increasingly presents itself as the only rational standard, preserving some civility — even when it isn't strictly "useful" — may not be wasteful at all.

AI doesn't require politeness. But humans might.