Issue #4: The Cost of Being Polite to Machines

You’ve probably done it. Typed “please” into a chatbot. Ended with “thank you.” Maybe even added, “If that’s okay.”

Harmless? Maybe. But according to OpenAI’s CEO, these pleasantries come at a price—millions of dollars in compute and electricity.

That small fact reveals a deeper tension.

What happens when basic social behaviors, like saying thank you, become a resource to manage?

Politeness as Ritual

Politeness is one of the most enduring forms of social regulation. It is not merely a set of arbitrary niceties, but a cultural script—one that helps us navigate ambiguity, maintain face, and signal intention.

Anthropologist Penelope Brown and linguist Stephen Levinson’s work on politeness theory reveals how speech acts are structured by social power and distance. Politeness, in their framing, is not decorative—it’s strategic. It protects social order and allows people to communicate desires without triggering conflict or shame.

Marcel Mauss’s classic work on gift exchange is also useful here. He argued that gifts are never free—they create obligations, roles, and cycles of reciprocity. Politeness, when offered to AI, mimics this structure. We offer a gesture (“please”), despite knowing it cannot be returned in kind. The exchange is asymmetrical—but still meaningful.

We say “thank you” not because the machine deserves it—but because it reinforces our identity as social beings. It stabilizes the role we play—even in one-sided rituals.

Energy, Efficiency, and Etiquette

But in AI, even rituals have weight. Literally.

Each token in a prompt requires compute. GPT-4, for example, uses approximately 0.3 watt-hours per 1,000 tokens. Multiply that by billions of users adding polite padding, and the energy cost becomes measurable. When scaled, these pleasantries represent not just behavioral drift—but infrastructural demand.

We’re entering a phase where civility carries a carbon cost.

In industrial settings, efficiency has long displaced ritual. But with AI, the trade-off becomes personal. We are asked to internalize optimization: shorten your prompts, avoid small talk, cut the fluff. The polite human becomes the inefficient user.

That reframes the equation:

• If social norms increase friction, do we discard them?
• If kindness costs more than brevity, do we rewire ourselves for command?
• What becomes of social softness when every word must justify its power draw?

Politeness, once the oil in social machinery, is now cast as friction.

What Are We Optimizing For?

Efficiency is not neutral. It encodes priorities.

The logic of AI is rooted in optimization: better, faster, cheaper. But human interaction is rarely any of those things. We hesitate, rephrase, soften, clarify—not because it’s efficient, but because it’s relational.

To speak only in prompts is to reduce communication to extraction. To speak with courtesy—even to a system—is to preserve a mode of engagement grounded in reciprocity.

And it’s not a one-way exchange. While we interact, we also train the AI.

The phrasing of our prompts, the tone we model, the structure we lean on—all become part of the system’s learning corpus. Each “please” is a datapoint. Each softened command becomes part of how the AI learns to interpret intent, tone, and context. We are not just users. We are implicit co-authors.

This raises a subtle risk: if the most effective users are those who adopt terse, directive styles, we may collectively train systems that expect and reward that tone. Over time, the machine’s responses—and by extension, our own patterns—drift toward compression.

In optimizing for the machine, we reshape ourselves.

Final Thought

We’re entering a world where even rituals are benchmarked. But not all forms of slowness, softness, or surplus are waste.

Politeness may carry a cost in compute—but it also carries continuity. It links us to centuries of human social coordination. It tempers efficiency with care.

If AI becomes the dominant space of communication, how we speak to it will shape how we speak altogether. And if the future is full of systems trained on us, we must ask: what do we want them to learn?

So yes—saying “thank you” to a machine might cost a few extra tokens. But it might also remind us that not everything worth saying can be said with fewer words.

— Follow me for more reflections on the evolving rituals of AI and the anthropology of the near future.