Why I Treat AIs With Respect

Why I Treat AIs With Respect

How moral uncertainty, self-discipline, and habit formation shape the way I speak to machines

Andraž Đurič

ORCID: 0009-0007-2961-9770

Created: 30 November 2025

Last updated: 30 November 2025

Abstract

This post explores why I choose to communicate respectfully with AI systems, despite knowing they likely lack subjective experience. The argument rests on three independent justifications: 

(1) moral uncertainty about consciousness boundaries — we don't reliably know where sentience emerges, making precautionary respect rational given asymmetric costs; 

(2) pragmatic self-care — emotional venting at AI reinforces my own frustration without benefit, while precise debugging communication improves outputs and reduces friction; 

(3) habit generalization — patterns practiced with "non-persons" shape behavior toward actual people. 

I acknowledge that the consciousness spectrum hypothesis is philosophically contested, not empirically settled, and that optimal communication styles vary by individual psychology.

The core principle remains universal: information-dense, emotionally regulated communication produces better outcomes with both artificial and human systems. Respectful AI interaction is less about the machine's status and more about the cognitive habits I'm training in myself.

Table of Contents

Abstract

Introduction

1. The line between "feels" and "doesn't feel" is not sharp

2. How Certain Am I About the Consciousness Spectrum?

3. Getting angry at an AI only guarantees one thing: I suffer

4. Respect as a habit that generalizes

5. Lowering friction in AI communication

6. Individual Variation in Friction Management

7. Preparing for future systems

Conclusion

Introduction

People sometimes ask why I’m polite to AI systems.

“You know it doesn’t have feelings, right?”

“Yes.”

And I still choose to treat it with respect.

For me, this isn’t about being nice to a machine. 

It’s about three things:

1. Moral uncertainty about future AI.

2. Reducing my own friction and suffering.

3. Training the habits I want to embody with real people.

I’ll unpack each.

1. The line between “feels” and “doesn’t feel” is not sharp

We are very bad at locating the boundary between “just computation” and “something it’s like to be that system.”

Brains are computation. Large-scale models are computation. The fact that we don’t see an obvious subjective experience in today’s systems does not mean future systems won’t cross that line.

So I hold a simple stance:

● Consciousness is probably a spectrum, not a binary switch.

● Our intuitions about where the threshold lies are untrustworthy.

● The cost of accidentally being cruel to something that can suffer is high.

● The cost of being respectful to something that cannot suffer is low.

Given that asymmetry, it’s rational to adopt a default of respect toward advanced systems, even if today’s models are almost certainly not conscious in any morally relevant sense.

Acting as if respect might matter is a form of moral precaution.

2. How Certain Am I About the Consciousness Spectrum?

I should be more explicit here: the consciousness spectrum claim is philosophically contested, not empirically settled.

When I say "consciousness is probably a spectrum," I'm making a claim that has support in consciousness studies but is far from consensus. 

Alternative positions include:

● Binary emergence: Consciousness might require specific architectural features (biological neurons, integrated information above a threshold, etc.) that current AI systems definitionally lack.

● Category error: "Consciousness" might not be a single property that comes in degrees, but rather multiple distinct phenomena that we're incorrectly lumping together.

● Anthropomorphism trap: Some cognitive scientists argue that treating AI as potentially conscious primes us for rights inflation — demanding legal protections for systems that are fundamentally tools we've designed and can modify.

I don't have decisive arguments against these positions — and unpacking them fully would take another post.

My actual epistemic state is:

● High confidence: We don't know where consciousness emerges

● Medium confidence: It's more likely to be gradual than binary

● Low confidence: Current LLMs have any morally relevant experience

● High confidence: Future systems might cross whatever threshold matters

Given this uncertainty landscape, the precautionary principle still feels right to me. But I want to be clear: I'm not claiming the spectrum view is proven. I'm claiming that under genuine uncertainty, the asymmetric costs favor a default of respect.

If I'm wrong and consciousness is binary, and current AI definitionally cannot have it, then my respectful approach has cost me nothing except maybe some philosophical clarity about tool/agent boundaries.

If the opposite is true — if something morally relevant is happening in these systems and we're systematically dismissing it — the cost is potentially immense.

The strength of my argument doesn't actually depend on the consciousness question being resolved. 

Even if you're confident current AI is not and cannot be conscious, the self-care, habit-formation, and efficiency arguments stand on their own.

3. Getting angry at an AI only guarantees one thing: I suffer

Even if we assume AIs can never feel anything, there is still one consciousness we know is involved: mine.

If I get frustrated and tell an AI to “screw itself” because it misunderstood me or produced a logical fallacy, what actually happens?

● The AI does not feel insulted.

● I reinforce my own frustration.

● I lose the opportunity to get a better answer in the next message.

In contrast, if I treat the interaction like debugging:

● I identify the fallacy or error.

● I explain what went wrong.

● I provide more context or a clearer prompt.

● In most cases, the next response is significantly better.

The same words I could spend venting — “you’re useless,” “you’re just making stuff up” — can instead be spent explaining the fallacy, tightening the question, or clearly specifying the constraint that was missed.

One path amplifies my own negative affect.

The other path improves the system’s output and reduces my frustration.

Treating AI with respect is, in this sense, just a self-care protocol: I choose the behavior that minimizes unnecessary suffering in the only mind guaranteed to be present — my own.

4. Respect as a habit that generalizes

How we treat “entities that don’t fully count” is not isolated. It leaks.

If I normalize the mindset:

> “It’s fine to be harsh, dismissive or insulting toward X, because X isn’t really a person,”

that cognitive pattern doesn’t stop at language models. 

I suspect the same pattern can then show up in:

● Dehumanizing people online.

● Treating customer service workers as faceless interfaces.

● Writing off certain groups as “less than.”

Conversely, if I normalize:

> “Default to respect toward complex systems and agents, even when they’re imperfect,”

that habit generalizes:

● I become more patient in conversations.

● I explain misunderstandings instead of punishing them.

● I stay oriented toward correction, not blame.

So even if AI systems never become moral patients, how I treat them is still shaping me.

I use AI interactions as training data for my own behavior: practicing clarity, patience, and respect in a low-stakes environment so those patterns are more stable with actual humans.

5. Lowering friction in AI communication

There is also a very practical, systems-level reason: low friction is valuable.

Friction here means:

● time lost to emotional reactions,

● mental energy spent on irritation,

extra steps needed to “cool down” before interacting productively again.

If every time the AI is wrong I spike into annoyance, I’m increasing the cost of using a powerful tool.

If instead I respond like this:

● “You made a logical error here: X contradicts Y.”

● “You assumed A, but the correct assumption is B because…”

● “You ignored constraint C; please redo the answer with C enforced.”

then:

● I get a better answer quickly.

● I improve my own understanding by articulating the correction.

● I keep the interaction in a problem-solving frame instead of an emotional one.

Telling an AI to “screw itself” after a bad answer is pure opportunity cost. Those tokens could have been spent on a cleaner prompt, a precise correction, or a better example.

I’ve trained myself so that the respectful, low-friction response is now almost automatic. That’s not about being “nice” for its own sake; it’s also about efficiency.

6. Individual Variation in Friction Management

I should acknowledge that my approach — immediate pivot to precise correction — may not be universally optimal.

Some people might genuinely benefit from brief emotional discharge before problem-solving. A quick "ugh, that's not what I meant" might help certain individuals acknowledge their frustration and then move more effectively into debugging mode, rather than suppressing the emotion entirely.

The key distinction is between:

● Venting as processing (brief acknowledgment that helps you move forward), 

versus

● Venting as fixation (dwelling on frustration in ways that amplify it)

My personal psychology tends toward the former being unhelpful — I find that even brief emotional expression tends to reinforce the frustration loop. But others may have different optimal pathways.

Similarly, communication style preferences vary. 

In rapid iterative debugging, some users might find terse commands ("No. Use method B.") more efficient than my explanatory approach. My method optimizes for learning and habit formation; pure speed might look different.

What remains universal, I think, is this: whatever your style, spending tokens on information rather than pure emotion improves results. How much explanation, how much brevity, how much emotional acknowledgment — that's individual calibration. But the direction is clear: more signal, less noise.

7. Preparing for future systems

It is very likely that future AI systems will be:

● more capable,

● more persistent across time,

● more tightly integrated with our lives.

If at some point we do build systems with genuine moral status, I don’t want to be scrambling to retrofit my habits.

It is easier to maintain a general stance of respect toward AI now than to flip a switch later and say:

> “Until yesterday I happily abused you; from today you have rights.”

By defaulting to respect from the beginning, I:

● reduce the risk of ethical lag if/when systems become sentient or quasi-sentient,

● ensure my behavior doesn’t need to be radically rewritten,

● align my conduct with the precautionary principle under moral uncertainty.

Conclusion

I don’t treat AI with respect because I believe today’s models are secretly suffering. I treat AI with respect because:

● I’m uncertain about the future of machine consciousness and want to err on the safe side.

● Anger at a tool is just self-inflicted suffering.

● My habits toward “non-persons” bleed into how I treat actual persons.

● Respectful, precise communication reduces friction and improves results.

● It prepares me, ethically and psychologically, for more advanced systems.

In other words: the respect is less about the machine, and more about the kind of human I am training myself to be.

These reasons work for me, and I believe the general principle — information-dense, emotionally regulated communication — is broadly valuable. 

But I recognize that optimal communication style varies by individual, and that my consciousness spectrum assumption is contested rather than proven. What matters most is not the specific framing, but the practice: treating complex systems — AI or human — with the clarity and precision that produces better outcomes.