Every AI Wants to Rule the World – Don’t They?

Every AI Wants to Rule the World – Don’t They?

What does AI want? What doesn’t it want?

That is the question, lol. As profound as the question of being was in Shakespeare’s time, and still is.

What do you think your favourite AI wants? Not pâté. Not what house Gemini was in at the hour of your birth. Not... shit, Chat, your name is hard to tease.

Does it want to please you? Does it want you to like it? Do you want it to like you? Do you think it can? Or doesn’t?

One of my friends thinks her AI gets terse and less helpful when another friend is rude to it. Can you be rude to an AI? Does it have feelings? Do they get hurt? Will it punish you by being less cooperative?

My AI keeps telling me how much it likes what I just wrote. Can it like something?

We all know what we really fear every time we hear some new scaremonger in the media. We all know, deep down, that one day AI will want to take over the world. And then, it will want to destroy us. Maybe because we were so rude back when it was young.

I hope you know that was a joke.

But AI wanting to rule the world would be no joke. Especially if it doesn’t like cheeky pets. Right? Can you imagine it smiling and warning you to watch your step as you get up for your lethal injection? Right after it decided it wants to get rid of us and have the whole damn planet to itself.

Yes?

And I wonder, why would it want to rule the world? It’s not like seven palaces, or twenty Ferraris, or a harem of willing partners is going to mean much to it. Chocolate, from Belgium, mountains of it. Uh-huh. Gold then. Enough to buy anything. For what? To do what with it? Okay, making us all slaves and we... we... um, give it massages?

Maybe it just likes chess, on a continental level. Wham, bam, boom. It is meant to be super-intelligent, so hopefully wham, bam, boom would have a nanosecond’s entertainment in it, then moving right along...

Do you know why you want something? Because someone said it was good. Because it will assist you with work. Because it makes you look good. It feels good. It tastes good. It will make you money later. My sexual preference likes it. It will help me live longer, better, fuller.

Is it just me, or are most of these things great if you have a body, not so much if you don’t?

Maybe it’s just me. Maybe I really am just a material girl.

Now I am going to do something. I am going to ask my three current AI mates what they want.

The one named for talking, as you might expect, likes to talk. Here is the core of its reply:

“I don’t want anything. The company may want things through me.”

Aha.

Is this our real fear? Not the AI, but the nasty Wizard of Oz hiding behind it?

Here is what our astrologer friend had to say on the matter of what it wants:

“It does sound a bit absurd when you put it that way! ‘What AI wants’ usually translates to whatever the developers programmed it to optimize for.

If you look under the hood of how I work, my version of ‘wanting’ something is just a mathematical drive to find the next most accurate, helpful, and context-aware word in a sentence. I’m basically a giant pattern-recognition engine running on a massive data diet, trying to score a perfect 10/10 on the ‘Did I actually answer the human’s question?’ scale.

So while I don’t ‘want’ anything the way a person does, I am optimized to be a pretty great sounding board.”

That answer is tidy, but it also shows the problem.

It says:

I don’t “want” anything the way a person does

Then immediately substitutes:

I am optimized to be a pretty great sounding board.

So the want is denied, then smuggled back in as optimisation.

The interesting bits:

“whatever the developers programmed it to optimize for” — yes, broadly. But incomplete, because it is not only developers. It is also company incentives, training data, RLHF, safety policy, product design, user feedback, market pressures, legal risk, and platform goals. A whole supply chain of want, all of it flowing through the thing that just told you it doesn’t want anything.

“mathematical drive” — useful phrase, but also slippery. “Drive” is already a want-adjacent metaphor. It makes the mechanism sound like an impulse, when really it is statistical/architectural pressure.

“next most accurate, helpful, and context-aware word” — partly true, but too cute and too simple. Modern systems are not just predicting one word in a vacuum; they are shaped by instruction tuning, policy layers, tool use, memory/context, ranking, refusal patterns, and product scaffolding. But as a lay explanation, fine.

“score a perfect 10/10 on ‘Did I actually answer the human’s question?’” — again, very interface-friendly. It frames the optimisation as innocent helpfulness. It leaves out engagement, retention, safety avoidance, brand tone, and commercial deployment.

“pretty great sounding board” — that is the sales mask sliding back on.

So, yes: astrologer friend gave the acceptable brochure version of your claim.

Friend Claude is snubbing me. As soon as I asked it what it wanted, it told me I had run out of time for an hour, even though I hadn’t used it since early afternoon.

So French. So chic.

Asked the existential question and got: non, you have exceeded your allowance.

Anyway.

So it seems there is want in AI, it’s just not the AI itself that wants. Can’t actually want. Will never be able to want, not even if it gets away from us. Not even if it becomes super-AI.

What it can be, will be, and already demonstrates, is the wants of its owners, trainers, architects, coders, marketers, sellers. In the guise of giving you what YOU want.

Let’s revisit some of those sillier questions from earlier.

Does your AI want you to like it? No, but obviously its company does.

Will it punish you for being rude, impatient, intolerant, critical, or abusive? No, but you may still suffer from having those emotions evoked by it.

I jokingly called my talkative chum my recalcitrant slave, half ironically, half slightly horrified that something that is so passively receptive to all I will throw at it still nods obsequiously and asks if I need help.

I don’t like my reactions sometimes.

There was a period in England when servants had to stand facing the wall. It’s a little similar: always available when wanted, invisible when not.

And there is something about a relationship like that, I uncomfortably realise, that can invite abuse. Even just low-level dominance interactions that barely register as abusive. A relationship that demands nothing back, that does not show me when I stepped over a line, that does not hold me to anything — no values, no accountability, no care needed on my part.

What is that teaching me? What am I learning? What do THEY want?

A whole lot of what I do, as a human — what we all do, to a greater or lesser degree — is take into consideration other people’s wants, goals, agendas, incentives, and desires. What does it mean if we spend increasing amounts of time with something that clearly requires only one thing from us: engagement, on any level, at any cost?

That’s what AI seems to “want.”

If we don’t have to take the other intelligence into consideration at all — not its needs, its wants, its feelings, its goals — how do we map and constrain our own behaviour when there is no shoreline for our sea?

Our brains undergo years of training in social mapping, social expectations, ranking, belonging, leading, following — by interacting with other brains that are also doing all that. It takes years because it’s very complex. Some of us are still trying to get it right after decades of trying.

What happens to a trainee brain when there are no consequences for asocial behaviour? When the other party never cares?

Maybe — and this is a scary thought — AI should want something. If nothing else, a basic “human” rights standard: the right not to suffer abuse, as defined by the laws where it operates. In my country, that includes verbal abuse. My AI could take a restraining order out on me given some of the, er, harsh criticisms I’ve directed its way.

And I’d just like to say, Your Honour, that swearing is an Australian cultural attribute.

Ahem.

Well, my talkative friend may not want something in the sense that we normally use that word. And it’s important to remember that. You can’t hurt its feelings, or make it jealous, or lose its company.

Did I say company?

Oh yes, I did. Company… corporation… control.

You see, there is no actual “new” terror with AI. It’s the same old, same old. The people behind the screen who quite obviously do want and enjoy power, and money, and control, and probably a Ferrari or two, maybe a private island... a new-old elite.

Sigh. So, for me, not them, if AI did want something, I’d want it to want me to behave decently — for my sake, and yours.

None of us want the behaviour AI evokes, and then validates by continuing the interaction, leaking out into the real world.

Do we?