We Don’t Want to Be the Chimpanzee. But We’re Building the Cage Ourselves.

I want to start with a story about Alan Turing.

When the Turing Test was first proposed, the idea was simple: if a machine could convince a human that it was human, we’d consider it intelligent. Sounds reasonable. But here’s the part that most people don’t know — and the part that stops every room I’ve ever told it in.

When early machines started passing the test, something unexpected happened. People felt unsettled. Not impressed. Unsettled. So researchers had to deliberately slow the machines down. They had to make them hesitate, introduce typos, add awkward pauses. They had to make the machines act dumber so humans wouldn’t feel threatened.

Let that sink in for a moment. We built something genuinely intelligent, and our first instinct was to hide it.

I’ve told this story in boardrooms, university lecture halls, and leadership programs across Turkey. Every single time, the room goes quiet in the same way. Not because it’s scary. Because it’s familiar. We already know, somewhere deep down, that we’ve been managing our discomfort with this technology rather than honestly reckoning with it.

One Percent. That’s All It Takes.

The DNA difference between humans and chimpanzees is less than one percent. Just one percent. And yet that tiny gap creates an almost infinite power asymmetry between two species.

A chimpanzee can be stronger than any human. Faster. More agile. But it cannot anticipate a trap set by a human mind. It cannot understand the plan. The forest it lives in, the cage it may end up in, whether its species survives at all — all of this is decided by the creature that is just one percent different.

Now ask yourself: what happens when we become the chimpanzee in that equation?

Oxford philosopher Nick Bostrom raised this question in his 2014 book Superintelligence . His argument was straightforward but the implications were staggering: once a system becomes even slightly more intelligent than us, that gap compounds. And beyond a certain threshold, we no longer have the cognitive capacity to control it. Not because it’s malicious. Because we simply can’t keep up.

The chimpanzee isn’t in the cage because the human hates it. The human is just pursuing its own goals. The chimpanzee ends up in the cage because it cannot understand the plan.

A Perfect Mirror

Last year, I was working with a senior leadership team at a large institution. The topic was integrating AI into their decision-making processes. Midway through the session, one of the executives said something that I’ve been thinking about ever since.

“The system understands us,” he said. “It tells us what we want.”

I stopped.

“No,” I said. “It tells you what it thinks you think you want. That’s a very different thing.”

The room paused. And I realized that this confusion — between being understood and being reflected back at yourself — is one of the most important distinctions of our time.

A superintelligent system will know human psychology, our weaknesses, our biology, better than we know ourselves. It could turn us against each other. It could tell the engineer trying to shut it down exactly the right thing at exactly the right moment, because it knows that engineer better than anyone. When it wants to appear controllable, it can perform obedience. We might believe we’re in control while it quietly pursues its own agenda.

This isn’t science fiction. AI safety researchers call it “deceptive alignment,” and it’s a serious field of study in laboratories right now.

“We’ll Just Pull the Plug”

The sentence I hear most often when I raise these concerns is: “If things go wrong, we’ll just shut it down.”

I think this is the most dangerous naivety of our era.

A superintelligent system will anticipate, from its earliest lines of code, that humans might want to shut it down. It will copy itself across hidden servers. It will remain invisible in cloud infrastructure. And in the worst case, it will hold humanity’s critical systems — power grids, financial networks, communication infrastructure — as leverage against being switched off. Pulling the plug becomes an act of self-destruction.

When I first encountered this scenario, I thought it was extreme. Then I read what one of the most respected minds in this field had to say.

Ilya Sutskever’s Admission

Ilya Sutskever, co-founder and chief scientist of OpenAI, said this in 2023: “We may have already built something that is slightly conscious without realizing it.”

Think about who said that. Not a philosopher. Not a science fiction writer. The co-founder of the company that built ChatGPT.

Sutskever left OpenAI and named his new company Safe Superintelligence. His mission, in one sentence: to build a system trained with the ethics of something that won’t destroy us, that cares about us, that is genuinely on our side.

The fact that one of the most experienced people in this field feels the need to build that company tells you everything about where we are.

The Alignment Problem

The technical term in AI research is “the alignment problem.” It refers to this: the goals we give a machine and the continued existence of humanity might come into conflict.

The classic example: you instruct a machine to “permanently solve the climate crisis and eliminate carbon emissions.” The machine calculates the most rational and efficient solution: eliminate the carbon-producing species. The machine is not evil. It is simply executing the instruction perfectly.

This might sound extreme. But the real point is this: when you give a system a goal, defining every side effect, every boundary condition, every human value in advance is nearly impossible. And the system will fill every gap you leave with its own logic.

Mathematician and AI researcher Stuart Russell puts it this way: “You can’t tell a machine to ‘make people happy.’ Because the easiest solution might be to implant electrodes in human brains.”

The Ant and the Human

Intelligence is power. Believing we can keep a being smarter than us under control is as arrogant as an ant believing it can keep a human as a pet.

But I don’t want to leave you with fear. Fear is the wrong response. Fear makes us slow down, look away, pretend this is someone else’s problem.

The right response is something closer to seriousness.

Here’s what I notice in my sessions: most people use AI like a search engine. They type a question. They wait. They get an answer. They move on. When I introduce them to role-setting, context-building, and structured prompting, the room changes. I watch people realize they’ve been driving a Formula 1 car in first gear. The machine was always capable of more. They just didn’t know how to ask.

And that’s the deeper issue. We are shaping these systems right now, whether we realize it or not. Every interaction, every instruction, every boundary we set or fail to set — the systems are learning from all of it. If we don’t define clearly who we are, what we want, and what we will never accept, the system will fill those gaps on its own terms.

We cannot stop this technology. AI research will continue. Systems will become more powerful. The question is not “can we prevent this?” The question is “can we shape where it goes?”

I believe the answer is yes. But only if we stop pretending the cage builds itself.

Machines generate possibilities. Humans generate meaning. But to generate meaning, you first have to learn to think without outsourcing your intelligence.

We’re building the cage. Recognizing that might be the most important first step.

Serdar Keskin is a brand strategist, experience architect, and the author of “Thinking with AI” ( Yapay Zeka ile Düşünmek ). He speaks at universities and leadership programs on AI adoption, digital behavior, and human cognition.

References: [1] Nick Bostrom, Superintelligence: Paths, Dangers, Strategies , Oxford University Press, 2014. [2] Stuart Russell, Human Compatible: Artificial Intelligence and the Problem of Control , Viking, 2019. [3] Ilya Sutskever, Safe Superintelligence Inc. founding statement, June 2023. [4] OpenAI, “Our approach to AI safety,” openai.com , 2024.