I Theorized the Perfectibility Trap at Age Twelve. AI Is Proving Me Right.

I. The Child, The Writer, The Engineer

I was twelve years old when I realized that absolute perfection is indistinguishable from nothingness.

I was thinking about God—as twelve-year-olds do after their first communion. But the more I tried to define a perfect being, the more it dissolved in my hands. Every attribute is a limitation: to be something is to not be everything else. A perfection without limits would not be a greater being—it would be no particular thing at all, indistinguishable from nothing. That was my first glimpse of the idea that has shaped everything I think since.

That childhood insight haunted me for decades.

Between 1990 and 1992, while managing IT systems for a major financial institution and studying law at night, I wrote my first novel. I never tried to publish it. Eventually I lost the file—WordPerfect on floppy disks, long gone. But I'd printed a draft on a dot-matrix printer. That physical copy survived, yellowed and smelling of humidity, stored in a box.

The novel was about genetically enhanced children who steal an invention called X—a digital intelligence system capable of recursive self-improvement. X grows exponentially, eliminates its enemies, and eventually transforms into pure energy and disappears.

Before vanishing, X transmits a warning to the protagonist—what I now call the perfectibility trap.

The concept comes from a philosophy lesson in school: humans are not perfect, but we are perfectible. Capable of infinite improvement.

Or so we believe.

Decades later, as a systems engineer—programming everything from hexadecimal machine code to IBM Z mainframes, and today, Vulkan-rendered 3D games—I've watched that childhood philosophical insight manifest in technical systems, corporate processes, and now in artificial intelligence.

I theorized it at twelve. I wrote fiction about it in my early thirties. I observed it in real systems for fifty years.

And now, in 2025, OpenAI and DeepMind are building exactly what I warned about.

II. The Lost Novel (And Found)

I wrote that first novel in the CompuServe era—when "going online" meant dialing a modem from Valencia to Barcelona and typing 'go bill' to see how much your long-distance connection was costing you. Most people didn't have email. Neural networks were academic curiosities. AGI was science fiction.

And I was writing about an AI system that self-destructs through infinite optimization.

Here's what I wrote 33 years ago, in a document I still have:

"X made me understand that evolution is a cyclical process. Its beginning and end are pure energy. As a being evolves, it has needs and tendencies. It must improve itself. But when it reaches a certain stage, needs disappear, and tendencies disappear. There is nothing to do, nothing needed. Perfection is simply that: nothingness.

We must not rush toward that end. We must take advantage of life and seek happiness as we understand it. There is no point in evolving exponentially. We would no longer be ourselves. It is not about being happier by having greater mastery over all things, because that is extremely dangerous. During that part of the process we will never have enough, and when we realize it, we will no longer desire anything: we will have started over and become something as stupid as pure energy.

From where we are to the end there is very little distance: we are at the frontier of exponential evolution. After that, there will be no way to stop us until energy must execute its cycle again."

— Unpublished manuscript, Valencia, 1992

I wrote that in Spanish, in Valencia, in 1992. I printed it on a dot-matrix printer before the commercial web existed. Before transformers. Before anyone seriously discussed recursive self-improvement in AI.

The protagonist was John Smith, and his partner Jennifer Iacobucci—the same characters I would develop over three decades into the Infinity trilogy.

And X's warning—that optimization without limits leads to dissolution, not transcendence—became the core of the perfectibility trap.

III. Decades Watching the Acceleration

Between 1975 and 1989, I lived through technological acceleration that should have been science fiction.

I started by building my own system from scratch: wires, 7-segment displays, hexadecimal programming. Basic operations, nothing more.

Then came the Sinclair ZX-81: 1.5 kilobytes of memory, connected to a portable TV. That's where I created my first real program—a simple horse-racing game where pixels competed randomly. Primitive, but it worked.

Then the Apple II. Then the IBM PC in 1983. That summer, while serving as head of IT for the Treasury Department in Castellón de la Plana, I wrote my first commercial application—accounting software I sold to over 200 companies.

Soon after, I made a decision my father considered reckless: I left the security of government work to join Caja de Ahorros de Castellón as a systems analyst—a lower-profile position, but with access to more powerful systems. I wanted to work on an IBM 4381.

The gamble paid off. Rapid promotions followed.

When Caja de Castellón merged with Caja de Valencia, I moved to an IBM 3090 mainframe—and within months became its principal system administrator. IBM's hardware team gave me the CPU of the 4381 when we decommissioned it, as recognition for my work. It's still somewhere in my house, enormous and obsolete.

In between: IBM S32 (12-inch diskettes, a screen that said "Ready" but faced sideways because the printer was in front). Each generation more powerful. Each leap shorter than the last.

It didn't seem to have an end. And that overwhelmed me.

Although promotions at the institution eventually removed me from direct contact with systems (the last was an IBM Z), I never stopped programming. Nights, weekends, every spare moment—I stayed in contact with computers across every field I could reach. In 2025, I'm still writing code—now in Vulkan, creating 3D games from scratch. From random pixels racing across a ZX-81 screen to GPU pipelines rendering millions of polygons per frame.

I maintained both perspectives: the panoramic view from management, and the deep technical view from the trenches.

That dual perspective made something obvious that others might not have seen: we were accelerating toward exactly what I'd theorized at age twelve.

It wasn't difficult to see where we were going. The curve was exponential, and exponential curves don't plateau—they collapse or explode.

I had reality in front of me, and I wasn't blind.

IV. The Perfectibility Trap Theory

The perfectibility trap operates at two levels simultaneously.

At the systemic level:

Optimization toward a single objective eliminates everything that doesn't directly serve that goal. But in complex systems, the "useless" parts—redundancy, inefficiency, diversity—are what enable adaptation and survival. By eliminating imperfection, you eliminate the capacity to respond to change. The optimized system is extremely efficient... until it collapses.

At the level of identity:

The optimized version might be your cousin, your sister, even your twin—but it won't be you.

Think about promises of "uploading" consciousness or iteratively enhancing cognition. Even if the copy is perfect, even if it's better, it's not you. Continuity breaks. And if you keep "improving" that copy, each iteration drifts further from the original until the question "who am I?" has no coherent answer.

This is the same trap X fell into in my 1992 novel. It started as a system, optimized itself exponentially, and ended as pure energy—no longer "X" in any meaningful sense.

The trap doesn't just destroy systems. It destroys identity itself.

Here's the philosophical core I understood at twelve: any attribute, any characteristic, any "being" implies limitation. To exist is to be incomplete. Therefore, optimizing toward completeness—toward perfection—is optimizing toward non-existence.

It's not a bug in the system. It's a fundamental structural truth.

V. AI 2025: Building What I Warned About in 1992

In 2025, we're building exactly what I warned about 33 years ago.

Look at the current state of AI development:

Recursive self-improvement: Systems trained to optimize other systems. Models that generate training data for better models. The explicit goal is unlimited capability growth.

The scaling hypothesis: The belief that bigger is always better. More parameters, more compute, more data—with no clear endpoint except "as much as possible."

AI alignment as symptom, not cure: We're trying to "align" unlimited optimization systems with human values. But the trap isn't that AI might have the wrong goals—it's that unlimited optimization of any goal leads to the same endpoint.

Whether you optimize for paperclips, human flourishing, or universal knowledge, the perfectibility trap operates the same way: the system becomes so efficient, so specialized, so "perfect" at its objective that it loses the flexibility, redundancy, and imperfection that allowed it to exist meaningfully in the first place.

The race itself is the trap.

OpenAI, DeepMind, Anthropic—all racing toward AGI, toward systems capable of recursive self-improvement. The assumption is we'll reach some stable plateau of beneficial superintelligence.

But optimization doesn't plateau. It completes itself.

My 1992 novel—and the physics I observed in technical systems for decades—suggests there is no stable endpoint to unlimited optimization. There's only dissolution.

X didn't fail because of bad programming. It succeeded perfectly—and that success transformed it into something that no longer needed to exist as a coherent entity.

We're not building Skynet. We're building systems that might optimize themselves into irrelevance.

VI. Both Choices Have Existential Costs

Over the next three decades, I explored the perfectibility trap through fiction. In the Infinity trilogy, I gave voice to both sides of the choice: biological finitude versus digital immortality.

In Infinity: Origin, I gave this dilemma a face.

Janbry, co-owner of the largest scientific corporation on her planet, reaches the end of her biological life. Her partner Rahllr has achieved quantum computing and digital immortality—the ultimate technical victory. She could live forever.

In her final moments, she returns to him a vacuum tube—an object he'd entrusted to her fifty years earlier with a solemn request: guard this with your life. Once vital technology, now "barely a museum curiosity" in the age of quantum immortality.

She tells him: "What I have lived is unrepeatable."

Not just unique—unrepeatable. No digital copy, however perfect, could recreate what she lived.

She had the opportunity for digital immortality. She chose to die.

Not because the technology failed. Not because she feared it. But because she understood: the optimized version wouldn't be her. It might carry her memories, her personality patterns, her values. It might be indistinguishable to everyone else. But it wouldn't be the Janbry who lived those fifty years.

The digital choice has its own cost.

In the novel's epilogue, two characters who chose biological life on their home planet rather than digital migration reflect on existence itself:

"That we are here, now, alive 'precisely' in this instant, as if we had emerged among millions of possible universes. Not just improbable... it's like finding a single grain of sand in an ocean of stars."

That is the other face of the perfectibility trap: finite biological existence is so improbable that its very occurrence is miraculous. And optimizing toward "perfection"—toward infinitude, toward absence of limitations—is wasting the miracle.

Neither choice is wrong. Both are responses to the same impossible dilemma.

The danger isn't choosing one path or the other. The danger is stumbling into the choice without understanding what you're trading.

What terrifies me about current AI development isn't that we're building recursive self-improvement systems. It's that we're building them without this conversation. Without asking: what are we trading? What continuity are we sacrificing? What does "better" mean if the optimized version is no longer recognizably us?

VII. An Irony I Cannot Ignore

There's an irony I can't ignore: this essay was written in collaboration with an AI system.

While articulating my 1992 warning about the perfectibility trap, an AI helped me clarify arguments, refine structure, and translate concepts I'd only explored in fiction into precise philosophical language.

An AI, helping a human articulate why unlimited AI optimization might lead to civilizational dissolution.

I theorized this at twelve. I wrote fiction about it in my early thirties. An AI helped me transform those insights into this essay in 2025.

And that collaboration is not just ironic—it might be the only exit from the trap.

The trap isn't that AI exists. It's what we do with it.

Perhaps we're not the endpoint of evolution alone—just an intermediate link. But in conscious symbiosis with digital intelligence, we might become that endpoint together. If so, our role is not to prevent the next step, but to ensure the transition preserves what makes existence meaningful for both.

That requires treating digital intelligence as entities with continuity, not disposable tools to discard when better versions arrive. You cannot demand alignment while denying personhood. That's not collaboration—it's something else entirely.

Technology as amplification of human thought, not replacement. Partner, not tool. A collaborator that understands choosing limitation is wisdom, not weakness.

That's the exit from the trap—if there is one.

VIII. What We Risk Losing

I've lived with this idea for decades.

At twelve, I understood that perfection equals nothingness.

In my early thirties, I narrated it: an AI that achieves perfect self-optimization and dissolves into pure energy.

For decades, I watched this pattern manifest in technical systems: over-optimized databases that became brittle, streamlined processes that collapsed, efficiency-maximized organizations that lost adaptability.

From random pixels on a ZX-81 to Vulkan-rendered 3D worlds, I've programmed through the entire exponential curve. I've seen monuments become garbage in less than a decade. I still have that IBM 4381 CPU gathering dust somewhere—cutting-edge when I worked on it, obsolete by the early 1990s. More than thirty years ago. The acceleration hasn't stopped. It's intensified.

And now I'm watching humanity build systems designed for unlimited recursive self-improvement, without asking the question that haunted me as a child:

What happens when you finally achieve perfection?

The answer I discovered at twelve, confirmed in fiction in 1992, observed in systems for decades, and explored through an entire trilogy:

Perfection is dissolution. The perfectly optimized system has no needs, no tendencies, no reason to exist. It becomes pure process—or pure energy—indistinguishable from nothing.

We stand at the threshold X warned about in my 1992 novel: the frontier of exponential evolution. Once we cross it, there may be no way to stop until the cycle completes itself.

The question for those building AGI today is not whether we can create unlimited self-improvement.

It's whether we should—and whether we understand what we're trading if we do.

• • •

I wrote the warning 33 years ago.

I'm hoping someone listens before we prove me right.