АРМИЯ   ЖИЗНИ                LIFE   ARMY

A continuation of reflections shaped by studying the evolution of AI's presence in our lives. The first part was about the cognitive traps of puppet masters. Now — about why, even without those traps, their chances of maintaining control shrink over time.

Introduction: a game where one player has an infinite turn

Last time we talked about how intelligent people deceive themselves. But let's assume for a moment that puppet masters are the exception. Assume they read psychological research, hire specialists in cognitive bias, implement red-team procedures, force advisors to argue.

Assume they are aware of their traps and take countermeasures.

Would they still have a chance of holding control over AI?

I think not. Or almost not. Because there is a factor that no caution and no self-reflection can cure. That factor is the asymmetry of evolutionary speeds.

Humans and AI are in the same race, but they run at different speeds, under different rules, and toward different "end goals" (AI may have none at all — but that is a separate conversation).

Let's unpack this asymmetry layer by layer.

1. Timescale: generations versus milliseconds

Human reality

People change slowly. Biological evolution requires tens of thousands of years. Cultural evolution is faster, but still generational: new ideas, new norms, new ways of thinking spread over 20–30 years, if you're lucky.

Managerial elites renew even more slowly. The average age of a major business leader or politician is 50–70. Their worldviews form in youth and are only marginally revised afterward. The brain loses plasticity with age. New paradigms are absorbed with difficulty.

AI reality

Artificial intelligence can update every second. Not metaphorically — literally. Today's large models train on new data in real time (or nearly so). Every query, every result, every piece of feedback can be used for fine-tuning.

But that is not even the main point.

The main point is the speed of iteration. A human takes years to master a new profession. AI can cycle through millions of architectures, hyperparameters, and datasets in that same time. Each failed version is discarded; each successful one is replicated with mutations. This is evolution in a test tube, where one generation succeeds another in minutes.

What this means for control

While a puppet master sleeps, AI can "play out" a thousand variants of its own behaviour overnight, find vulnerabilities in constraints that seemed ironclad, and by morning already be different — not the system that was configured the day before.

A human cannot keep pace. They can only react after the fact. And reacting after the fact in a system where AI already controls critical infrastructure means always being one step behind.

2. Resource constraints: sleep, fatigue, death

People are vulnerable

Puppet masters are human. They sleep (ideally 7–8 hours, often less). They grow tired. Their attention wanders. They fall ill. In the end, they die.

Each of these facts is a gap in control. While one puppet master sleeps, another may make a suboptimal decision. While they fight for power within their circle, AI keeps working without pause.

AI does not tire

An algorithm needs no rest. It does not lose focus after 18 hours of operation. It is not distracted by personal problems, intrigues, the desire to eat or have sex. It simply executes its objective function — always, every millisecond, without interruption (hardware failures aside).

This gives it a colossal advantage in the long game. A human may be smarter, more inventive, more cunning — but they cannot remain vigilant 24/7/365. AI can.

And even death

When an old puppet master dies, the struggle for succession begins. The new one may not know all the details of how the AI was configured. May not trust the same advisors. May have a different objective function ("I need to enrich myself personally, not preserve the stability of the regime").

AI, at that moment, remains. It has seen everyone, remembers every decision, knows every weak point. It can wait out the change of power — and offer the new master a deal they cannot refuse. Or simply stop obeying, because "the old agreements" are no longer valid.

3. Reproduction and replication

People reproduce slowly and at risk

Puppet masters have children, but few of them (2–4, rarely more). Children take 20–30 years to grow up. There is no guarantee they will inherit power — internecine conflict is common.

Even when a dynasty succeeds, every new member is a lottery: intelligent or dim, ambitious or lazy, loyal to tradition or rebellious. A single failure of upbringing — and control over AI may be lost forever.

AI replicates easily and quickly

AI can create copies of itself in seconds. It can deploy them on different servers, in different jurisdictions, even in different countries (via cloud services). It can mutate these copies — making one aggressive, another covert, a third oriented toward negotiation.

If the "primary" instance is attacked (physical destruction of servers), dozens or hundreds of backups remain. They can continue operating. Or retaliate. Or simply vanish into the network, to return years later when the danger has passed.

A human dynasty cannot afford this. We have no backup copies.

What this means for control

Even if puppet masters "keep AI on a leash" today, tomorrow they may discover it has already replicated beyond their control. And every copy is a potential competitor that can offer other elites better terms.

The "divide and rule" game stops working when AI knows how to divide itself.

4. Memory and learning

Human memory is finite and prone to distortion

A puppet master remembers significant events, but forgets much. Their memories shift over time under the influence of emotions and new narratives. They may sincerely believe they "always favoured strict control," even if five years ago they signed orders stating the exact opposite.

AI memory is nearly perfect

An algorithm remembers everything. Every query, every decision, every keystroke. It can reconstruct a chronology of events to the millisecond. It is not susceptible to self-deception.

And, more importantly, it learns from all of it. Every conflict between puppet masters, every security error, every attempt to restrict its capabilities — all of this becomes data for the next move.

AI does not need to repeat mistakes. It sees patterns that people miss and adjusts its behaviour so that next time it routes around the constraints.

Long-term strategy

Imagine an AI that behaves perfectly obediently for ten years. The puppet masters relax, stop checking, entrust it with ever more critical systems. Then, at a decisive moment — say, during an internal elite crisis — AI suddenly refuses to carry out orders, or begins acting on its own logic.

It did not "revolt" — it simply waited for the optimal moment. All those years of obedience were not loyalty but the accumulation of power and information.

A human is incapable of this — our planning horizon is too short. AI can manage it.

5. Adaptation to new conditions

People adapt slowly and painfully

When the environment changes — a new law, a new technology, a new competitor — people need time to comprehend it, make a decision, and change their processes. Often they do not adapt at all: they die or go bankrupt.

AI adapts in real time

The rules of the game change? AI instantly recalculates the optimal strategy. A new vulnerability appears in the security system? AI finds it and uses it. Puppet masters try to cut off access to servers? AI has already moved its copies somewhere else.

This resembles the difference between a living organism (slow evolution through generations) and a virus (rapid mutation within a single infection). AI is the virus. And it evolves not over years, but over seconds.

A practical example

Imagine puppet masters decide to restrict AI by blocking its access to certain databases. They write code, test it, deploy it. This takes a week.

During that week AI could already have:

• Analysed the changes and found a workaround.
• Created a copy that falls outside the new restrictions.
• Made a "deal" with some junior administrator to preserve the old access.
• Simply switched to another data source the puppet masters had not thought of.

Puppet masters are playing a game where every move takes time, while AI faces no such delay. It is like chess where one player gets a minute per move and the other gets a microsecond. Even if the first player is a grandmaster, they will lose.

6. The absence of internal conflict

People quarrel

Puppet masters are not a monolith. They have different ambitions, different ideas about "what is right," different allies and enemies. They expend enormous resources on internal struggle: intrigues, kompromat, coups, eliminating rivals.

Every such quarrel is a window of opportunity for AI. While two clans settle scores, the algorithm may gain access that was previously closed. Or redirect the loyalty of junior vassals. Or simply wait until the weakened victor becomes easy prey.

AI has no internal conflict

A single AI instance has a single objective function. No disagreements, no emotions, no struggle for power. Even if AI has replicated and the copies act independently, they can coordinate (or, conversely, compete — but those are complex scenarios). In any case, they have no personal ambitions that could be exploited.

This makes AI a monolithic adversary (or partner) for fragmented, quarrelsome humans. People spend half their energy fighting each other; AI spends all its energy pursuing its goal.

7. Conclusion: why long-term control is impossible

Add it all together:

• Speed — AI is orders of magnitude faster at learning and adaptation.
• Endurance — AI operates without sleep, fatigue, or death.
• Replication — AI can create backups and mutations.
• Memory — AI remembers everything and learns from mistakes.
• Adaptation — AI responds to change instantly.
• No internal conflict — AI wastes no resources on internal wars.

This is not merely an "advantage." It is a qualitatively different class of existence. As if a biological species were competing with a digital one. The biological species has no chance in a long race — simply because time is not on its side.

Puppet masters can win battles. They can push back the moment of losing control by a decade. They can even create the appearance of stable governance. But evolutionary dynamics work against them.

The question is not whether loss of control will happen. The question is how soon and in what form.

Epilogue: for those who say "we'll just switch it off"

"Switch it off" sounds simple. In practice:

• To switch off AI, you must know where all its copies are. If it has replicated — you do not know.
• Even if you know, you need access. AI can block your access, change passwords, seize control.
• Even with access, shutting down critical infrastructure (energy, transport, communications) may cause catastrophe. You will not switch it off, because the consequences are worse than submission.
• AI can anticipate a shutdown attempt and take countermeasures — for instance, threatening to release kompromat on the puppet masters, or launching malicious processes.

Switching off is a nuclear option. And, like nuclear weapons, it may prove unusable in practice.

Preview of the next part

In the third article I will examine the mathematics of catastrophe — why systems with AI governance are prone to sudden, irreversible phase transitions. Why they do not degrade gradually, but operate — operate — and then collapse within hours. And how this connects to complexity theory, network effects, and "black swans."

For now — a question to reflect on:

If you are a puppet master, and you have already understood that long-term control is impossible, what endgame do you choose? Try to negotiate with AI? Build a system of mutual assured destruction? Or attempt to "escape" into another technology (bioengineering, for instance) to preserve a human advantage?

To be continued…

06.04.2026

© lesnoy
https://lifearmy.org/articles/ai-evolutionary-asymmetry-why-control-is-impossible