AI, Physics, and the Myth of Limitless Intelligence

Over the past few years, conversations about artificial intelligence have increasingly drifted toward dramatic narratives: runaway superintelligence, machines replacing humans, or AI systems eventually controlling civilization. These ideas make for compelling science fiction. But when examined through the lenses of physics, engineering, economics, and technological history, the picture looks far more grounded—and arguably more interesting.


This post reflects a long discussion exploring a simple question: What are the real limits of AI?


AI systems rely on enormous computational infrastructure. The training of large models requires vast clusters of GPUs, huge amounts of electricity, and specialized semiconductor manufacturing. All of this sits on top of an industrial ecosystem involving chip fabrications, supply chains, cooling systems, and energy grids.

Unlike software alone, this infrastructure cannot scale infinitely. Semiconductor progress—long driven by Moore’s Law—is already slowing as transistor sizes approach atomic-scale limits. Even if engineers squeeze out a few more generations of improvement, the exponential growth that fueled computing for decades is approaching physical constraints.

AI cannot escape this reality. Intelligence implemented as computation must obey the same laws governing all physical systems: thermodynamics, energy conservation, and information limits.

It is tempting to imagine that intelligence, once sufficiently advanced, could transcend its physical substrate. But intelligence is ultimately a form of computation, and computation requires energy, memory, and time.

Fundamental limits such as the Landauer principle (minimum energy required to erase information) and the Bekenstein bound (maximum information density within a physical system) remind us that every intelligent process—human or artificial—is bound by the laws of nature.

Even a hypothetical superintelligence would still be constrained by:
- finite energy
- finite computational speed
- finite memory
- communication delays
Nature does not permit infinite leaps.

History shows that technologies rarely grow indefinitely. Instead, they follow a recognizable pattern:

1. Discovery phase – rapid breakthroughs and excitement  
2. Expansion phase – massive investment and rapid scaling  
3. Infrastructure phase – stabilization and incremental improvements

Electricity, automobiles, and the internet all followed this trajectory.

Smartphones are a recent example. Between 2007 and about 2015, each new generation brought dramatic advances. Over the last several years, improvements have become incremental: slightly better cameras, faster processors, longer battery life. The core concept has stabilized.
AI may be following a similar path.

Today’s AI industry resembles the infrastructure build-out phase of earlier technological revolutions. Companies are investing billions in data centers, GPUs, and cloud infrastructure. Many AI startups operate at heavy losses while investors fund rapid expansion.

This pattern echoes previous technology bubbles, such as the dot-com boom of the late 1990s. In those cases, massive infrastructure was built before the economic value was fully proven.
Eventually, investors begin asking a simple question:
Where is the return on capital?
When expectations outpace reality, markets correct themselves. Historically, this correction often produces a crash followed by consolidation and more sustainable growth.

AI systems consume significant electricity. Large training runs can require energy comparable to that used by small towns. Data centers compete for power with homes, hospitals, transportation systems, and industry.

Civilizations prioritize essential services. Societies are unlikely to shut off schools or hospitals simply to power AI clusters.
As a result, the growth of AI is tied directly to the expansion of energy infrastructure—a process that typically unfolds over decades.

Much of the public anxiety surrounding AI comes from treating it as a potential rival intelligence. But a more accurate perspective may be simpler: AI is a tool.
Just as engines amplified physical strength during the industrial revolution, AI amplifies cognitive tasks:
- analyzing data  
- generating text  
- assisting software development  
- optimizing complex systems

Younger generations already seem to view it this way. To them, AI is simply another utility—like search engines or calculators.
That perspective may ultimately prove the most stable one.

None of this means AI will be insignificant. Even within physical and economic constraints, AI could become enormously powerful. But instead of an uncontrolled intelligence explosion, a more plausible trajectory might look like this:

AI capability grows rapidly →  hardware and energy constraints appear →  economic pressure forces consolidation →  AI stabilizes as a foundational infrastructure.

At that stage, AI may resemble technologies like electricity or the internet: deeply embedded in society, extremely useful, but no longer surrounded by constant hype.

The idea of limitless intelligence makes for compelling stories. But nature tends to move through continuous processes rather than dramatic leaps.

Every powerful technology eventually encounters the same reality: the laws of physics, the limits of resources, and the constraints of human civilization. AI will likely be no exception and that may be perfectly fine.

The arguments presented here reflect principles from physics, engineering, and historical technology cycles. However, the precise trajectory of AI development remains uncertain, and different technological breakthroughs could shift timelines in unpredictable ways. This is just a thought experiment that led to this perspective. 

Popular posts from this blog

Try to Behave Ideally: A Personal Journey

  Try to Behave Ideally: A Personal Journey Back in 2003, a single piece of advice from my professor lodged itself in my memory. He said, “Try to behave ideally in every situation, encounter, and challenge that you face in your life.” At the time, I balked. “If I lived like that,” I argued, “I’d be a monk—sacrificing the full spectrum of human emotion.” He smiled and pointed out the power word: try. No one can be ideal all the time, he said, but aspiring toward that goal is its own path. I agreed, but only skimmed the surface of what “ideal” might truly mean. Wrestling with “Ideal” Lately, that old admonition has been surfacing more and more. I find myself asking: What is “ideal,” anyway? Is it fixed or does it change from person to person? Should I rigidly follow my own belief system, or defer to someone else’s? In a societal context, what guides me when ideals collide—my ideal, theirs, or the common good? These questions tumbled around in my mind, demanding deeper ...
Read more

Morality, Power, and Choice: A Systems View

1. The Birth and Function of Morality Almost all evolutionary theories and philosophical traditions converge on one foundational principle: survival. In an environment filled with external threats—predators, climate instability, scarcity, and competing groups—individual survival alone is fragile. Under such conditions, cooperation emerges not as an ideal but as a strategy. Four coordinated hands are stronger than two isolated ones. This marks the origin of morality—not as virtue, but as a behavioral constraint that stabilizes cooperation. Morality, in this sense, is not uniquely human. Wolves hunt in packs, chimpanzees organize in clans, and elephants operate as extended families. What humans did differently was not invent morality, but externalize it—by naming it, teaching it, remembering it, and enforcing it consciously. Morality therefore functions as an early coordination technology. It reduces internal friction, increases predictability, and enables collective response to external...
Read more