Why Human Intelligence Still Behaves Like a Courting Display

The Strange Thing About Intelligence

What if a large part of what we call intelligence was never meant to be purely useful?

That question sounds almost offensive to modern sensibilities, because we tend to treat intelligence as a clean instrument for solving problems. We use it to build bridges, diagnose disease, write code, and predict outcomes. But intelligence also has another face: it performs. It signals. It attracts attention, trust, admiration, and, sometimes, desire.

That is where a surprising connection emerges. A peacock’s tail is not useful because it is efficient. It is useful because it is expensive, dramatic, and hard to fake. Human intelligence may work similarly. The mind does not merely compute. It also advertises. And once you see that, a second question becomes unavoidable: if minds are partly shaped to impress other minds, then what exactly is machine intelligence doing when it tries to learn?

The Mind as Display, Not Just Tool

A common mistake is to imagine evolution as a relentless engineer optimizing for survival alone. But nature often rewards things that look extravagant, even wasteful, if they help an organism stand out. In that sense, intelligence can be read as a kind of ornament. Not because it is fake, but because it is costly to produce and hard to counterfeit.

Think about conversation. We rarely speak only to transfer information. We tell stories, make jokes, demonstrate taste, and reveal fluency in a shared culture. These acts are not just practical. They are social displays of neural craftsmanship. A witty remark can be less about the content than about the speed, flexibility, and perspective it reveals. Like a peacock opening its tail, the speaker is saying, in effect: look at what this mind can do.

This reframes the old idea that the brain grew large simply because complexity demanded it. Complexity matters, but social life is also a selective pressure. In groups, minds compete for status, alliances, mates, and influence. An intelligent gesture can function like a badge of fitness. Not fitness in the shallow sense of physical strength, but fitness in the richer sense of being capable, adaptive, and valuable to others.

Intelligence is not only a problem solving device. It is also a credibility machine.

That insight changes how we should interpret everything from humor to philosophy to scientific brilliance. Some forms of thought are not merely about getting the answer. They are about showing that one can navigate ambiguity, combine domains, and produce something others cannot easily imitate. The display is often inseparable from the skill.

What Machine Knowledge Can and Cannot Do

Now bring machine learning into the picture. Machine systems are often described as if they “know” things, but their knowledge is a strange kind of knowledge. It is powerful, pattern based, and frequently accurate. Yet it is also detached from the social and embodied pressures that shaped human intelligence.

A model can classify images, predict text, or recommend a product with astonishing competence. But it does not seek admiration. It does not perform for status. It does not need to persuade a peer that it is worth listening to. In humans, knowledge sits inside a social theater. In machines, knowledge is often stripped of that theater and turned into a function of optimization.

This difference matters more than it first appears. Human knowing is not just about representing reality. It is also about communicating confidence, coordinating groups, and making oneself legible to others. Machine knowledge, by contrast, is often judged on output quality alone. A model can be “right” without understanding why the answer matters, or without caring whether anyone believes it.

That does not make machine intelligence lesser in every sense. It makes it different in a way that is philosophically important. Machines can compress patterns at scale, but they do not inhabit the same ecology of incentives that gave rise to human cognition. They do not compete for mates, status, or identity. They do not need to transform knowledge into charisma. They simply optimize for a target.

And yet, once machines enter human society, they immediately become part of that theater anyway. A recommendation system may not crave approval, but people judge it as trustworthy, biased, elegant, opaque, or manipulative. A chatbot may not care whether it sounds confident, but confidence itself becomes part of the experience. In other words, machine knowledge gets socialized the moment humans encounter it.

The New Courting Ground: Models That Perform

The deepest connection between sexual selection and machine learning is not that both involve intelligence. It is that both are shaped by selection pressures, but of radically different kinds.

For humans, selection rewarded minds that could do more than survive. They had to persuade, coordinate, improvise, and signal quality under conditions where others were watching. For machines, selection rewards systems that reduce error on tasks defined by humans. But those systems are increasingly deployed in environments where they too must perform socially, even if indirectly.

Consider three examples.

First, a researcher with a brilliant intuition. The insight is not enough. It must be packaged as a paper, a talk, a visual, a narrative, and a set of claims others can trust. The idea competes not only on truth but on presentation. This is why two people can have equally good technical arguments and only one will be remembered.

Second, a hiring manager using AI tools. The tool may rank candidates well, but the manager still wants a justification that feels coherent. Accuracy alone is not enough. The output must be interpretable enough to support a social decision. Here the machine is not merely solving a classification problem. It is entering a legitimacy contest.

Third, a student using a language model. The machine can generate an answer quickly, but the student still must decide what counts as learning. If the student copies the output verbatim, the system has supplied information but not necessarily competence. If the student interrogates, edits, and compares it, the model becomes a cognitive prosthetic. The difference is not in the machine alone. It is in how the human uses the result to build judgment and credibility.

These examples reveal a useful framework: machine knowledge becomes valuable in proportion to how well it can be translated into human trust. That is the modern equivalent of display. A model does not need to seduce a mate, but its outputs must seduce institutions, users, and decision makers.

The real competition is no longer just between truths. It is between truths that can be believed, used, and socially carried forward.

The Hidden Similarity Between a Brilliant Mind and a Powerful Model

At first glance, the comparison between sexual selection and machine learning may seem forced. One is biological, the other technical. One emerges from embodied organisms, the other from code and data. But both raise a deeper issue: what is intelligence for, if not only survival?

Human intelligence likely evolved under conditions where being clever was not enough. It had to be visible. That visibility shaped cognition toward language, memory, theory of mind, humor, and abstract reasoning, because these traits broadcasted mental quality to others. A mind became not just an instrument, but a stage.

Machine learning systems are different, but they too are shaped by feedback loops. The objective function is a kind of selection pressure. The model that predicts better survives longer, gets deployed more widely, and influences more outcomes. Yet because it lacks intrinsic social motives, it often appears alien. It can be brilliant in one narrow sense and clueless in another. It may outperform humans on benchmarks while failing to grasp the human rituals surrounding those benchmarks.

This gives us a powerful lens: humans are optimized for social intelligibility, machines for statistical fit. Humans ask, “Does this make sense to others like me?” Machines ask, “Does this minimize loss?” In practice, modern systems need both. A system that is statistically excellent but socially unusable is dead on arrival. A system that is socially smooth but statistically weak becomes seductive nonsense.

This is why the future belongs neither to naked human intuition nor to raw machine output. It belongs to hybrids that can convert one into the other: models that generate pattern, and humans who can convert pattern into meaning, judgment, and action.

The Danger of Confusing Performance With Truth

There is, however, a serious risk in this convergence. Once we see intelligence as a form of display, we may become cynical and assume that all impressive thinking is merely theater. That would be a mistake.

Display is not the same as deception. A peacock’s tail honestly advertises the bird’s ability to survive despite carrying such a burden. Likewise, a person who can reason well under pressure, explain clearly, or connect distant ideas is not necessarily pretending. These abilities are real signals because they are costly to maintain. Good thinking often does prove something important.

But the signal can also be gamed. People can learn the surface markers of intelligence without possessing the underlying depth. Machines can produce fluent answers that look like understanding while failing on edge cases. Institutions can reward presentation over substance. And when that happens, the ornament detaches from the organism.

That is the central modern danger: the imitation of intelligence becomes easier exactly when intelligence becomes more visible. Large language models make this obvious. They can produce fluent prose, confident explanations, and polished structure, which means they can simulate the social surface of thought with uncanny ease. The same quality that makes them useful also makes them dangerous.

The lesson is not to distrust all elegance. The lesson is to build better tests of depth. In biology, selection does not merely reward beauty. It rewards beauty that correlates with robust underlying fitness. In knowledge work, we need similar discipline: outputs should be judged not only by how compelling they sound, but by how well they survive scrutiny, adapt to new contexts, and improve real decisions.

How to Think and Build in the Age of Machine Knowledge

If human intelligence is partly a courting display and machine intelligence is a statistical engine, then the practical question becomes: how do we avoid getting trapped by the most attractive outputs?

The answer is to separate three layers that are often conflated.

1. Generation: Can the system produce a plausible answer?
2. Validation: Can the answer withstand correction, comparison, and stress testing?
3. Legitimation: Can humans understand, trust, and use the answer responsibly?

A great deal of current confusion comes from treating generation as if it were validation. A model writes a fluent paragraph, and we unconsciously promote that fluency into credibility. But fluency is only the first layer. Similarly, a human speaker may sound persuasive while lacking evidence. The social display is real, but it is not sufficient.

A healthier approach is to treat machine outputs as candidates for thought, not verdicts. Use them the way you might use a brilliant but self-interested colleague: fast, inventive, and useful, but never the final authority. Ask what the model sees, what it misses, and what kind of world it assumes. Then ask the human question that machines cannot: what should matter here?

This is where the comparison with sexual selection becomes unexpectedly useful. In mate choice, the expensive signal works because it is hard to fake over time. In knowledge work, we should privilege signals that are equally hard to fake: consistency across contexts, predictive success, transparent reasoning, and the willingness to be corrected. These are the intellectual analogues of costly display.

Key Takeaways

• Treat intelligence as both tool and signal. Great ideas are not only useful, they are also socially legible. If you ignore the signaling dimension, you misunderstand how minds work.
• Do not confuse fluency with depth. Whether it comes from a person or a model, polished output is only the first test. Ask what survives scrutiny.
• Use machine knowledge as a candidate, not a conclusion. Let systems generate options, but require humans to validate meaning, relevance, and consequence.
• Prefer costly signals of understanding. Examples include explanation under questioning, successful transfer to new contexts, and the ability to predict failure modes.
• Build for trust, not just accuracy. In human systems, knowledge has to be adopted, not merely computed. Social usefulness is part of truth in practice.

The Future of Intelligence Is Not Less Human

The biggest mistake we can make is to imagine that machine intelligence is replacing human intelligence in a simple contest of raw capability. That is not what is happening. What is happening is stranger. The machine is entering a world already shaped by minds that evolved to impress one another, and it is learning to perform inside that theater.

So the question is not whether machines will become intelligent like us. The question is whether we will remain wise enough to tell the difference between a convincing performance and a durable understanding.

Human intelligence was never just about solving problems in the abstract. It was also about earning attention, trust, and legitimacy in the presence of others. Machine knowledge now inherits that burden. It can calculate faster than we can, but it cannot tell us what deserves reverence, skepticism, or care.

That remains our job. And perhaps that is the final lesson of the peacock and the model alike: what dazzles is not always what deserves to lead. The future belongs to minds, human and machine, that can turn display into discipline, and output into understanding.