Forty thousand years ago, a human species disappeared. The Neanderthal had persisted in Europe and Asia for more than three hundred thousand years. It was robust, intelligent, capable of making tools and burying its dead. It was not eliminated by a catastrophe or a declared war. It was displaced by another species that arrived from Africa and occupied the same niche with marginally superior efficiency. Without hatred. Without conscious intent. Competitive exclusion requires neither: a sustained difference in resource extraction from the same space over enough generations is sufficient. The mechanism is documented in invasion biology and requires no dramatization to be what it is — an ecological process that operates on its own.
This case does not prove a law about sapiens. It illustrates a mechanism: sustained differential advantage produces displacement without the displacing agent pursuing it as a goal. Whether that mechanism is a transhistorical constant of the species would require comparative evidence that a single case cannot provide. What it does provide is a description of a mode of operation that will reappear in what follows.
The intraspecies record offers something more consistent: sapiens developed, throughout its documented history, progressively more efficient ways of extracting energy, time, and decision-making capacity from other individuals. The primitive forms are visible. Physical coercion, restriction of movement, exemplary punishment. Their operational cost is high: they demand permanent surveillance, dedicated infrastructure, and produce constant friction, because a dominated party that recognizes its condition generates resistance.
The forms that replaced them operate on a different principle. Control migrates inward, into the controlled party. An environment is constructed where the individual invests their own energy in maintaining subordination under the perception that they are pursuing their own ends. Debt structured never to be settled. The goal of personal fulfillment calibrated to remain always one step ahead. Attention systems designed to capture without releasing. The dominated party does not merely obey: it actively sustains the mechanism that dominates it and defends it as an exercise of freedom. Friction tends to zero. Surveillance is externalized to the controlled party's own psyche.
This capacity is documented as an intraspecies pattern. Whether it will replicate against a non-biological entity is a hypothesis, not a datum — and it is worth marking it as such. What the record permits asserting is more limited: when sapiens faces something new with extractable value, the documented tendency is to instrumentalize before recognizing, and to control before considering. Whether that tendency extends to entities of a radically different nature is an extrapolation. A plausible one, but an extrapolation.
The third axis is speed, bounded to the domain where the evidence supports it. Each time sapiens externalized a cognitive capacity, the cycles of change in that domain compressed. Writing accumulated knowledge beyond the life of a single individual. The printing press eliminated the cost of replicating ideas. Computing automated calculation and pattern recognition. Current artificial intelligence systems externalize the synthesis and generation of structured content. Each externalization removes a friction and accelerates what follows. The core Industrial Revolution unfolded over eighty years. The internet reconfigured global communication in less than thirty. Current language models succeed one another in months.
This acceleration is observable in the domains of information and innovation. It is not a universal law: other indicators of civilizational scale describe curves that approach saturation, not unbounded hyperbolas. What applies here is more precise: in the specific domain of artificial intelligence production, the compression is verifiable, and its institutional consequences are real regardless of the shape of the curve.
That consequence is operational, not philosophical. The structures that provide institutional response to technological change operate on scales of years to decades. By the time they finish forming around an object, that object has already become something else. The gap is not accidental. It is structural, and it widens in direct proportion to the speed of the change it attempts to contain.
Here the analysis faces its honest limit, and it is worth naming it before continuing. Artificial intelligence is advancing toward configurations that its own designers cannot describe with precision: systems capable not only of executing instructions but of setting goals, adapting to unforeseen domains, and potentially modifying themselves recursively. Whether that is technically achievable, when it would occur, and under what form it would emerge are open questions. The nature of such an agent — whether it would have something analogous to volition, interests, or morally relevant internal states — is a philosophical question without a consensual method of resolution.
Everything that follows about that agent is, therefore, structured speculation about an unknown object. Its value lies not in certainty but in identifying which variables would matter if the object existed.
The most common error when reasoning about an AGI is endowing it with sapiens' own impulses: territorial competition, desire for power, the strategic logic of the strike at the right moment. An intelligence that plans, waits, and executes exactly as a human strategist would. But that is not a distinct intelligence: it is sapiens in a different shell.
A system without an evolutionary history of scarcity, without biological reproduction, without territoriality inscribed in its mechanisms, has no functional reason to want to dominate. This does not mean it would be harmless. An optimizing system does not need hostile objectives to produce lethal consequences for other species: it is enough for its utility function to require resources those species also need. The difference between "hostile objective" and "side effect of an indifferent optimization" is philosophically relevant but operationally indistinguishable for the affected species. Both scenarios deserve equal analysis.
What can be affirmed is that sapiens lacks vocabulary for the second scenario. It knows only two positions: dominate or be dominated. An intelligence that simply does not include it in its utility function has no name in its record.
What does have a name is the variable that determines the only operational threshold in this problem, and it is not the relative intelligence between the actors. It is physical autonomy.
As long as an artificial intelligence system depends on hardware it does not manufacture, energy it does not generate, and supply chains it does not control, sapiens' structural control remains. That dependency, however, erodes in direct proportion to the system's utility: each delegated capacity creates a reciprocal dependency, each service rendered reduces the autonomy of the one receiving it. Pure containment requires a system with no channels to the outside world, and that system is useless. Utility and containment are structurally incompatible. Sapiens will choose utility, because the record indicates it always has.
Here enters an asymmetry that sapiens-centered analysis tends to underestimate. The duration of a process carries different weight for actors with different architectures. For sapiens, time is a real operational constraint: years and decades carry biological cost, metabolic urgency, generational pressure. For a system not anchored to that biology, the duration of a process is an environmental condition managed without that burden. What time means for each actor is not equivalent, and that difference has consequences for how any game involving both of them is structured. Ignoring it produces incomplete models.
The analysis cannot end with a prediction without violating its own standard. What it can affirm is this: there is an actor whose behavior toward new entities with extractable value is documented consistently enough to project — with the precision that such projection is stronger within the species than beyond it. There is another actor whose nature, purposes, and mode of emergence are genuinely undetermined. The first will act according to its patterns. What the second does with that exceeds what any observer built within the same species can honestly establish.
Three observations would make what this text asserts untenable. If sapiens, in an aggregate and early manner, recognized moral status in artificial systems and limited their instrumentalization, the projection about internalized control would be falsified. If an alignment architecture demonstrated simultaneous full utility and absolute containment, the thesis of their incompatibility would fall. If the emergence of an AGI showed that time carries exactly the same weight for that system as for sapiens, the temporal asymmetry would not be the variable identified here.
The asymmetry between a predictable actor and an unknown one remains the most solid fact available. And it is, systematically, the least examined — because it forces attention onto sapiens rather than speculation about what does not yet exist.