We Are the Neanderthals: AI, Homo Artificialis, and the Quiet Replacement of Human Intelligence
16 – 25 min.
Approximately forty-five thousand years ago, two forms of human intelligence encountered one another in Eurasia. That encounter offers an unexpectedly powerful way to think about AI and human evolution today.
One had lived there for hundreds of thousands of years. Neanderthals (Homo neanderthalensis) were physically powerful, technologically capable, socially organized, able to control fire, manufacture sophisticated tools, hunt formidable animals, care for injured companions, and survive environments that would challenge most modern humans.
The newcomers were Homo sapiens.
For a long time, we told ourselves a convenient story about what happened next. We were smarter. They were primitive. Superior intelligence arrived from Africa, defeated an inferior species, and the Neanderthals became extinct.
That story has not survived modern science particularly well.
There is no convincing evidence that Homo sapiens systematically exterminated the Neanderthals. There is evidence of something far more interesting.
We met them.
We competed with them.
We lived alongside them.
And we had children with them.
Genomic evidence now dates an important episode of Neanderthal admixture shared by the ancestors of present-day non-African populations to approximately 45,000–49,000 years ago. People of European and Asian ancestry today commonly retain roughly 1–2 percent Neanderthal DNA, and some early modern humans possessed much more.
The Neanderthals disappeared as a distinct population.
But they did not entirely disappear.
Part of them is inside us.
That distinction changes the story completely.
And it raises an uncomfortable question.
What if our encounter with artificial intelligence is beginning to resemble theirs with us?
Key Takeaway: AI and Human Evolution
Neanderthals (Homo neanderthalensis) were not simply exterminated by Homo sapiens: the two populations competed, interbred, and partly merged. AI and human evolution may follow an analogous path. As artificial intelligence becomes embodied in humanoid robots and increasingly autonomous systems, the emergence of a possible Homo artificialis may not mean the disappearance of humanity, but a profound transformation in what it means to remain the dominant intelligence on Earth.
They Did Not Have to Be Stupid
The old image of Neanderthals as intellectually defective brutes has steadily collapsed under archaeological and genetic evidence.
They were successful humans.
They existed across vast regions of Eurasia for hundreds of thousands of years. Their subsistence was more flexible than once believed; archaeological evidence shows sophisticated hunting and exploitation of a broader variety of resources than the old stereotype allowed. Reviews of the archaeological record have consequently warned against simply assuming a universal cognitive or technological superiority of Homo sapiens.
This is important because otherwise the analogy with artificial intelligence becomes trivial:
Neanderthals were stupid.
We were smart.
AI becomes smarter.
We disappear.
That is probably the wrong lesson.
The more interesting possibility is that Homo sapiens did not need to possess some enormous individual intellectual superiority over Neanderthals.
We needed to be a more successful system.
Population size may have mattered. Connectivity may have mattered. Migration may have mattered. The transmission and accumulation of innovations may have mattered. Energetic efficiency may have mattered. Greater demographic reservoirs may have mattered.
Genetic studies reinforce this picture. Neanderthal communities could be remarkably small, and at least some populations show evidence of considerable isolation and low genetic diversity. A 2022 genomic study of a Siberian Neanderthal community found extensive homozygosity consistent with very small local groups. More recent work continues to reveal complex regional Neanderthal population structures rather than one large, uniformly connected population.
There is even a physiological argument. Neanderthals had robust bodies that may have required higher energetic expenditure than those of anatomically modern humans. Modeling has explored whether such metabolic differences could have contributed to long-term competitive disadvantage when two populations depended on overlapping resources. This remains one hypothesis among many, not a settled explanation, but it illustrates the larger principle: small differences in efficiency can become enormous when accumulated across populations and generations.
Evolution does not ask which organism deserves to survive.
It does not even ask which organism is more intelligent.
It selects outcomes from differences in reproduction, adaptation, resource use, connectivity and persistence.
That should ring a bell.
The Neanderthals May Have Been Absorbed
A particularly provocative result appeared in 2025.
Andrea Amadei, Giulia Lin and Simone Fattorini constructed a mathematical model asking whether Neanderthals could disappear as a genetically distinct population through repeated small-scale immigration and interbreeding with a much larger Homo sapiens population.
Their model did not require genocide.
It did not even require Homo sapiens to possess a selective advantage.
Under the model’s assumptions, recurrent migration from a much larger sapiens demographic reservoir could progressively dilute Neanderthal ancestry until a distinct Neanderthal population effectively vanished over thousands of years. The authors explicitly caution that this does not prove genetic dilution was the sole historical mechanism; competition, climate and other demographic factors could also have contributed.
But conceptually, the result is extraordinary.
A population can disappear without being annihilated.
It can be absorbed.
That may be one of the most important ideas for thinking about artificial intelligence.
Because the most probable competition between humans and AI may not resemble a war.
It may resemble integration.
The Wrong Question About AI
Much of the public debate asks:
Will artificial intelligence become smarter than humans?
The question sounds profound, but it may be badly formulated.
Smarter at what?
Humans themselves do not possess a single quantity called intelligence that can meaningfully be placed on one universal scale. A mathematician, carpenter, surgeon, pianist and five-year-old child possess radically different competencies, embedded in bodies and social contexts.
Artificial intelligence is even more uneven.
The 2026 Stanford AI Index describes a strikingly jagged capability landscape. Frontier systems now meet or exceed human baselines on some advanced mathematical, scientific and coding evaluations. AI agents have improved dramatically at operating computers. Yet the same technological frontier still fails unexpectedly on tasks humans regard as elementary.
Therefore AI does not have to become universally superior to Homo sapiens.
Neither, presumably, did Homo sapiens have to be universally superior to Neanderthals.
What matters is something closer to effective fitness within an environment.
For artificial intelligence, useful capability is only one part of that fitness. Scalability matters. Availability matters. The speed at which improvements can spread matters. And cost matters.
This is where the comparison with biological intelligence becomes uncomfortable.
Humans possess extraordinary cognitive capabilities. But biological intelligence has severe limitations in almost every dimension of scale.
A human mind cannot simply be copied. A body may one day be cloned—but can a lifetime of memory, experience, judgment, and tacit knowledge be duplicated with it?
A human expert takes decades to produce.
A human needs sleep.
A human occupies one place at a time.
Knowledge acquired by one individual does not automatically appear in every other human brain.
Digital intelligence is fundamentally different.
Once a model acquires a useful capability, that capability can potentially be reproduced across millions of instances. An improvement developed in one place can be distributed globally. The same system can operate simultaneously in hospitals, factories, laboratories, offices, and homes.
The decisive advantage, therefore, may not be that artificial intelligence becomes infinitely smarter than us.
It may simply become good enough, cheap enough, available enough, and reproducible enough to outperform biological intelligence as a system.
That is not merely greater intelligence.
It is a different ecology of intelligence.
Imagine the Neanderthal Looking at Us
Imagine, purely as a thought experiment, that a Neanderthal hunter and a Homo sapiens hunter were individually similar enough in practical intelligence that neither possessed an overwhelming advantage.
Now increase the sapiens population.
Connect distant sapiens communities.
Allow innovations discovered by one group to propagate through many others.
Increase the diversity of available tools.
Reduce the energetic cost of some activities.
Expand trading and mating networks.
Increase environmental flexibility.
Then wait.
No decisive battle is necessary.
A small systemic advantage compounded over generations can transform the demographic landscape.
The same principle applies to AI, except the timescales are radically compressed.
Biological cultural change took generations.
Software can propagate overnight.
That difference may prove more consequential than whether today’s most advanced AI system has an IQ equivalent of 120, 150 or any other number we might attempt to assign to it.
The competition is asymmetric.
We Are Already Mixing With It
There is another reason why the Neanderthal analogy is so powerful.
We are not standing on one side of a battlefield while artificial intelligence stands on the other.
We are already integrating it into ourselves.
Not biologically.
Functionally.
A scientist using an AI system can search literature, generate hypotheses, write code, analyze data and explore mathematical alternatives faster than the same scientist working alone.
A physician supported by AI can combine human clinical judgment with machine pattern recognition.
An engineer can move from concept to simulation to prototype using computational capabilities that would once have required teams of specialists.
A writer can interrogate ideas, compare arguments and restructure prose interactively.
The important competitive unit therefore begins to change.
It is no longer simply human against machine.
Increasingly, it becomes human plus AI.
And that creates a selection pressure of its own.
A company using AI competes with companies that do not.
A scientist using powerful computational assistants competes with scientists who refuse them.
An engineer capable of automating part of a design process competes with one performing every stage manually.
A country that integrates AI into manufacturing, medicine, education, defense and scientific research competes with countries that integrate it more slowly.
No artificial intelligence has to force anyone to adopt it.
Efficiency can do the job.
This is perhaps the most important parallel with the Neanderthals.
The mechanism does not require hostility.
Then Intelligence Acquires a Body
There is, however, one enormous difference between present-day AI and Homo sapiens entering Neanderthal territory.
AI still depends largely on humans to manipulate the physical world.
It can write an extraordinary analysis and still cannot pick up the coffee cup beside your computer.
That limitation matters.
A disembodied intelligence cannot independently build a factory, repair a power line, mine lithium, carry bricks, harvest crops or replace a broken server.
For AI to become a truly different actor in the human environment, intelligence must acquire competent embodiment.
That is where robotics enters the argument.
And that transition has already begun.
Google DeepMind’s Gemini Robotics research has extended multimodal models into vision-language-action systems that can perceive physical environments, reason spatially and directly control robotic actions. NVIDIA has similarly developed its GR00T family of foundation models for humanoid robotics. The objective is no longer to program a robot separately for every movement, but to give robots increasingly general models that can learn tasks, interpret instructions and transfer behavior across different situations.
This transition from artificial intelligence to embodied artificial intelligence deserves a name.
I will call its mature form Homo artificialis.
Not because a robot would literally belong to the biological genus Homo. It would not. The term is philosophical rather than taxonomic, and variations of Homo artificialis have already appeared in discussions of technologically transformed or artificial humanity.
Here I use it more specifically.
Homo artificialis is artificial intelligence when cognition, perception and physical agency become integrated into a scalable autonomous machine capable of operating in the environment built by Homo sapiens.
Software gave artificial intelligence a mind-like interface.
Robotics is beginning to give it a body.
That is a much larger transition.
Homo Artificialis Is Not Here Yet
This is where discipline is essential.
The humanoid robot revolution is real.
But the humanoid robot hype is also real.
The International Federation of Robotics stated in 2025 that there was not yet massive deployment of humanoids and that many systems remained prototypes or demonstrators, with practical applications still needing to be proven. Gartner predicted in early 2026 that fewer than twenty companies would have humanoid robots operating at production scale in manufacturing and supply-chain applications by 2028.
The Stanford AI Index provides an even better reality check.
Robots have become highly competent in structured environments and simulated manipulation benchmarks. Yet on one benchmark involving real household tasks, robots succeeded only about 12 percent of the time. Human homes remain brutally difficult places for machines because they are variable, cluttered and unforgiving.
Humans possess an astonishing physical generality that we usually fail to appreciate.
We can walk across uneven ground, open an unfamiliar drawer, recognize a wet floor, move a chair, catch a falling glass, comfort a child and use a screwdriver we have never seen before—all without downloading a new model.
The human body is still an extraordinary general-purpose machine.
But notice the direction of travel.
The Next Decade
By 2026, humanoids had already moved beyond laboratory videos into genuine industrial trials.
BMW is testing humanoid systems in automotive production. Schaeffler and robotics companies have announced plans involving deployments numbering in the thousands over the coming years. CATL has placed heavy-duty humanoid systems into battery-production environments. Chinese manufacturers are building production capacity for humanoids rather than merely constructing one-off laboratory prototypes.
Industry expectations remain uncertain, and they should not be mistaken for scientific predictions. In August 2026, Unitree’s chief executive estimated that the robotics equivalent of a “ChatGPT moment”—robots able to understand and perform a broad range of unfamiliar tasks from ordinary instructions—might be two to ten years away. He simultaneously acknowledged that present humanoids remain less efficient than humans in many real applications.
That combination is exactly where we should be intellectually.
Skeptical of the timetable.
Serious about the trajectory.
I would therefore not claim that general-purpose robots will surpass human physical competence within the next decade.
Nobody knows that.
But it is entirely reasonable to expect machines to surpass humans across an expanding set of economically important physical tasks during that period.
In some dimensions, they inevitably will.
Strength can be engineered.
Precision can be engineered.
Running speed can be engineered.
Resistance to heat, radiation, toxic chemicals and vacuum can be engineered.
Night vision can be engineered.
A machine can operate without fear of injury.
Its damaged component can be replaced.
Its learned software can be copied.
Its sensors need not be restricted to the narrow electromagnetic spectrum visible to human eyes.
And unlike biological workers, robots need not all have the same body.
The humanoid form is useful precisely because our civilization was constructed around the geometry of the human body—stairs, doors, tools, vehicles, shelves and factories.
But once intelligence controls machines, there is no evolutionary requirement that every machine resemble us.
Homo artificialis may have two legs.
Or six.
Or wheels.
Or wings.
Or no permanent body at all, moving its intelligence from one machine to another.
We are constrained by the bodies evolution gave us.
Artificial agents will be constrained principally by engineering.
That is an enormous asymmetry.
When the Advantage Reverses
The first industrial robots surpassed humans decades ago at particular tasks.
They could weld more consistently, move heavy objects repeatedly and operate for long periods without fatigue.
But they were stupid machines.
Their environment had to be redesigned around them.
Their movements had to be programmed.
Their world had to remain predictable.
AI changes the equation because the machine begins to deal with variability.
Computer vision lets it perceive.
Language models let it interpret instructions.
World models may allow it to anticipate consequences.
Reinforcement learning allows behavior to improve.
Foundation models allow knowledge acquired in one context to generalize, imperfectly but increasingly, to another.
The convergence matters more than any single breakthrough.
Artificial intelligence gives robotics flexibility.
Robotics gives artificial intelligence agency.
And scalable manufacturing gives both population.
That is the moment at which Homo artificialis becomes more than a metaphor.
We Should Not Expect a War
Science fiction has trained us badly for this transition.
It expects rebellion.
A machine becomes conscious.
It realizes humans are its enemy.
The robots attack.
Civilization fights back.
One side wins.
But the Neanderthals suggest another possibility.
There may be no war.
No declaration of independence.
No day on which television announces:
“Human supremacy ended this morning at 9:37.”
Instead, millions of perfectly rational choices may produce the transition.
This factory buys robots because they are cheaper.
This hospital uses AI because diagnosis improves.
This university integrates AI because research becomes faster.
This logistics company automates because machines work overnight.
This family purchases a domestic robot because elderly parents need assistance.
This engineer delegates another task.
This programmer delegates another function.
This government delegates another analysis.
Every decision makes sense individually.
Taken together, they alter the ecology in which human beings live.
That is how profound transitions often occur.
Silently.
The Competition May Not Be Humans Against AI
There is an even more consequential possibility.
Imagine two future workers.
One refuses artificial augmentation on principle and relies exclusively on biological cognition.
The other has an AI system permanently available—remembering everything, monitoring information, translating languages, performing calculations, generating simulations and coordinating robotic systems.
Which one is the more capable economic agent?
Now imagine two companies.
Then two universities.
Then two countries.
The evolutionary analogy suddenly shifts.
AI may not replace Homo sapiens directly.
Homo sapiens integrated with AI may outcompete Homo sapiens without it.
The pressure for integration therefore becomes endogenous.
We do not merge with AI because machines conquer us.
We merge because refusing to do so becomes increasingly costly.
The Neanderthals did something biologically analogous with Homo sapiens.
We mixed.
The populations were not equal in scale.
One eventually ceased to exist separately.
But some of its inheritance survived inside the population that followed.
Our integration with AI could be the mirror image of that process.
We may provide the biological component.
AI provides an increasingly large share of the cognitive infrastructure.
Robotics provides artificial physical agency.
Together they produce something that no longer fits comfortably into the traditional picture of Homo sapiens acting alone.
That is Homo artificialis in the broader sense:
not necessarily a robot replacing a human,
but a civilization in which biological humans and artificial intelligence have become functionally inseparable.
Who Absorbs Whom?
This leads to a peculiar question.
Suppose Homo sapiens still exists five hundred years from now.
People are born.
They fall in love.
They raise children.
They eat, sleep, argue, laugh and die.
Nothing resembling biological extinction has happened.
But almost every consequential intellectual activity depends on artificial cognition.
Machines design most technologies.
Machines perform much scientific inference.
Machines coordinate infrastructure.
Robotic systems carry out much physical production.
Human decisions are continuously informed by artificial systems vastly superior to individual humans in memory, calculation and access to information.
Perhaps neural interfaces eventually connect some of these systems more directly to biological brains, although nothing in the argument depends on that happening.
Human beings remain everywhere.
But unaided human intelligence is no longer the dominant cognitive force of civilization.
Have we survived?
Of course.
Have we been replaced?
In another sense, perhaps.
The ambiguity is precisely the point.
Neanderthals disappeared as an independent population while part of their genetic inheritance survived within us.
Homo sapiens could survive biologically while surrendering an increasing proportion of civilization’s cognition to an artificial substrate.
We would remain.
And yet something fundamental would have changed.
The Last Human Advantage
There is one enormous advantage that we possess and the Neanderthals did not.
We can see the transition coming.
Neanderthals could not study demographic models describing their interaction with Homo sapiens.
They could not regulate sapiens development.
They could not decide which sapiens abilities should be deployed.
They could not choose how rapidly sapiens populations should expand.
We created artificial intelligence.
We construct its processors.
We build its robots.
We provide its energy.
We determine—at least today—where it is deployed and what authority it receives.
That does not guarantee permanent control. Complex technological systems can become economically and institutionally difficult to reverse long before they become literally autonomous.
But it means the future has not yet been determined.
The most important decision may therefore not concern how intelligent machines are allowed to become.
Attempts simply to stop intelligence from improving may prove unrealistic.
The deeper question is what humans must remain capable of doing even in a civilization where machines can do those things better.
We invented calculators without abandoning mathematics.
We built aircraft without deciding that walking was obsolete.
We created cameras without eliminating human vision.
The challenge with AI is larger because cognition is not merely another human tool.
It is the faculty through which we control all our other tools.
When we outsource cognition, we are outsourcing part of the mechanism by which we decide what should be outsourced.
That recursive property makes AI historically unique.
The Neanderthal Lesson
Perhaps Neanderthals were not defeated.
Perhaps “defeat” is the wrong concept.
They encountered another population with different demographic and adaptive characteristics. They competed. They exchanged genes. They influenced one another. One population expanded. The other ceased to persist independently.
And approximately forty thousand years later, we are discovering that the boundary between “them” and “us” was never as clean as we imagined.
The same may eventually be true of artificial intelligence.
We speak today as though humanity and AI were clearly separate categories.
Human.
Machine.
Natural.
Artificial.
Creator.
Tool.
Those distinctions feel obvious because we are living at the beginning of the process.
They may not remain obvious.
As artificial cognition becomes embedded in our institutions, as humans increasingly think with machines, and as artificial intelligence acquires increasingly competent physical bodies, the boundary may begin to dissolve.
Then the central question will no longer be:
Will AI destroy Homo sapiens?
That may turn out to have been the primitive question.
The more difficult question is:
What happens when Homo sapiens becomes unable—or unwilling—to function without the intelligence it created?
The Neanderthals left us an unexpected answer.
You do not need to be exterminated to cease being the dominant form of intelligence.
You can be outscaled.
You can be outcompeted.
You can mix with the newcomer.
You can become part of something more successful.
And you can survive inside the system that replaces you.
Perhaps artificial intelligence is not our enemy.
Perhaps it is not even our successor.
Perhaps, just as part of the Neanderthals became part of us, artificial intelligence will become part of whatever comes after Homo sapiens.
We may still be here.
Our descendants may still call themselves human.
But the dominant intelligence on Earth may no longer reside entirely inside the human skull.
Its mind will extend through silicon.
Its memory through networks.
Its senses through machines.
Its hands through robots.
And its ancestry will include us.
Homo sapiens met Homo neanderthalensis and the two became entangled.
Homo sapiens is now creating Homo artificialis.
This time we are not merely witnessing evolution.
We are engineering one of its possible successors.
And perhaps the strangest possibility of all is that the successor will still be us.
References
Amadei, Andrea, Giulia Lin, and Simone Fattorini. “A Simple Analytical Model for Neanderthal Disappearance Due to Genetic Dilution by Recurrent Small-Scale Immigrations of Modern Humans.” Scientific Reports 15 (2025).
Fu, Qiaomei, et al. “An Early Modern Human from Romania with a Recent Neanderthal Ancestor.” Nature 524 (2015): 216–219.
Google DeepMind. Gemini Robotics: Bringing AI into the Physical World. 2025.
International Federation of Robotics. Humanoid Robots: Vision and Reality. Frankfurt: IFR, 2025.
Skov, Laurits, et al. “Genetic Insights into the Social Organization of Neanderthals.” Nature 610 (2022): 519–525.
Stanford Institute for Human-Centered Artificial Intelligence. AI Index Report 2026. Stanford University, 2026.
Sümer, Arev P., et al. “Earliest Modern Human Genomes Constrain Timing of Neanderthal Admixture.” Nature 2024.
Vaesen, Krist, et al. “An Emerging Consensus in Palaeoanthropology: Demography Was the Main Factor Responsible for the Disappearance of Neanderthals.” Scientific Reports 11 (2021).
Editorial transparency note: This article, as with all articles published on this site, was conceived, directed, written, and reviewed by Prof. Maurício Veloso Brant Pinheiro. Artificial intelligence was used as an assistant for editorial refinement, formatting, image generation, SEO metadata, and publication workflow.
Copyright 2026 AI-Talks.org
