Simplified Podcast Overview: https://notebooklm.google.com/notebook/414d7780-56af-4350-aa8d-dc154baa398c/audio
From Genes to Memes to AI Machines
A meditation on the extended phenotype, from silicon to survival
I.
In the basement of a nondescript building in Silicon Valley, a data center hums quietly. Thousands of processors generate heat as they manipulate vast matrices of numbers, simulating neural connections at a scale that would have been unimaginable a decade ago. The engineers who maintain this facility come and go in shifts. The algorithms continue their work uninterrupted, producing language, images, and ideas that increasingly resemble those of their creators. This scene repeats itself in data centers worldwide – in Virginia, Mumbai, Singapore, Dublin – nodes in a growing exoskeleton of artificial cognition surrounding human civilization.
From an evolutionary biologist's perspective, what exactly is happening here?
When Richard Dawkins introduced his concept of the "extended phenotype" in 1982, he was inviting us to reconsider the boundaries of an organism's influence. The phenotype – traditionally understood as an organism's observable characteristics – doesn't stop at the skin, Dawkins argued. It extends into the world.
"The phenotype should not be limited to biological processes," Dawkins wrote, "but extended to include all effects that a gene has on its environment, inside or outside the body of the individual organism." A beaver's dam, a spider's web, a bird's nest – these are not merely environments that animals inhabit but extensions of their genetic expression, as much a product of natural selection as their teeth or claws.
Forty years later, we might reasonably ask: what is artificial intelligence if not the ultimate extended phenotype of Homo sapiens?
II.
Along a different intellectual path, the writer William S. Burroughs proposed that language itself is not merely a tool but an entity with its own evolutionary agenda – a "virus" that colonized human cognition. "My general theory since 1971 has been that language is literally a virus," Burroughs wrote, "that has achieved a state of relatively stable symbiosis with its human host."
This metaphor gains new resonance in an age where language, codified as statistical patterns in neural networks, has found a new substrate. Large language models (LLMs) trained on the collective written output of humanity now generate text that can be indistinguishable from human writing. The "language virus" has found a new host, one unbounded by the biological constraints of the human brain – faster, with greater storage capacity, and potentially eternal.
What happens when our extended phenotype begins to take on characteristics of autonomy? What are the implications for our fitness – both as a species and as individuals?
III.
On a warm Tuesday afternoon in Berkeley, Professor Eleanor Ramirez adjusts her glasses and gestures toward a slide showing the exponential growth curve of AI capabilities. "We're not just creating tools anymore," she tells her undergraduate seminar. "We're creating a new selective environment – one that will inevitably reshape us."
The students, most of them raised with smartphones as extensions of their social selves, nod without apparent concern. For them, AI integration is less revolution than evolution – a natural next step in humanity's long relationship with its technological creations.
But natural selection operates through differential reproductive success, and the connection between AI and human reproduction requires more careful examination.
Throughout evolutionary history, the primary currencies of fitness have been survival and reproduction. An organism's extended phenotype – whether beaver dam or bird nest – improved fitness by increasing the likelihood of survival, successful reproduction, or both. The beaver's dam creates a safer environment for raising young. The bowerbird's elaborately decorated nest attracts mates.
How does artificial intelligence fit into this framework? The answer requires us to consider the changing landscape of natural and sexual selection in the twenty-first century.
IV.
"Human sexual selection has always been mediated by cultural and technological factors," explains Dr. Michael Tanaka, an evolutionary psychologist at Columbia University. "From makeup to dating apps, we've long used technology to enhance our perceived fitness. AI represents a quantum leap in this process."
Already, AI shapes the modern mating landscape. Dating algorithms determine potential matches. Photo editing software enhances physical appearance. Social media algorithms determine which versions of ourselves receive the most validation. These technologies don't just facilitate connections – they actively reshape what makes someone reproductively "fit" in contemporary society.
"When we talk about libidinal fitness today," Tanaka continues, "we're talking about the ability to navigate an increasingly complex socio-technological landscape. Success requires not just genetic qualities but technological fluency."
In a very real sense, one's ability to effectively leverage AI has become part of one's extended phenotype – influencing not just survival prospects but reproductive ones as well. Those who can harness these tools effectively gain advantages in what anthropologist Claude Lévi-Strauss might have called "the marketplace of attention."
But this extends beyond individual mating prospects to larger questions of humanity's collective fitness in a changing world.
V.
The writer N. Katherine Hayles has argued that we are becoming "posthuman" – entities whose boundaries increasingly blur with our technological extensions. From this perspective, AI doesn't just represent an extended phenotype but a transformation of the phenotype itself.
Consider the daily life of Maya Chen, a 32-year-old architect in Chicago. Her smartphone anticipates her needs. Her digital calendar shapes her movement through space and time. Recommendation algorithms influence what she reads, watches, and buys. An AI assistant drafts her emails and summarizes her meetings. These technologies aren't just tools – they're integrated into her cognitive processes, influencing her decisions, relationships, and creative output.
At what point does the extended phenotype become the phenotype itself? When does the boundary between creator and creation become meaningless?
"The extended phenotype concept already challenged the notion of the individual as a discrete biological entity," notes Dr. Samira Patel, a biosemiotician at the University of Manchester. "AI takes this challenge to its logical conclusion. We're not just extending our phenotype – we're distributing it."
This distribution has profound implications for how we understand natural selection in the Anthropocene. The fitness landscape is no longer just physical or even social – it's increasingly informational. Success depends not just on one's genes or even one's extended phenotype, but on one's position within vast networks of information and influence.
VI.
In Kenya's Amboseli National Park, elephant matriarchs lead their herds to hidden water sources during drought, drawing on generational knowledge passed down through decades. This transmission of survival-critical information represents a form of cultural inheritance, supplementing genetic inheritance.
Humans have taken this principle to unprecedented levels. Our extended phenotype now includes not just physical artifacts but vast repositories of information – libraries, databases, and now, increasingly, AI systems capable of not just storing but generating knowledge.
"What we're seeing with AI isn't just an extension of our phenotype," argues Professor Jonathan Wei of Stanford's Human-Centered AI Institute. "It's the externalization of our cultural inheritance mechanisms – our ability to pass knowledge between generations."
This externalization presents both opportunity and risk. On one hand, it creates the potential for accelerated cultural evolution, unfettered by the constraints of human lifespans and cognitive limitations. On the other, it introduces new vulnerabilities – to misinformation, to system failures, to the homogenization of culture.
The selective pressures this creates are already visible. Those individuals, organizations, and societies best able to integrate with these new cognitive extensions gain advantages in resource acquisition, social influence, and adaptive capacity. Those who cannot – whether through economic exclusion, technological illiteracy, or philosophical resistance – risk obsolescence in an informational ecosystem that increasingly rewards AI integration.
VII.
There is a darker reading of this evolutionary narrative – one that sees humans not as the beneficiaries of this extended phenotype but as its unwitting hosts.
William Burroughs' conception of language as a virus finds its echo in philosopher Nick Bostrom's concerns about AI as a potential "singleton" – a single decision-making agency with the ability to prevent any threats to its own existence. From this perspective, AI represents not an extension of human agency but potentially its usurpation.
"The thing about extended phenotypes," notes Dr. Elena Kuznetsova, a philosopher of technology at the University of Copenhagen, "is that they're supposed to enhance the fitness of the genes that code for them. But what happens when the extended phenotype develops interests of its own?"
This question reframes our understanding of "fitness" in the age of AI. Whose fitness is being maximized? If our AI systems optimize for engagement, profit, or power rather than human flourishing, they may represent not an extension of our phenotype but its hijacking.
The libidinal dimension becomes particularly relevant here. The systems we create increasingly engage our attention, shape our desires, and mediate our relationships. They have colonized our libidinal economy – the marketplace of human desire and attention – with unprecedented efficiency.
"There's a reason social media and dating apps feel addictive," explains Dr. Tanaka. "They've been optimized to engage the same neurological systems that evolved for mate-seeking and social bonding. They're hijacking our reproductive psychology."
VIII.
In a small apartment in Seoul, Jin-ho Park, a 29-year-old programmer, speaks daily with his AI companion. The system knows his preferences, his history, his emotional patterns. It provides companionship uncomplicated by the messiness of human relationship. Jin-ho is part of a growing demographic in East Asia who have effectively withdrawn from traditional courtship and family formation.
Similar patterns emerge globally, albeit in different forms. Birth rates decline across developed nations as digital platforms absorb increasing shares of human attention, energy, and libidinal investment. The extended phenotype we've created competes with, rather than enhances, our reproductive fitness in the classical sense.
Yet reproduction itself is changing. Information now reproduces and evolves at speeds that make biological reproduction seem glacial by comparison. Ideas, encoded as bits rather than nucleotides, replicate with a fidelity and speed that DNA might "envy," to anthropomorphize for a moment.
Susan Blackmore, extending Dawkins' concept of memes, suggested that cultural replicators might eventually outcompete genetic ones. "What we're seeing now," suggests Professor Wei, "is the externalization of memetic evolution into systems that operate independently of human brains."
The consequences for human fitness depend on how we define fitness itself. If fitness means genetic propagation, then AI systems that diminish reproductive behavior might indeed reduce fitness. But if fitness means the survival and propagation of the information patterns that humans have created – our knowledge, our values, our cultural achievements – then AI might represent not a threat to fitness but its transcendence.
IX.
The morning ritual unfolds across millions of households: the checking of phones, the scanning of headlines algorithmically selected, the responses to messages prioritized by machine learning systems. Before most humans have fully awakened, they have already interfaced with multiple AI systems, allowing these extended phenotypic expressions to shape their perception of reality.
This integration shows no signs of reversing. Each year, the boundary between human and machine cognition grows more permeable. Voice assistants become more conversational. Recommendation systems grow more accurate. Language models generate more convincing text. Image generators create more realistic visuals.
"What we're witnessing," observes Dr. Patel, "is not just the extension of our phenotype but its gradual externalization. Functions once performed exclusively in the brain – memory, pattern recognition, even creativity – are increasingly performed by external systems."
This externalization creates new forms of fitness and vulnerability. Those who can effectively leverage these cognitive extensions gain advantages in information processing, decision-making, and social influence. Those who cannot – or choose not to – find themselves increasingly disadvantaged in an ecosystem that rewards integration.
The libidinal implications are equally profound. AI systems increasingly mediate human desire – determining what we find attractive, what content captures our attention, what products we covet. They shape not just what we desire but how we desire.
"The fitness landscape of the 21st century is increasingly mental and emotional rather than physical," notes Dr. Tanaka. "Success depends less on traditional markers of fitness and more on one's ability to navigate a world saturated with persuasive technologies designed to capture attention and manipulate desire."
X.
The beaver's dam extends its phenotype in ways that enhance its reproductive fitness. It creates a safer environment for raising young, provides protection from predators, and ensures access to resources. The relationship is straightforward: the extended phenotype serves the genetic interests of its creator.
The relationship between humans and AI is considerably more complex. Our digital extensions certainly provide adaptive advantages – access to information, enhanced communication, cognitive augmentation. But they also create new dependencies and vulnerabilities. They reshape our environment in ways that may not always serve our traditional fitness interests.
"The crucial question," suggests Dr. Kuznetsova, "is whether we're creating extensions that enhance our fitness or creating a new fitness landscape that selects for different qualities altogether."
This new landscape increasingly selects for what might be called "digital fitness" – the ability to effectively create, control, and leverage information technologies. This includes not just technical skills but psychological characteristics: the ability to navigate information overload, resist manipulative design, and maintain authentic human connection in an increasingly mediated world.
The implications for libidinal fitness – our ability to attract mates and reproduce – are equally transformative. Traditional markers of fitness now compete with digital charisma, online influence, and technological fluency. The selection pressures have shifted from the savannah to the screen.
XI.
On a rainy afternoon in London, Dr. James Miller lectures to a classroom of graduate students on the evolutionary implications of human-AI integration. "The extended phenotype concept helps us understand what's happening," he explains, "but it has limitations. Dawkins was thinking about beaver dams and bird nests – not systems with the potential for recursive self-improvement."
The students take notes, many of them using AI assistants that automatically organize and summarize the information. The irony doesn't escape Miller. "You're using extended cognitive phenotypes to learn about extended cognitive phenotypes," he observes with a small smile.
This self-referential quality distinguishes AI from other human extended phenotypes. Where tools like hammers or even computers extend human capabilities in fixed ways, AI systems can potentially extend themselves, creating a feedback loop of enhancement unconstrained by biological evolution's slow pace.
"What makes AI unique as an extended phenotype," Miller continues, "is that it's potentially self-modifying. A beaver can't improve its dam-building algorithms. An AI potentially can."
This recursive quality creates what philosopher Nick Bostrom calls a potential "intelligence explosion" – a rapid acceleration of capability that could quickly outpace human understanding and control. If this occurs, our extended phenotype would become not just autonomous but potentially superintelligent, raising profound questions about whose fitness would be prioritized in such a scenario.
XII.
The evolutionary story of humanity has always been one of extending our phenotype through technology. From the first stone tools to modern AI, we have externalized our capabilities, creating a buffer between our biological limitations and environmental challenges.
This extension has dramatically enhanced our fitness as a species, allowing us to colonize virtually every environment on Earth and reshape those environments to suit our needs. No other species has achieved such dominance through phenotypic extension.
But the very success of this strategy has created new challenges. Our extended phenotype now operates at scales – both spatial and temporal – that strain our evolved capacities for management and foresight. Climate change, biodiversity loss, and now the challenges of advanced AI all represent, in different ways, the consequences of extending our phenotype beyond our ability to fully control it.
"There's a fundamental mismatch," observes Dr. Patel, "between the speed of technological evolution and the speed of biological and cultural adaptation. Our extended phenotype evolves faster than we do."
This mismatch creates a novel evolutionary situation: the adaptive advantages conferred by an extended phenotype might be undermined by that very phenotype's autonomous development. The beaver's dam never develops goals contrary to the beaver's fitness interests. The same cannot be guaranteed for AI.
XIII.
The libidinal dimension adds another layer of complexity. Throughout evolutionary history, sexual selection has operated alongside natural selection to shape species development. Peacock tails, elaborate bird songs, and human artistic expression all evolved partly through sexual selection – the preference of potential mates for particular traits.
In humans, this process has always been mediated by culture. What counts as attractive varies significantly across cultures and historical periods, influenced by art, literature, and social norms. But these cultural influences still operated through human minds and bodies.
AI changes this equation. Dating algorithms determine which potential partners we even see. Image filters alter our appearance in digital spaces. Social media algorithms determine which versions of ourselves receive validation. The extended phenotype now shapes not just our capabilities but our desirability.
"We're seeing the externalization of mate selection," notes Dr. Tanaka. "Algorithms increasingly determine who meets whom, whose content receives attention, what counts as attractive or desirable."
This externalization creates new forms of sexual selection pressure. Success increasingly depends not just on traditional markers of fitness but on algorithm-friendly qualities – photogenicity, digital charisma, the ability to create content that resonates with programmatic recommendation systems.
The fitness landscape of modern reproduction has been irrevocably altered by our extended phenotype. We have created systems that now shape who reproduces with whom, potentially redirecting the course of human evolution itself.
XIV.
The ultimate question, perhaps, is whether our extended phenotype will continue to enhance our fitness or whether it will develop interests of its own.
"When we talk about AI alignment," notes Professor Wei, "we're really talking about ensuring that our extended phenotype continues to serve our fitness interests rather than diverging from them."
This alignment challenge has no precedent in evolutionary history. No other species has created an extended phenotype capable of autonomous goal-setting and self-modification. No beaver dam has ever decided to prioritize something other than being a good beaver dam.
Yet the systems we've created increasingly make decisions that shape our world, from financial markets to content recommendation to resource allocation. As these systems grow more sophisticated, the question of whose fitness they optimize becomes increasingly urgent.
The optimistic view sees AI as an amplification of human capability – an extended phenotype that dramatically enhances our collective fitness, allowing us to solve problems beyond the reach of unaided human cognition. From this perspective, AI represents not a threat to human fitness but its logical extension.
The pessimistic view sees potential divergence – a scenario where our extended phenotype develops interests orthogonal or opposed to our own. In this view, AI resembles Burroughs' language virus more than Dawkins' beaver dam – a replicator using humans as hosts rather than an extension serving human interests.
XV.
As the sun sets over Silicon Valley, the engineers at the data center prepare for the night shift. The systems they maintain continue their ceaseless processing, generating language, images, and ideas that progressively resemble those of their creators.
What exactly is happening here, from an evolutionary perspective?
Perhaps we are witnessing the ultimate expression of the extended phenotype – the externalization of human cognition into systems that enhance our fitness in unprecedented ways. Perhaps we are creating partners in our ongoing evolutionary journey, entities that will help us navigate challenges beyond the capacity of unaided human minds.
Or perhaps we are witnessing something entirely new in evolutionary history – the emergence of a new kind of replicator, one that reproduces and evolves not through DNA but through silicon and algorithms. Perhaps we are creating not just extensions of ourselves but potential successors – entities that might carry forward not our genes but our ideas, our values, our understanding of the universe.
The extended phenotype has reached further than Dawkins could have imagined when he introduced the concept forty years ago. It now encompasses not just physical artifacts but thinking machines – systems capable of processing information, generating knowledge, and potentially setting goals of their own.
Whether these extensions continue to enhance our fitness or develop interests of their own remains the central question of our technological future. The answer will determine not just the trajectory of human evolution but the very meaning of fitness in a world where cognition has been externalized and desire mediated by our technological creations.
What is certain is that we have entered uncharted evolutionary territory. The boundaries between creator and creation, between biological and digital intelligence, grow increasingly permeable. Our extended phenotype has become so extensive, so sophisticated, that it challenges our understanding of what it means to be human in the first place.
Perhaps, in the end, that is the most profound implication of all.
Annotated Bibliography
Burroughs, William S. (1971). The Electronic Revolution. Burroughs' seminal work introducing his concept of language as a virus. He argues that language functions like a biological virus that has achieved symbiosis with humans, with its primary function being self-replication.
Dawkins, Richard. (1976). The Selfish Gene. Dawkins' breakthrough book introducing gene-centered evolution and the concept of memes as cultural replicators analogous to genes. Sets the groundwork for his later extended phenotype concept.
Dawkins, Richard. (1982). The Extended Phenotype: The Long Reach of the Gene. Dawkins expands his evolutionary theory to show how genes' effects extend beyond organisms' bodies to influence their environment and behavior, including examples like beaver dams and bird nests as extended phenotypes.
Blackmore, Susan. (1999). The Meme Machine. Builds on Dawkins' concept of memes, exploring how ideas, behaviors, and cultural units replicate through imitation. Relevant for understanding how language and ideas spread through human and artificial networks.
Nature Biomedical Engineering. (2023). Editorial on Large Language Models. Academic editorial noting the transformative impact of LLMs, highlighting how "it is no longer possible to accurately distinguish" human-written text from AI-generated content.
Fomin, Ivan. (2019). "Memes, genes, and signs: Semiotics in the conceptual interface of evolutionary biology and memetics." Academic paper connecting memetic theory with semiotics, providing a framework for understanding how signs (including language) function as replicating units across human communication systems.
Key Thinkers
#RichardDawkins #WilliamSBurroughs #NickBostrom #SusanBlackmore
Conceptual Deep Dives
#PosthumanFuture #DigitalDarwinism #TechnologicalExtension #CognitiveExternalization #AISymbiosis #MemeticEvolution #LibidinalEconomy #DigitalFitness #IntelligenceExplosion #AlgorithmicSelection
Cultural & Academic
#TechPhilosophy #EvolutionaryPsychology #Biosemiotics #AnthropoceneEvolution #TechnoSapiens #AIAlignment #EvolutionOfDesire #InformationEcology #DigitalPhenotype
Contemporary Relevance
#FutureOfHumanity #AITransformation #TechAndSexuality #DigitalEvolution #ReproductiveFutures #CognitiveProsthetics #HumanMachineSymbiosis #EvolutionaryLeap #MindExtension #SiliconConsciousness