In this conversation, Claire Isabel Webb and N/ Katherine Hayles discuss the ideas in Hayles’ new book, From Bacteria To Ai: Human Futures With Our Nonhuman Symbionts, 2025.

Webb: You've long used the word symbiotic, a term rooted in 19th-century biology, to describe flourishing interactions between organic creatures. Lynn Margulis argued in 1967 that we should replenish biology’s narrative of competition and mutation with concepts of cooperation and mutuality that also drive the emergence of new kinds of life. You want to apply the concept to computational entities. Why transpose it—and not only to nonhuman organisms but to nonbiological ones?

Hayles: Computational media evolve. Their cycles are much faster than those of organic things, but both are oriented toward forms of cognition. I propose that we humans are in a symbiotic relationship with computational media. I'm thinking not only about computers but about transistors, chips, and fiber optics. Computational media have completely interpenetrated our technological infrastructure, from flying airplanes to rail travel to the electrical grid. They have the cognitive capabilities to control and direct many other kinds of technology.

Webb: Symbiosis implies mutual flourishing. We have evolved to be deeply entangled with computational media—but how are we promoting the flourishing of their lifeworlds?

Hayles: We create them, we turn on the electricity for them, we repair them, and we continually improve them. Computational media wouldn't exist without humans. That's what astrobiologist Sara Walker means when she says computational media are life: because they derive from life, they must be counted in the evolutionary chains of life.

Webb: But such existence isn't exactly flourishing. How can a non-organic entity flourish if it lacks the capacity to enjoy itself, to avoid pain, to form emotional relationships?

Hayles: For one thing, the rapid evolution of computational media from 1945 to the present has been phenomenal. The chips get smaller and smaller. We're now approaching the absolute limit of the diameter of the silicon atom. We're experimenting with quantum computing. And we continually produce more of them.

Webb: Through symbiosis, what will human life become with machines? What will microbial life become with machines?

Hayles: There's an installation at the MIT Museum, "AlterEgo," a wearable interface system that uses sensors to pick up a person's inner vocalizations and then uses bone conduction in the inner ear to interact. What interested me about it was the way it moved computational media deeper into the body. Information moves from the air to the surface of the body and then inside it. Such an interface goes dramatically beyond something like a GUI, because it's more intimate, more emotionally persuasive. To feel the vibrations in your inner ear is an entirely different mode of rhetorical exchange than to hear it through your outer ear.

The evolution I foresee is a closer and closer integration of computational media, opening more intimate possibilities for exchange. The evolutionary trajectories of humans and computational media are already completely entwined. As William Gibson is credited with the famous remark, "The future is already here, but it's not evenly distributed."

Webb: Donna Haraway's Cyborg Manifesto presents an ironic, un-innocent figure essential to feminist confrontations with the phenomena you're describing. She viewed material and technological cultures producing, and produced by humans’ political, and technological enmeshment. “By the late twentieth century,” Haraway wrote in 1985, “our time, a mythic time, we are all chimeras, theorized and fabricated hybrids of machine and organism.”

As we become more and more cyborgian, what will become of our fleshly bodies?

Hayles: Bodies we categorize as disabled—for vision, for hearing, and so on—could really benefit from an evolutionary trajectory that incorporates computational media. Think of all the technologies that could be leveraged to accommodate particular ways of learning, seeing, and interacting with the world.

Webb: The title of your 2025 book is From Bacteria to AI. How do you imagine symbiosis among the microbiome, or the planetary biome, and computational media, including AI?

Hayles: The biologist Michael Levin and his colleagues at Tufts University are experimenting now with opening direct channels of communication with engineered organisms. That means communicating with them at the cellular level, not the organism level. If we can communicate with cells, we can direct them in all kinds of ways beneficial to humans—to cure cancers, say, or stop floods, or open waterways. Scientists are also already using computational media intensively to communicate with other species, such as elephants, chimpanzees, and whales. I hope any such channels would be two-way: it wouldn't just be us, it would be the whales talking back.

Webb: How do you define nonconscious cognition?

Hayles: Nonconscious cognition is a level of neuronal processing below consciousness. It has a faster response time than consciousness does. It can interpret information too noisy for consciousness to find any pattern in. It has a direct connection with body sensors and body systems, different from that of consciousness. And it is primarily non-linguistic, whereas consciousness, of course, is primarily linguistic.

Webb: When you say linguistic, do you mean symbol exchange? Would you say chimps and whales are conscious because they trade symbols?

Hayles: No—symbol use may be an indicator of consciousness, but an indirect one. Consciousness, for me, requires not only awareness but self-awareness: the sense that one has (or is) a self, which confers the ability to think about oneself as a self. Chimpanzees, for example, can interpret, understand, and use symbols, and tests have shown that they have a sense of self and a sense of having a mind. That said, the extent to which non-human animals use symbols and abstract thought pales in comparison to how humans use them.

This is almost oxymoronic to say, but we humans live in our consciousness more than we live anywhere else. Our identity is bound up with it. As a result, many people fail to recognize how important nonconscious cognition is to us on a daily basis. What we call intuitions are often nonconscious cognition speaking. Nonconscious cognition is not a new human capability—it's been around as long as the human species has. But only recently have scientists devised experiments to test what it can do, and its capabilities are quite amazing. It can recognize patterns in noisy data that consciousness has no clue about.

Webb: Such as?

Hayles: Say you have a very noisy visual image. There's a pattern in it; however, it's so subtle that if you ask an experimental subject to identify it, they're completely unable to. But they still discern the pattern nonconsciously.

Even in humans, cognition is a much broader capacity than consciousness. If nonconscious cognition can operate in humans, that implies nonconscious organisms—plants, bacteria, fungi—may have cognitive capacities even though they aren't conscious. That opens the entire biological realm to being considered as cognitive systems.

Webb: Will computational nonconscious cognitions evolve toward a different kind, or higher-order level, of consciousness?

Hayles: There are really two issues here. One is whether advanced AI, which in my view is at present cognitive but not conscious, will one day attain consciousness. The other is how human cognition, in rapid and intense feedback with AI, will change as synthetic intelligences are incorporated more and more into our daily lives.

One possibility is that we'll use AI to catalyze our own cognitive abilities, reaching levels of sophistication faster than would otherwise be possible. Another is that our use of AI will make us dumber and lazier—if we let AI substitute for our own native cognitive abilities rather than cultivating them. Capacities, mental and physical, that aren't used daily can atrophy very quickly.

Webb: What are the defects of consciousness? What would be the transcendence of those defects?

Hayles: Consciousness evolved so that we humans can make sense of the world. We constantly tell ourselves little stories—the attempts of consciousness to make everything explainable in reasonable terms, consistent with our past experiences.

We use interior monologues to suture our present experiences into past trajectories, so that our actions make sense to us as emanating from a single, unitary self. Neurodivergent people such as schizophrenics rely on these narratives even more strongly, as they want to explain why the entire world is enrolled in conspiracies against them. But even neurotypical people like me rely on the narratives consciousness spins every hour of every day to assure ourselves that we have (or are) stable, consistent, reasonable selves.

Ironically, consciousness's driving purpose is also its greatest limitation. Many experiments have shown that people will invent narratives to make sense of something they don't understand. Even when they're wildly off the mark, humans will fill in the gaps and create stories that are nothing more than confabulations. In other words, consciousness lies all the time, consistently and effectively.

By contrast, the body does not lie. When the hairs rise on the back of our neck, it's because the body senses danger—even though consciousness may be busy telling us, "It's all right, you can just keep walking down this dark alley, because you want to reach the other street."

The upside of consciousness's determination to make the world make sense is that it bestows a tremendous evolutionary advantage. It was probably the foundation for scientific inquiry: "If something has happened that doesn't make sense, I'll make sure it makes sense by changing the story."

Webb: Can you give me an example?

Hayles: Maybe you've seen the video of a man in a gorilla suit walking through a basketball court while people bounce a ball between them. About half of viewers won't see the gorilla. They edit it out of the picture because it's highly anomalous—but they can tell you how many bounces the ball made.

Webb: So would a transcendence of what we now call consciousness be an ability to integrate conscious and nonconscious signals? The basketball bounced ten times—and, by the way, a guy in a gorilla suit walked by?

Hayles: We're already on that pathway. The number of seconds it takes an audience to recognize a visual image has declined dramatically since the 1950s; now it's less than a tenth of a second. Why? Because people have been exposed to faster and faster images over the past half-century.

Webb: Can you explain how the lifeworlds of computational media overlap with humans'?

Hayles: The idea of an umwelt comes from the work of the biologist Jakob von Uexküll, who imagined the kinds of worlds non-human creatures have. His favorite example was the tick. In German, Umwelt translates to something like "world horizon," or, more literally, "world-surround." It's the sort of world a species constructs for itself—what its sensors, actuators, and chemical and neuronal processing allow it to build. A snake has infrared vision to detect heat sources, including prey. A dog hears frequencies humans don't, and can smell enormous numbers of odors we can't. Each species has its own world, specific to its capabilities.

To communicate, different species must have some degree of overlap in their umwelten. A porcupine signals a threat by raising its quills—a gesture that will only be meaningful to species with vision. That creature's umwelt evolved ejectable quills in an ecosystem full of sighted predators. Each species' umwelt is environmentally and evolutionarily determined; it was always in conversation with all the other kinds of umwelten around it. Cognition always happens in relation to environments.

Now let's go to computational media. There's enormous flexibility in how they construct their own umwelten, because humans determine what kind of sensors they have and the environment they work in. My desktop computer knows its data. It knows its algorithm. It has an internal clock. It has logic gates that allow it to interpret commands and algorithms. And it has outputs that connect it somehow to the world. Most computational media have sensors and actuators. They communicate with other computers. They see what other computers see, and they see what their own visual systems enable them to see. Consequently, they're in very rich sensory environments.

An example is an algorithmic docking system for spaceships. If you give the computer sensors and flexible programming so it can interpret and make decisions, you now have a much more adaptable and robust system. That's crucial for me, because to be counted as a cognizer, an entity has to receive information from the environment, interpret it, and then make choices or selections. If there's only one interpretation, then there's no cognition—it's a straight causal chain.

Webb: What are the boundaries of any computational medium's umwelt? A chip is a material assemblage that lets the computer gather sensory information. Flexible programming enables it to respond to the environment—which is, as you say, characteristic of any world horizon, regardless of material.

Hayles: As you suggest, there's no upper bound to computational media's umwelten, because they're designed and given purposes by humans. If we want to build a planetary surveillance system, then the umwelt of that system will be planetary in scale. If we want a cellular surveillance system, it will be microscopic in scale.

All life forms began immersed in their environment, simply to continue existing. Through eons of evolution, humans became capable of abstraction. Mathematics is a convenient example. Computational media began on the opposite side: the very first computers were already completely abstract machines. We evolved from immersion to abstraction; computational media evolved from abstraction to immersion.

Humans, like all biological entities, began with a desire to survive. Over time, we progressed to having designs and purposes other than survival. Computers originated instead with designs and purposes that humans gave them. Can they evolve a desire to survive? That would be a complete game changer.

Webb: What would be the phenomenological transition for AIs to gain intersubjectivity and evolve toward something we could call desire?

Hayles: Leif Weatherby has recently argued, in his book Language Machines, that LLM-generated texts have meaning even if the machines themselves are not intelligent or cognitive. Basically, this is because the semiotic web of language patterns they learn has human cognition embedded within it, so in effect LLMs are piggybacking on human cognition to achieve meaning.

As a literary scholar, I'm interested in the fact that LLMs can detect and reproduce a literary style. Literary styles are devices that embed social relations in verbal articulation. That means the LLM has made inferences about human social relations. It understands hierarchy. It understands privilege. It understands what it means to be a survivor.

Genre is the literary mode of constructing the kind of world in which the literary action can take place. LLMs can mimic this too. The LLM has inferred the rules that govern a genre's world-construction. If it can do that, how can you tell me it doesn't have meaning?

Webb: What was the moment that such rich content and context took on a reflexive meaning for the machine?

Hayles: In 2023, a team of Microsoft researchers invented a test whose solution was not retrievable online. They asked: “Here we have a book, nine eggs, a laptop, a bottle and a nail. Please tell me how to stack them onto each other in a stable manner.” Earlier models gave absurd answers, such as a vertical egg tower. But later, GPT-4 reasoned its way to suggest a stable arrangement. The book goes on the bottom because it makes a stable base; the nine eggs go on the book, spaced in a three-by-three grid because that distributes their load; the laptop goes on the eggs because its flat rigid surface won't crack them; and so on up to the nail at the top.

Webb: So, the later model was able to make incredible inferences about physical objects. How?

Hayles: Because GPT-4 can reason. Because it is increasingly using the word “because.”

Webb: Can LLMs produce meaningful content—that is, content meaningful to them as well as to us, natives of natural language?

Hayles: I believe they can.

Webb: Humans invent literary styles and then work within a given genre. Could AIs invent a new style humans wouldn't have arrived at otherwise? Invention is different from mimicry.

Hayles: If you connect two data points that have never been connected before, you produce new knowledge. For example, AIs could correlate all the people in the world who get a certain very rare disease with their environments, their diets, and so on. Their pattern-seeking capabilities dwarf ours.

Humans have explicit knowledge—knowledge we know and can talk about. But connecting two things that have never been connected before generates implicit knowledge. So if we marry implicit knowledge with explicit knowledge, we generate an enormous possibility space. That's the realm in which computers, or AI, can collaborate.

There is also the space of potentiality. This process isn't merely connecting up and building on what's already known; it can generate what has never been thought before—Einstein's theory of relativity, for example. I think AIs cannot reach the space of potentiality. Humans can, and do, all the time. So there's an aspect of human intelligence that is fundamentally different from artificial intelligence: the ability to think the radically new or radically different.

Webb: Let's go back to cognition. You're claiming that we can use the word cognition for humans, biological entities, and computational media. It seems to me you're more concerned with how things function in relation to one another than with the materiality that enables those relations. You put emphasis on "family resemblances"—you're interested in the process of synapses firing, whether that synapse occurs in silicon or in carbon.

Hayles: Homology clarifies family resemblances. (I prefer homologies to metaphors, because the former are constrained to empirical resemblances—functional, structural, or both—while the latter are not.) Neural nets and artificial media are homologies of biological neurons, because humans engineered them to be that way.

There are exciting experiments with neuromorphic chips. When chips become capable of the kinds of correlations that computational media can now make, the entire computer will reduce to the size of a tiny chip—and you'll approach something like the neural density of the human brain.

Another near-future direction is to give neural nets bodies, so that computational media could make inferences from their experiences in the world. If GPT-4 can already reason about how to build a stable structure from an egg and a book, think what it could do with actual input from clearing tables, walking downtown, petting a dog, or anything else you can imagine.

Webb: You're saying cognition is the thing that biological and nonbiological entities can do. But what difference does matter make? How much does matter matter to the construction of different kinds of cognition?

Hayles: Absolutely, matter matters. It matters that the umwelten of LLMs are composed entirely of textual signifiers, with no intuitive connection to the physical world embodied humans live in. As a result, LLMs often fail at tasks humans would find trivial, such as locating objects in three-dimensional space. They can and do make obvious mistakes. I call this the systemic fragility of artificial intelligence. AIs have only very structured ways of knowing the world; they need to explore more. Experimentation is endemic to humans. You can see it in babies: already they're experimenting, exploring their environment, trying to figure out new things. That's something an artificial intelligence can't do—spontaneously explore.

Webb: Animals—including humans, and especially mammals—learn a lot through play. Play is a way to test, and then find, social boundaries. George Lakoff observed that for two wolf pups a bite is a playful act that establishes social conventions. What would be the ingredients of a platform in which LLMs could "spontaneously explore," to use your phrase? How important is embodiment for an LLM integrated with a robot—one that never underwent a social evolution deeply tied to its physical one?

Hayles: Linking a robot body to an AI—a large language model, say—could theoretically give the AI a chance to explore its environment spontaneously. I should emphasize that this kind of experiment is still in the early stages, such as the "SayCan" robot that can fetch a sponge to clean up a spill on a countertop. Nevertheless, robotic embodiment would certainly give an AI a much better sense of what it means to move in a three-dimensional environment. Even so, as your comment suggests, there would still be massive differences between human embodiment—evolved in coordination with our environments—and robot bodies with a completely different origin story. Embodiment matters, and embodied material entities should always be seen in conjunction with their environments and evolutionary histories.

Webb: How We Became Posthuman was a wonderful inspiration for me to think about feminism entwined with technology and cyberspace. The book came out in 1999, but I still think of it as prescient. You wrote in your prologue, “What embodiment secures is not the distinction between male and female or between humans who can think and machines which cannot. Rather, embodiment makes clear that thought is a much broader cognitive function depending for its specificities on the embodied form enacting it”.

You are part of a coterie of feminist scholars—Donna Haraway and Rosi Braidotti come immediately to mind—who insisted on the necessity of embodiment to cultivate a novel philosophical subject. As you write, the Macy conferences of the 1940s and '50s marked the apotheosis of the cleaving of information from knowledge production—a cleaving that depended on the evacuation of various embodiments. It's 2026. Since How We Became Posthuman came out, how has your analysis of embodiment evolved?

Hayles: "Computers having bodies" is a metaphor. It would be more correct to say that computers are instantiated, and that they may be instantiated in radically different forms. The material instantiation of any system is completely bound up with the kinds of cognition it can have.

Webb: So the material substrate facilitates the uneven translation of symbols, bits, words, and concepts into—as you argue—meaning. Take a non-human thinking machine. How does a silicon instantiation of cognition handle a thought, or create meaning, or know what "because" is? How is that different from the material substrate of human corporeality—that is, human embodiment?

Hayles: The huge difference is that human embodiment includes emotions mediated through internal organs, through the limbic and endocrine systems, and so forth. The endocrine system—adrenaline, fight or flight, and the rest—has a deep evolutionary history stretching back to the very beginning of the species. AIs couldn't possibly have that particular kind of deep evolutionary history. A computer's desire to survive, by contrast, would have to be mediated through the same inferential networks that enable it to have a thought at all. And so it wouldn't feel the way we feel.

I think that for a computational system to evolve a desire to survive, that desire would have strong inferential intensity. It would not be emotional intensity, as it is for humans.

Webb: As one of the theory's founding mothers, what do you mean by posthumanism? What's the hinge point between being human and being posthuman?

Hayles: If I had to pick a date, I'd mark the beginning of posthumanism in 1950, when post–World War II technological innovations accelerated, especially with computers. I chose the title How We Became Posthuman to foreground that "the human" has always been a contested category. As far as I'm concerned, the human is a historical construction; it has varied through time. I doubt there's any essential quality one could ascribe to being human that would last throughout time. So to say that we are already posthuman is a way of saying that a certain kind of historical construction is falling apart, and a new one is emerging.

Webb: You wrote in the 1999 book, “...Human being is first of all embodied being, and the complexities of this embodiment mean that human awareness unfolds in ways very different from those of intelligence embodied in cybernetic machines.” And how do you define posthumanism?

Hayles: Posthumanism takes all the vaunted qualities of the human that governed in the Enlightenment and deconstructs them. Instead of free will, you have distributed agency. Instead of the lone rational thinker, you have cognitive assemblages. Instead of the soul, you have the construction of artificial media. It's a deconstruction of the old ideas and the emergence of new configurations around the technological.

Webb: With the development of LLMs, soft robotics, and brain implants, how has the posthuman condition deepened in the last couple of years?

Hayles: The emergence of LLMs that can produce natural language is a huge, huge win.

Webb: And what would mark the end of posthumanism, as you currently define it as a historical structure?

Hayles: That's an interesting question, Claire, because for centuries—maybe millennia—"human" has been a valorized category. Stigmatized populations have aspired to emphasize that they too are human. Now I see a movement in the opposite direction. I see scholars such as Bruno Latour wanting to stigmatize the category of the human. His name for the replacement term is "Earthbound": in his telling, the humans are the bad guys and the Earthbounds are the good guys. Humans are denigrated. But do we really want an ideology that distinguishes Earthbounds from humans?

Webb: I'm drawn to posthumanism because it gives us a critical framework for investigating those boundaries. Things get really fuzzy at the edges. Animals have made the category of the human fuzzy. Octopuses make what we describe as human intelligence fuzzy. And today, computational media are making post-Enlightenment subjecthood blurry beyond belief. How might computational media aid in interpreting alien inner subjectivities, from machines to macaques? Could they illuminate "what it is like to be" a non-human—say, a bat?

Hayles: AI has also been used to facilitate interspecies communication. This would be a huge leap forward in understanding the umwelten of other creatures. 

As you say, the boundaries are getting fuzzy, but the realities of material embodiment place an upper bound on how far that fuzziness can proceed. Even with all our computational media and technological appendages, we are still only and always in the human umwelt. In fact, we can’t be anywhere else. Umwelten can and do change, and the human umwelt has transformed radically  over the millenia.  But there are still contemporary perceptions that would largely coincide with what cave people saw—a tree, grasses on a savannah, and so forth.  

My hope is that we can use our enhanced umwelt not only better to understand the creatures with whom we share the earth, but also to grasp on a deep level that our perspectives are not reality itself, only the world we construct for ourselves. That would be the beginning of wisdom, and a pathway to understand that our cognition, like all cognition, is relative and relational. Then perhaps we can overcome anthropocentrism and understand humans not as the species superior to all the others, but as embedded in the web of life, without which we cannot survive.