Distributed Intelligence

Intelligence is not the monopoly of the brain. It unfolds in bodies, machines, and living environments. Faced with the fantasy of an omniscient AI, let us instead cultivate ecosystems of small intelligences in dialogue.

Questioning the brain-king, from living beings to machines

We have traditionally associated intelligence with the human or animal brain, following the Western conception of the distinction between body and mind, an image that seems perfect and natural to us, corresponding to the visible separation between head and body. If it has always seemed obvious to us that our brain commanded our body, with our mind and soul housed inside, it was actually our preconceived notion of things that gave us this vision, but by no means an established scientific reality.

This traditional vision has been challenged by neuroscience for quite some time now. We now know that our intestine, for example, is filled with neurons, which researcher Michael Gershon called “the second brain” in his pioneering work in 1999. Beyond even neurons, our microbiota directs our food consumption much more than our neurons do, whether they are in our brain or in our intestine. As Giulia Enders emphasizes in The Discreet Charm of the Intestine (2015), “we are a walking ecosystem”.

The experience of a week-long fast allows us to experience this obviously, an experience I practice every year. During the fasting week, stopping food intake triggers the phenomenon of cellular autophagy (cells “clean themselves” by degrading their damaged or useless components). Among this great cleaning that takes place in the body, the intestinal microbiota undergoes a profound rebalancing: bacterial diversity increases, beneficial bacteria regain a better place and pathogenic strains reduce in number, thus allowing a form of “renewal” of the intestinal terrain. After this week, the body has acquired a new intuitive “intelligence,” which pilots itself, it no longer asks for the same foods, particularly those toxic foods we were used to but couldn’t, even with efforts of will, remove from our diet.

We can speak of intelligence, even in the absence of neurons, because there is a modality of “reasoning” or algorithm, a programming of organs, desire, need, a mobilization of our body’s entire machinery that escapes our conscious control, which is managed by the rebalancing of our inner ecosystem.

Unicellular or simple multicellular organisms, devoid of brain yet capable of complex adaptive behaviors, illustrate a form of distributed intelligence in the metaphorical sense. One could object that this is not intelligence in the strict sense, their behaviors being determined by biological regulations rather than neural learning. Nevertheless, their responses to the environment, often non-linear and partially unpredictable, testify to an adaptability that serves their survival and broadens our understanding of what intelligence can be in living beings.

Automatic robots without a central brain also reveal a form of intelligence in their emergent behaviors. Simple robots, equipped only with local reflexes, can manifest complex behaviors, and even a form of “life,” without any centralized control. The “Robotic Ray” experiment (video at the top of this article), presented by its designer Sung-Jin Park in the journal Science in July 2016, almost 10 years ago, shows an artificial creature, a small ray, half-machine half-living, that lives and moves autonomously in water. The gold skeleton and elastomer body are associated with 200,000 living cells genetically programmed to react to light, draw energy from it and produce the contractions necessary for swimming. For now, this “being” follows the light. If we equip it with more complex sensors and algorithms, it will be able to do many other things...

My philosophy lessons with drones

My numerous experiences with drones starting in 2013 made me concretely discover this reality. Drones are computer objects packed with sensors and algorithms, designed to pilot their four propellers, to be able to stabilize and move, automatically taking into account the specificities of the environment that surrounds them. They are completely devoid of intelligence in the neuronal sense of the term. Yet, I have experienced many times their surprising and singular behaviors, linked to their form of reasoning programmed by human beings but confronted with the complexity of reality.

Depending on wind, sun, reflections, what the vertical camera perceives, altitude, surrounding obstacles, speed, drones can have behaviors that completely escape us. We find ourselves facing an entity with unpredictable behaviors, because the context of the world is infinitely rich. Even if the algorithms seem mastered, it is impossible to predict all the situations in which the drone will find itself.

I remember a particularly revealing experience. My drone, above a street, was trying through its algorithm to remain fixed by spotting elements below it: the uniform background of asphalt, houses, signs, parked cars. It was trying to maintain exactly the same relative position in its coordinate sensor. But drones from about ten years ago, I discovered to my cost, did not have the ability to distinguish a stopped car from a moving car. Thus, my drone followed cars, constantly correcting its position so that the car would always remain in the same place in its sensor, believing it was staying still! Result: my drone was running away, dragged by passing cars, and it was extremely difficult to get it out of this erroneous reasoning. Same above water, this same drone would race without being able to stop it, to the opposite shore of the lake.

From centralized super-brain to distributed intelligence

The great fear, quite legitimate, concerning the deployment of artificial intelligences is that of generalized surveillance, total control, a sort of super-brain possessed by American multinationals who would decide everything in our place and insert themselves ever closer to our brain, our lives, our gestures, our perceptions.

But this super-brain, if it exists technically in the form of very large centralized models, remains a fragile construction. It is extremely energy-costly, depends on massive infrastructures, and remains disconnected from what it claims to pilot: bodies, environments, reality in movement. This “top-down” conception recalls a first stage of robotics: a central brain charged with commanding robots packed with sensors. The results proved unconvincing. These machines were energy-consuming, slow, fragile and not very autonomous.

The real breakthrough came from a second approach: the dissemination of intelligence in each of the robot’s organs, its foot, its arm, its joints. Each part dialogues locally with the others, producing emergent coordination, more flexible and more robust. It is this principle of “distributed intelligence” that made possible much more efficient robots, capable of saving energy and resources.

This logic, far from being a simple metaphor, corresponds to the reality of living beings. Our body is not a monolithic unit piloted by a single center, but an immense ecosystem. Spinal reflexes, autonomy of the digestive system, hormonal regulation, microbiota activity: so many “local intelligences” that cooperate with the central brain without being totally subordinated to it.

The vision error consists in believing that piloting would come from a single sovereign place. In reality, the efficiency and resilience of a being, like a technical system, depend on the distributed diversity of its intelligent instances and their permanent interactions.

The “centralized super-brain” is therefore not a solid long-term perspective: it is technically possible but structurally limited. The future of AI, like that of robotics, will be played out more in the articulation between local intelligences, connected but autonomous, than within a single oversized machine.

Woody Allen had staged humorously, in Everything You Always Wanted to Know About Sex But Were Afraid to Ask (1972), the idea of a brain orchestrating a body like a factory full of departments. A very funny image, which reflects this centralized vision, and ultimately questions its inefficiency very well.

Towards sovereignty through disseminated intelligences

Large language models (LLMs), made famous by ChatGPT in November 2022, were initially centralized systems. You had to access this very large brain remotely to benefit from it. But gradually, and almost from the beginning actually, via software like LLM Studio or Ollama, you can download a language model to your own computer. You can run artificial intelligence locally, question it, dialogue with it, ask it to generate texts or images without connection to a central server.

Certainly, the power, that is, the degree of sophistication of the model, depends on the power of the local computer. Models running on our personal machines have more limited intelligence than those hosted in large data centers, these very powerful infrastructures but also very energy-consuming and ecologically destructive.

There is therefore a form of initial disappointment: sentences are less well formed, fewer parameters are available, reasoning seems more basic. But in reality, if we take the image of the robot’s body or the human body, we don’t need the same form of intelligence everywhere. The intelligence of a kneecap or an intestine is not of the same nature as that of a brain that manufactures language and conversations. As Antonio Damasio writes in Descartes’ Error (1994), “the body and brain form an integrated organism”.

What seems to me to be the future of artificial intelligence, in a perspective of sovereignty freeing us from dependence on large American multinationals, is a conception that I propose to name “disseminated intelligences.” It is not about seeking a superior form of central intelligence, but about cultivating ecosystems of small intelligences, perhaps each limited, but which, in their mutual, horizontal and dialogical functioning, can produce an efficiency superior to that of a big intelligence that would believe it could master everything.

This approach echoes the concept of “stigmergy” developed by zoologist Pierre-Paul Grassé (1959) to describe how termites build their nests: no central plan, but emergent coordination through the shared environment. To be more precise, it’s a mechanism of indirect interactions that allows the coordination of individuals through traces left in the environment. For termites, everything begins when a worker makes a mud ball, impregnates it with pheromones and deposits it in a place. Another termite, attracted by this chemical trace, will come to deposit its own ball in the same place or just next to it. This process, repeated countless times and without central planning, results in the construction of remarkable structures such as pillars, arches and galleries, which gradually form the collective nest.

The particularity of this mode of organization lies in its simplicity: each individual acts according to local rules, stimulated only by the state of the environment, without direct communication or leadership. This makes the system extremely robust, adaptable and scalable. Successive actions tend to reinforce and amplify each other, giving birth to an edifice whose global coherence is never explicitly programmed, but results from the aggregation of individual choices.

This principle is observed in other social insects, like ants, which also use pheromones to mark paths between food sources and their colony. In sum, stigmergy allows a colony to collaborate effectively through interactions mediated by the environment, giving rise to complex architectures and emergent collective intelligence, today a source of inspiration for human organizational models and decentralized digital systems.

I allow myself to redisplay here the “Game of Life” invented by John Conway in 1970, which functions on the same principle as ants or termites, but in a computer. It is as I had programmed it in the early 1980s on my small personal computer. I develop details on this subject in the article Already organic machines.

Experimenting and playing with micro-intelligences

I believe we really need to play, experiment and explore the possibilities of these disseminated intelligences. Get out of this capitalist headlong rush that always seeks more mastery, more power, this myth of infinite growth. Because this is not a natural functioning but demiurgic, utopian and destructive.

I therefore invite, particularly in the educational field, not only to learn to best use the generative artificial intelligences on the market, but also and especially to install them on one’s own computer, to play with objects like the Raspberry Pi, these mini-computers that can be embedded in small objects or even in clothing. We can equip them with sensors, motors, algorithms and artificial intelligences, certainly limited, but capable of dialoguing with their environment.

Cultivating the humility of small networked intelligences

We must cultivate this critical distance and this knowledge that intelligence is not a superior abstraction, a sort of new god, but rather a relevant function to integrate everywhere, at all levels, in all elements of bodies, whether they are machine bodies, living bodies, or even the social body.

Let us cultivate disseminated micro-intelligences without asking each one to be “superior,” rather than a supposed great intelligence, which would probably prove much more limited than the set of small beings in relation to each other. This approach advocates constructive humility rather than belief in total mastery, and a culture of connection and dialogue that founds the environments forming life itself.

As Gregory Bateson wrote in Steps to an Ecology of Mind (1972): “The unit of survival is not the organism nor the species, but the flexible system organism-in-its-environment.” It is this systemic and disseminated vision of intelligence that we must today embrace, to build a more resilient, more democratic and more living technological future.