Ant colonies, bee swarms, and the case for collective intelligence
Dear Reader,
I have been writing for months about minds in different substrates. Machine minds, animal minds, plant minds. What I have not yet written about, and what belongs in the picture, is the possibility of collective minds. Minds that exist not in a single body but in the coordinated activity of many small ones. Ant colonies. Bee swarms. Termite mounds. Slime mold populations. These are systems where individual organisms behave in specific ways, and where the aggregate behavior of the whole exhibits properties that no individual has.
Whether these collectives count as minds is disputed. Whether they exhibit intelligence is not. The distinction between having intelligence and having a mind is exactly the distinction I have been circling around since the first piece in this series, and superorganisms are perhaps the sharpest place to see it.
Let me walk through what these systems actually do, what makes them philosophically interesting, and what they tell us about the possibility of intelligence in AI systems that are themselves collectives of smaller processes.
Start with what a superorganism actually is. The term was coined by William Morton Wheeler in 1911, applied to ant colonies. His point was that a colony behaves in ways that are not reducible to the behavior of individual ants. Individual ants follow simple rules. They respond to local chemical signals. They perform specific tasks based on their caste and the situation they encounter. Individually, an ant is not particularly smart.
But an ant colony as a whole solves problems that no individual ant could solve. Colonies find the shortest path between food sources and the nest. They optimize the distribution of foraging effort based on food availability. They regulate the internal temperature and humidity of the nest. They manage caste allocation based on colony needs. They coordinate defense against predators. They make collective decisions about when to migrate, where to build, and how to allocate resources.
The mechanism by which they do this is what biologists call stigmergy, from a Greek word meaning "mark from work." Individual ants leave chemical trails, and the trails influence the behavior of subsequent ants. When many ants interact through these trails over time, complex patterns emerge without any central coordination. There is no leader. There is no plan. There is only the aggregate effect of many small local decisions, and the aggregate produces sophisticated collective behavior.
This is not just an interesting curiosity of ant biology. It is a specific solution to the problem of distributed coordination that biology has discovered. The same basic principles show up in bees, termites, wasps, slime molds, and even in some bacterial communities. Different substrates. Different scales. Same fundamental pattern. Local interactions producing global intelligence.
Let me tell you what specifically strikes me about these systems.
The first thing is that the intelligence is real. Ant colonies, treated as decision-making systems, solve problems that would take substantial computational resources to solve algorithmically. Route optimization, resource allocation, defense coordination, response to environmental change. These are hard problems, and colonies solve them at scale with no central processing, no explicit algorithm, and no obvious way for any individual ant to know what it is doing.
The second thing is that the individual components are simple. An ant, considered on its own, is a small nervous system responding to local cues. There is no interesting cognition happening in the ant. What is cognitive, if anything is cognitive, is happening in the aggregate. This is a case where the whole is emergently more than the parts, in a way that is both technically impressive and philosophically interesting.
The third thing is that the behavior is genuinely responsive. A colony is not just following a fixed program. It adapts to novel situations. It learns, in a distributed sense, over the lifetime of the colony. Old colonies solve problems differently than young colonies. Colonies in specific environments develop specific patterns of behavior that suit those environments. All of this happens without any individual ant knowing what the colony is doing.
The fourth thing, and this is the philosophically loaded one, is that there is a real question about whether the colony itself is having any experience. Individual ants presumably have very limited experience, if any. But the colony processes information, responds to its environment, and behaves in ways that in a single organism we would associate with cognition. If we ask whether there is anything it is like to be a colony, the question is not obviously nonsensical. It might have no meaningful answer. But it might not be dismissable either.
The specific case of honeybees deciding where to build a new nest is worth walking through, because it has been studied in extraordinary detail and it reveals something specific about how collective decisions work.
When a bee colony grows large enough, part of the colony splits off with the old queen. This group needs to find a new location for its nest. Scouts fly out and investigate potential sites. They return to the swarm and communicate their findings through the waggle dance, a specific movement pattern that encodes the direction and distance to the site.
But here is the interesting part. The scouts do not just describe what they have found. They advocate for it. A scout who has found a good site dances more vigorously and for longer than a scout who has found a mediocre site. Other bees observe these dances and either fly out to investigate the advertised sites themselves, or join in dancing for a site they have already visited and evaluated.
Over hours or days, the swarm collectively converges on a decision. The dances for the best site become more vigorous. The dances for worse sites fade away. Eventually a critical threshold is reached, and the whole swarm launches together to move to the chosen location. The choice is genuinely collective. No individual bee decides. The swarm as a whole decides, through a process that has been called consensus democracy by the entomologists who study it.
What makes this remarkable is that the process reliably selects the best available site, according to specific criteria that we can measure. Cavity size, entrance orientation, proximity to water. The swarm does not just make a decision. It makes a good decision. The mechanism by which many small local behaviors aggregate into optimal global choice is one of the more elegant results in behavioral biology.
And it is happening without any central processor, without any planner, without any individual bee knowing why. The intelligence is genuinely in the collective.
Let me get to the philosophical implication, because it is direct and it applies to what I have been writing about.
If intelligence can exist in a collective without being present in any of the individual components, then the question of what has intelligence is more complex than the standard framing allows. We usually assume that intelligence is a property of specific systems, brains or brain-like structures, and that it is either present or absent depending on whether the system in question has the right architecture.
Superorganisms show that this framing is at best incomplete. The colony is intelligent. The individual ants are not, or are only in a very limited sense. Intelligence is a property that emerges from the specific pattern of interactions among the ants, not something located in any one of them. Which means intelligence can be present at multiple levels of organization simultaneously. An ant may have some minimal intelligence at its own level. The colony has substantial intelligence at its level. These are different intelligences, in different places, and they coexist without one being reducible to the other.
For AI, this suggests that we should be thinking about intelligence not just as a property of individual systems but as a property of ensembles. Multi-agent AI systems, which are becoming increasingly common, may exhibit collective intelligence that is not present in any of the individual agents. Large language models orchestrated in complex pipelines may collectively do things that no single call could do. Networks of AI systems interacting through shared protocols may develop emergent behaviors that were not designed and that no individual system explicitly performs.
Whether any of this rises to consciousness is a further question. What is clear is that intelligence, in the sense that matters for behavior and problem-solving, is a property that can arise at collective levels. The interesting future of AI is not just building single systems that are smarter. It is understanding how systems combine into larger systems that exhibit properties none of them have alone.
Let me tell you about a specific moment when this became vivid for me.
I was reading a paper on ant colony behavior at the same time I was designing a multi-agent AI pipeline at SECJUR. The pipeline was supposed to review legal contracts. Multiple specialized AI agents, each responsible for a specific aspect of the review, would coordinate through a shared workspace to produce a final assessment.
At some point during testing, the system produced an output that I could not trace back to any single agent. The final assessment included an observation that none of the individual agents had explicitly generated. It had emerged from the interaction. One agent had noted a specific clause. Another had noted a specific case law reference. A third had triggered a specific risk category. And the combination, filtered through the workspace, had produced a synthesis that was genuinely new.
I traced through the logs carefully. Each agent had produced its own output. Each output was traceable to its own reasoning. And yet the final synthesis was not present in any of the individual outputs. It was the specific interaction pattern that produced it.
I do not want to overstate this. The synthesis was probably present implicitly in the combined outputs, and a more careful analysis might have shown that it was inevitable given the inputs. But the experience of watching a multi-agent system produce something none of its parts explicitly contained was, functionally, an experience of collective intelligence in a system I had built. And it shifted my thinking about what these systems can do.
If intelligence can emerge from the interaction of components that individually do not have it, then AI systems are structurally similar to biological superorganisms. Not the same. But similar in the specific respect that matters. The intelligence is a property of the whole, not of any of the parts. And building intelligence, in this framework, is about designing the interaction patterns, not just designing the individual components.
There is a specific philosophical question that superorganisms sharpen, and I want to name it.
If a colony is intelligent, is it conscious? This is the question that would make the intellectual stakes of studying superorganisms very high, if it could be answered. Individual ants have some minimal experience, probably. Colonies process information at a level of complexity that would justify consciousness talk in a single organism with the same processing. So the question is whether the colony as a whole has some form of collective consciousness that is emergent from but distinct from the consciousness of the individual ants.
Nobody knows how to answer this. The theories of consciousness I have written about vary in their predictions. Integrated Information Theory would say the colony has some level of phi that reflects the specific integration of the components, and this phi might or might not be greater than the sum of the individual ant phis. Panpsychism would say the colony has some form of experience that is emergent from but not identical to the experiences of the individual ants. Illusionism would say the colony does not have any experience, but neither do the ants, and the question is confused. Physicalism would say we need to know more about the neuroscience.
What is clear is that the question is not settled by our usual assumption that consciousness is a property of individual brains. Superorganisms are a case where the individual brains are simple, and the interesting behavior is happening at a scale that no individual brain reaches. Whether consciousness scales with intelligence, or whether it is tied to individual brains regardless of the intelligence they produce, is exactly the question that superorganisms make impossible to avoid.
For AI, this maps onto the question of whether an AI collective is conscious in a way that its individual components are not. If we build networks of AI agents, and those networks exhibit intelligence that no agent has, do the networks have experience that no agent has? This is not a question we are anywhere near answering. It is a question we should probably start taking seriously as multi-agent AI becomes more common.
I want to close with something that has been growing in my thinking as I have written this newsletter series.
For most of the last few centuries, the dominant Western framework has treated minds as concentrated in individual organisms, and specifically in the brains of complex vertebrates. This framework has been productive scientifically. It has also, I think, been narrower than the phenomena it was supposed to describe.
What the accumulation of research on octopuses, corvids, plants, whales, and superorganisms suggests is that mind, or something functionally like mind, is more widely distributed than the individual-brain framework can accommodate. It shows up in different substrates. It shows up at different scales. It shows up in different configurations. The specific human version is one instance of a much broader phenomenon that we are only starting to understand properly.
This does not mean human consciousness is not distinctive. It probably is, in ways that we are still figuring out. But being distinctive is different from being unique. The universe has been experimenting with intelligence and possibly with consciousness in many different ways for a very long time. We are one experiment among many, and we are now conducting experiments of our own with AI. The scientific and philosophical work of the next century is going to be, in large part, understanding how all of these experiments relate to each other.
This is where I have been heading through this cluster of articles. The specific cases keep showing that the space of possible minds is larger than we thought. Machine minds are one region of that space. Biological minds are another. Collective minds are a third. Which regions overlap, and how, is where the interesting work is.
Cluster B of this series is now complete. In the next cluster I want to shift focus. Instead of writing about the range of possible minds, I want to write about what is happening to human minds under the specific pressures of modern technology. The attention economy, dopamine loops, algorithmic influence, and the collapse of shared reality. These are the specific ways our own minds are being shaped by systems we built. And they are the context in which any conversation about AI has to happen. Stay with me.
— Transmission Sent —
Niklas Hanitsch
Reference materials
- Deborah Gordon — Ant Encounters: Interaction Networks and Colony Behavior (2010)
- Thomas Seeley — Honeybee Democracy (2010)
- E.O. Wilson and Bert Hölldobler — The Superorganism (2008)
- Iain Couzin — Collective Behavior in Animals and Robots (Annual Review of Ecology, 2018)
- Peter Miller — The Smart Swarm (2010)
- Guy Theraulaz et al. — Swarm Intelligence in Insect Societies
- https://en.wikipedia.org/wiki/Superorganism
- https://en.wikipedia.org/wiki/Swarm_intelligence
Continue reading
- The octopus knows something we do not
- Crows, ravens, and the intelligence we underestimated
- Do plants communicate? What the science actually shows
- Panpsychism is making a comeback, and here is why
- Emergence, and why consciousness might not need a cause
Frequently asked questions
What is collective intelligence? Collective intelligence is the intelligence that emerges from the coordinated activity of many individual components, without being present in any single component. Ant colonies, bee swarms, and slime mold populations exhibit collective intelligence in biology. Multi-agent AI systems and human organizations can exhibit it too. The intelligence is a property of the interaction pattern, not of any individual part.
What is stigmergy? Stigmergy is a mechanism of coordination in which agents interact indirectly through modifications to a shared environment. Ants leaving chemical trails is the classic example. Each ant leaves a trace, and subsequent ants respond to the traces, producing complex collective behavior without any direct communication or central coordination. Stigmergy is one of the main mechanisms behind biological swarm intelligence.
Are ant colonies conscious? Unknown. Individual ants presumably have very limited experience if any. Colonies process information at a scale and complexity that would suggest consciousness in a single organism. Whether consciousness can emerge from the coordinated activity of many small components, without being present in the components, is exactly the question superorganisms raise. Different theories of consciousness give different answers.
How do superorganisms make decisions? Superorganisms make decisions through processes that aggregate many individual small decisions into a collective outcome. Honeybees selecting nest sites do this through the waggle dance and reinforcement dynamics. Ants selecting food sources do it through chemical trail feedback. The specific mechanism varies, but the general pattern is many local decisions reinforcing each other until a collective consensus emerges.
What does collective intelligence teach us about AI? Collective intelligence shows that intelligence is a property that can emerge from interactions between components, not just from individual system architecture. This is directly relevant to multi-agent AI systems, which are becoming increasingly important. It suggests that the interesting future of AI may lie not just in building smarter individual systems but in designing interaction patterns that produce intelligence at scale.