EXPLORE / WHY FLOCKINATOR
Intelligence may live in an agent. It may also emerge through relationships. Flockinator brings those interactions into light and sound, so we can notice what is worth understanding.
A visual and auditory microscope for collective behaviour.
01 / WHY MAKE IT PHYSICAL?
A sudden synchrony. A call answered across the wall. One node that hesitates while its neighbours commit. A pattern that disappears when communication is interrupted.
We are pattern-seeking animals. Motion, spacing, contrast and rhythm give us ways to recognise relationships that are hard to follow across separate streams of logs. The wall is designed to put those senses to work: each node has a visible place and can have an audible presence.
That is why Flockinator exists. It turns an agentic system into something we can watch and listen to while AI explores strategies for individual objectives and for the flock as a whole. A surprising pattern becomes a question we can investigate.
Human perception is also fallible. A display can make unrelated events seem connected. The microscope helps us discover questions; recordings, comparisons and repeatable tests help us answer them.
How observation becomes evidence ↗02 / THREE LENSES ON INTELLIGENCE
Change the boundary you draw around the system, and a different kind of behaviour becomes visible.
THE INDIVIDUAL
An agent receives observations, uses a rule or policy to choose an action, and affects its environment. It may have memory and an objective. Its policy might be simple code, a learned model or a larger AI system.
What can this node know and do on its own?
INTER-INTELLIGENCE
One agent’s action changes what another observes or does. Messages, imitation, disagreement and trust can make coordination possible. They can also spread error. The connections are part of what we study.
What changes when these nodes interact?
EXTRA-INTELLIGENCE
Here, “extra-intelligence” is a working name for useful capabilities that extend beyond an individual agent: distributed across a group, its connections, shared environment and feedback over time.
What can the whole system achieve that one node cannot?
“Inter” and “extra” are lenses for this project, not a settled scientific taxonomy. Research often discusses these questions as collective intelligence, distributed cognition and emergence. The boundaries and mechanisms are still active subjects of inquiry. Couzin’s review of collective cognition ↗
Simple local rules can produce coordinated motion without any individual planning the whole shape. Reynolds’ boids model demonstrates this through separation, alignment and cohesion. A recognisable flock does not, by itself, tell us that learning has occurred. The original model ↗
Optimisation asks what improves: coverage, accuracy, energy use, resilience, or something else. Helping one node may hurt the group. Flockinator lets us look for the trade-offs and test whether an apparent collective advantage survives a different task or condition.
03 / THE PARTS OF THE MICROSCOPE
The controller holds the consolidated experiment view. The individual panels make the agents’ behaviour available to our senses.
Shared state, timing, objectives and a view of the whole experiment.
Local states, decisions and exchanges, expressed through light and sound.
Look for a motif, hear a rhythm, identify a surprise and mark the moment.
Observation → a testable question → comparison → a revised strategy → another run
Holds the simulation clock and world state, hosts the planner and learning lab, and provides the consolidated overview. The experiment defines what each simulated agent is allowed to observe. The controller’s complete view and a node’s limited view serve different purposes.
Build the controller ↗Gives each node its own processor, identity and visible position. The panel turns state and transitions into a legible signal; its ESP32 drives the display and interfaces with local peripherals. A physical node and an AI agent are related roles, not interchangeable terms: a simulated policy may run on the controller or host rather than on the panel.
Build an individual node ↗The speaker gives a node a voice; the microphones let it listen. This supports the intended auditory microscope and experiments with acoustic interaction. The selected firmware and sound mapping must be checked: audio output, audio sensing and a working node-to-node acoustic protocol are separate capabilities.
Explore sound and sensing ↗Onboard sensors, optional proximity or light sensors, and communication links can give nodes different observations and ways to influence one another. Panel borders currently provide communication cues. An optional ARGB/WLED grid can extend that visual language into the spaces between nodes.
Read the panel language ↗Node cards and controller archives preserve the records produced by the chosen experiment. They let an interesting moment be revisited alongside its configuration and decisions. A successful file receipt establishes storage, not learning or an intelligent outcome.
Follow the experiment workflow ↗Reliable power and a serviceable mounting system keep the experiment readable and repeatable. The gaps between panels preserve each node’s identity. Brightness changes, resets or failed connections can otherwise resemble changes in behaviour.
Plan the wall and its power ↗Implementation reference: the Korvo standalone controller documentation. The public benchmark results come from host simulations. The controller’s learning lab and physical-node acceptance are separate parts of the project.
04 / WHERE AI FITS
The research ambition is to watch AI develop and refine strategies for a flock and its individual nodes, while making the resulting behaviour perceptible. An optimiser could change a policy, communication rule, allocation of work or response to uncertainty. The experiment supplies objectives and constraints against which to compare those choices.
Keep three roles visible: the agents acting inside the experiment, the optimiser or model-assisted development process proposing changes, and the observer noticing patterns and deciding what to test next. They need not run on the same hardware or at the same timescale.
A proposed observation exercise, not a reported Flockinator result. Sound is a designed cue only when the experiment’s audio mapping is implemented and recorded.
The microscope is successful when it helps us find a better question—even when the next experiment shows that our first explanation was wrong.
From observation to the next run ↗05 / BEYOND THE FLOCK
Flocks and schools make coordination easy to see. Related questions arise in living tissues, adaptive networks and information left in a shared environment. The mechanisms—and the appropriate meaning of “intelligence”—differ.
In flocks, schools and insect groups, interactions can turn local information into group-level responses. A useful question is how information travels, and when a minority can guide—or mislead—the group.
Collective cognition · Couzin, 2009 ↗
Physarum slime mould can form adaptive transport networks. Tero and colleagues compared their efficiency, resilience and cost with infrastructure networks. A network can also be one large cell with many nuclei, complicating the boundary between an individual and a collective.
Adaptive network design · Tero et al., 2010 ↗
Development and regeneration require cells to coordinate at tissue scale. Frameworks such as Levin’s TAME explore whether concepts of agency and problem-solving help explain this organisation. That is a research perspective, with definitions and mechanisms to investigate.
Multiscale cognition · Levin, 2022 ↗
Erez and colleagues found chemical communication in certain bacteria-infecting viruses that influences later infection outcomes. This is a striking example of information extending beyond one individual event; it does not establish conscious intent or a general intelligence in viruses.
Communication between phages · Erez et al., 2017 ↗Original conceptual illustrations. These examples motivate questions; they do not establish that Flockinator reproduces the biological mechanisms. Human teams and their tools offer another setting in which to ask where an individual’s contribution ends and a system’s capability begins.
06 / THE INVITATION
Flockinator is one experiment in translating agentic behaviour into light, sound and motion that human senses can explore. It is a discovery tool and an artwork: a microscope for relationships we might otherwise miss, and a mirror for our own assumptions about intelligence, intention and cooperation.
The instrument shapes what we can notice. The signals we choose to display, the objectives we set and the patterns we expect all influence our interpretation. An unexpected rhythm may reveal something about the agents—or about the way we are looking.
New scenarios, recordings and experimental overviews will keep this an active investigation. Contribute code, critique an experimental method, or propose another way to make emergent behaviour perceptible and test what it means.