Collective Behavior | Towards AI
Collective Behavior can be seen all around us. What really defines collective behavior models is that there is no global orchestrator that directs any power, but rather many individual agents locally communicating or interacting with each other, which allows them to synchronize or coordinate with the population.
Collective Behavior can start to be modeled in two modes, the Lagrangian or Eulerian. For the Lagrangian model, the modeling strategy is individually based, where each individual’s movements are simulated based on simple rules for their interactions: terms of distances between group members. This then allows the observance of the emerging collective motions. Past research of fish migration and movement, focused on genetically fixed patterns, was highly influenced by this form of modeling. Nearly every environment on earth has inhabitants that heavily rely on their physical surroundings; environments that are subject to varying degrees of change and strongly favor individuals who have the ability to learn.
The Eulerian modeling approach for aggregation, combined with Projective Simulation (previous blog post is all about this), was used to study locust movements. This was primarily focusing on the evolution of animal densities using equations similar to statistical mechanics or fluid dynamics. The results for the locusts study showed that “the solitary state of locusts became unstable as the population density reached a critical value.” This can be extremely beneficial for understanding how to handle invasive species like insect invasions.
“Understanding how social influence shape biological processes is a central challenge, essential for achieving progress in a variety of fields ranging from the organization and evolution of coordinated collective action among cells, or animals, to the dynamics of information exchange in human societies.”
“During plagues, the Desert Locust has the potential to damage the livelihood of one-tenth of the world’s population.” (Food and Agriculture Organization of the United Nations)
Statistical modeling with new AI techniques has great potential in illuminating the hidden mechanisms behind collective behavior. They allow for the expansion of empirical studies that help and have led to, the development of new mathematics and improvements of numerical simulation algorithms.
The rest of this blog will highlight the research done at the Max Planck Department of Collective Behavior, based at the University of Konstanz in southern Germany. They currently have three labs led by Iain Couzin, Director of the Department, Damien Farine, and Alex Jordan. Quick flex for their reputation, “The Max Planck Society has a world-leading reputation with 33 Nobel Prizes awarded to their scientists.”
I’ll be focusing on the work of Dr. Iain Couzin and his team’s research on fish schools: “his work aims to reveal the fundamental principles that underlie evolved collective behavior.”
School of Fish
Useful system to study since:
- They have to solve many challenges in an unpredictable environment.
- Each individual takes in sensory data and reacts in particular movements that coincide with the population.
Animals coordinate movement together to avoid predators as a unit. Predictions are difficult for Collective Behavior since there are many emerging properties. Some or most can be unexpected properties from the individual components. This is why schools of fish, a simple organization, is still so mysterious to us.
What rules do they use to translate sensory cues?
Challenges and Lab Setup
It gets fairly difficult when trying to track fish that move at fractions of a second! Modern technology allowed them with very high frame rate cameras to track each individual’s movement hundreds of times per second.
It was conducted in a large indoor tank full of fish. They used four roof-mounted cameras that were time synced together, and then graphically stitched the streams into one big view of the tank.
The goal was to use mathematical and computational techniques to find out how individuals take in complex sensory information and transform it into movement within fractions of a second.
Quantitatively proves and reveals for the first time how the emerging collective behaviors arise from the interactions of individual agents’ movements within a system. Through hundreds of trials presented to fish from nine families, Dr. Couzin’s team managed to show that fish escape decisions are governed by a conserved set of decision-making rules.
How does the behavior of an individual affect the behavior of the group?
Collecting and testing Real-World Data
The study took place in the red sea, looking particularly at damselfish.
Groups of 3–4 individuals up to 20–30 since tracking data with wild animals is difficult and adding collective behavior adds more complexity. Smaller groups allow for better detail and visualizations.
Three cameras were set up around the coral head where damselfish live in. The cameras shot 120fps since fish can all get back into the coral within 5 to 6 frames.
- Then characteristics to each individual fish were applied, like whether male or female, small or large, dominant or subordinate…
How can those characteristics affect the behavior of the group when reacting to stimulus?
- An Ipad that flashes (stimulus)
- WHO reacts first, second, third?
- WHO has visual access to stimulus?
- WHO sees which other individuals (social information)?
Once the tech has advanced enough, the team aims to study groups of hundreds of thousands of these individuals, while also knowing exactly which individual is which.
Looking at animal collectives, ignorance and being uninformed is actually a very positive thing. The uninformed individuals, in a way, managed to democratize the group’s behavior by preventing extremist individuals from having a disproportionate influence on the group.
The Media, where the same information is broadcasted to millions of individuals.
- Essentially erodes the capacity for collective intelligence, which relies on not being told what to think but rather on the evidence found by each individual in order to solve the problem.
- We find again and again in animal groups that they have evolved strategies to avoid having overly correlated information.
But in human society, do we possibly rely too much on information?