Project Description

A basic feeling that the theoretical framework and also the methodology of psychology does not sufficiently reflect evolutionary and energetic aspects of emotions and a principal curiosity about what would happen if one approaches this field using numerical simulation methods well known and established in evolutionary biology and in other fields motivated us to start this project.

Goal of this research is to find out if emotion-like patterns of function and energy mobilization emerge, through evolution, in a population of so-called autonomous agents. These agents move around and underlie energetic selection pressure. They have to develop appropriate survival strategies. Emotions are conceptualised as particular patterns of energy consumption, related to elementary motor and cognitive behaviour, like moving towards or away from a detected object and widening or narrowing the range of perception. By exposing the agents to changing environmental challenges, we like to see if a set of different emotion-like functions will appear, with a typical pattern of energy consumption and behaviour. We are also interested to explore if emotion-like behaviours are similar on the individual and on the group level. The general aim of the project is to develop a conceptual and computational framework that allows to model, run and analyze more complex emotion-cognition dynamics and, thus, to study elementary aspects of of emotions. It should be consistent and generic enough to serve other research topics in the future. At the same time, the attempt is to design the model as minimal and simple as possible in order to reduce and manage the automatically arising complexity.

The simulation model is based on a population of autonomous agents that have the capacity to perceive their environment and to move. Depending on what an agent detects in its vicinity - food, predator, fellows, or none of that - it activates a respective functional program which is genetically coded (see sample 1). In order to support this, the agents are equipped with a set of genes that each code for a specific behavioural aspect in each of the different modes - movement in respect to a detected object and attention spend for the observation of the environment. Additionally, there are genes that code for perceptive priorities concerning the object types in the environment of an agent. All gene loci underlie evolutionary change in the sense of mutation and selection.

Sample 1. Autonomous agents (medium sized circles) are moving within an action space. Small grey circles are food units, the brightness represents the energy content. Large moving circles (dark red/orange) are predators that are hunting agents. Depending on the current object in the focus of attention, the agents are in food mode (yellow), in predator mode (red), in fellow agent mode (green) or they currently do not detect an object (blue).

The model is insofar entirely energy-based as each action of an agent diminishes the amount of energy that the agent accumulated so far. Moving slowly or quickly and perceiving at short or long distance diminishes the agent's energy proportionally to the square of the step size respectively to the square of the distance. Consuming energy-rich or low-energy food units increases the agentís energy proportionally. From a certain amount of contained energy, food units can only be eaten by a conjoint action of agents. That forces agents to cooperate, to form groups and to act in a synchronous way. Running completely out of energy, as well as meeting a predator leads to instant death. Evolution is simulated through a defined number of generations, each with a specific life span (number of action steps). Within one action step each agent 1. observes its vicinty resulting in a new object in the focus of attention (which in many cases is the same as in the previous step), it 2. switches the functional mode according to the type of the detected obejct and it 3. performs a movement towards or away from the detected object as coded by the genome. An agent surviving until the end of its life span reproduces asexually. The number of genetically identical offspring is proportional to the amount of energy that it accumulated during its life span. Thus, situation-appropriate behaviour is transmitted from generation to generation, while unsuccessful behaviour tends to be eliminated. Evolutionary change is implemented by random mutations of the genetic program at reproduction. The patterns of behaviour and the associated energy consumption in each individual mode of function can freely evolve. For reasons of simplicity the model uses distinct generations, i.e. all agents reproduce at the same time at the end of a lifespan and the parents die.

Selective scenarios
The model allows to simulate evolutionary processes under a bandwidth of different selective conditions. It is very interesting and a often lot of fun to see how the agents react on them. An important aspect is the number of new food units that are randomly distributed in each action step and their energy content. For instance, the same amount of energy can be introduced by few energy-rich food units (see smaple 2) or by many low-energy units (sample 3). Despite the fact that the systems energy dissipation is the same, it will cause the system to favour different patterns of behaviour, since the competitive situation differs. In principal, agents have to develop effective strategies to deal with the competition about rare ressources. A further selective pressure can be introduced through predators that are hunting agents.

Sample 2. Few new energy-rich food units per action step cause a behaviour with very fast and therefore energy consuming movement towards food. The system can accomodate only for a low number of agents.

Sample 3. Many new low-energy food units per action step introduce the same overall energy amount into the system as in sample 2. However, competition among agents is much lower, more relaxed pattern of bahaviour evolves and the number of agents that survive is considerable higher.

Grouping and the fractal structure of affective behaviour
A basic hyptohesis of affect logic is that there is a fractal symmetry (similarity) of affectve-cognitive function on different social levels, ranging from the individual level to large groups. In order to explore this in the context of the given model, agents have to be exposed to selective conditions that favour the creation of smaller and larger groups. This can be done by increasing the energy content of food units so that only groups of agents can access them. Forming groups and synchronous behavior however is a more complex pattern and it takes a population many generations of evolution to adapt to such conditions. The model therfore allows to start a simulation with small food units and increase the energy content continously over generations, until the desired food size is reached. This allows the population to adapt and to follow. In this way, agents evolve respective strategies and can form groups with numbers of up to 50 members.

Sample 4. Depending on their energy content, large food units can only be 'cracked' by a number of agents. In order to make them accessible, agents therfore have to cooperate and to operate in groups. This is a behaviour that can evolve when it is favoured by the specific frame conditions. What you observe is an ephemeral self-organizing pattern that is caused by mutual attraction of agents and synchronous movement towards food. Agents can be rendered with a mode-specific color or blue with an intensitiy that represnts the size of the respective group.