One for One and All for All (Thesis)
There are several systems dealing with autonomous agents working as a group. In most of them, however, agents work for a common goal, not contemplating individual objectives. When two or more goals are available, many systems lack believability, for the choice is based mainly on utility. We aim at creating a system which allows agents in a group to decide in a believable manner whether to follow an individual or a group’s goal.
Extending Rui Prada’s SGD Model, which successfully implemented agents with personality based on the Five Factor Model, we created a model based on Psychology and Sociology theories, allowing agents to decide what is more important, the group or themselves. This choice is based on agents’ Egoism and on the attraction and influence they feel from the group. In turn, these depend on other factors such as personality, social relationships, outer stimulus, self-esteem or group’s and agents’ success.
We implemented our model on agents playing a semi-competitive game, Power Pentagram. On this game, agents must cooperate to win the different levels, at the same time they should compete in order to be in advantage against other members in the group.
The tests on our model showed that it successfully reflects different behaviors for groups with different constitutions, with results that would be expected for human groups.
Keywords: Autonomous Agents, Egoism, Self-Esteem, Individual Goal, Group Goal, Believability, Individuality