Work Plan Introduction Sample Clauses

Work Plan Introduction. For non-trivial behaviour to emerge, a population of agents should have a demanding environment confronting them with a challenge they can meet only by becoming advanced. Designing such environments and challenges is, therefore, a fundamental requirement for our project. We meet this requirement by seeking inspiration in the social sciences and setting up artificial worlds that originate from challenging problems there. On the other side of the coin are the agents. These must be suited for the environment(s) in question, but not be pre-engineered for the specific environmental challenge. We have to maintain a balance between agent properties that are hard-wired, i.e., properties that agents do have, and emergent properties that agents might have, i.e., that need to be acquired by adaptation and learning. We attempt to achieve such a balance by a minimalist approach, trying to keep the hard-coded intelligence at a minimum, enriching the lean agent structure only if necessary. We make a distinction between individual and evolutionary learning on the one hand, and social learning on the other. As for the first two, we intend to use existing mechanisms adjusted to our purposes. That is, we enable the agents to learn from the situations they encounter and make them evolvable, but major innovations concerning these mechanisms are not among our initial targets. The opposite is true for the social learning mechanism. In particular, we shall develop social learning mechanisms that do not work through social facilitation or imitation. Instead, we will implement mechanisms that allow passing knowledge by means of an evolved language to others within the same generation (with no permanent knowledge store such as a library). Our approach is based on perceiving a population as a collection of entities that process data and execute some machine learning algorithms to generate models matching these data. In this way, individuals might eventually have models which match the experience of large numbers of population members. In developing social learning systems we will build on the “newscast model” of computing, developed by an FP5 project3. In particular, we will investigate social learning mechanisms which allow information dissemination through epidemic protocols which have proved to be highly robust and effective in simple data mining tasks (▇▇▇▇▇▇▇▇▇, Jelasity, and ▇▇▇▇▇ 2003). Within the NEW TIES project we shall develop a far-reaching generalisation of this procedur...