UNIfying COllaborative, Reflective, Neurogenetic (UNICORN) Agent Framework

We propose to develop a new agent architecture based on a new agent framework, code-named the UNICORN (UNIfying COllaborative, Reflective, Neurogenetic) agent framework, which will combine the strengths of our own neurogenetic agent framework, enriched bythe theory of Emergence (and the collaborative, self-organizing Swarms), with the strengths of other types of agent architectures, including the reflective processes. In Phase I, we will analyze strengths and weaknesses of current leading architectures forintelligent agents and human behavior models, 2) develop a comparative framework to identify common and unique strengths and weaknesses, and 3) use that framework to design a new architecture that shows significant improvements over existing architectures.We will attempt to develop a `neurogenetic' agent framework, as an integral part of the comparative framework, to model, simulate, evaluate and compare the leading architectures for intelligent agents. The `neural' part of the neurogenetic agent frameworkcan be used to model various learning/reasoning capabilities, and perhaps even autonomy, while the `genetic' part can be used to model `interactions' among agents, the importance of which cannot be over-stressed. Furthermore, the `Swarm Intelligence' canbe fully integrated into this framework, broadening the types of interactions what can be modeled and simulated, when empirically evaluating and comparing existing agent architectures. Our proposed UNICORN agent framework centers on the evolutionary,neurogenetic approach to creating and controlling intelligent agents, which has been concept-proved in two of our government funded projects, the NIMA (National Imagery and Mapping Agency) funded EVA project (EVolving Intelligent Text-based Agents) and theAFRL (Air Force Research Lab) funded EMMA project (Evolving and Messaging decision-Making Agents). However, in this proposal we actually go one step further and merge our ideas of the neurogenetic agents with one of the central ideas in multi-agentsystems, namely, the concept of emergence and its associated concepts of emergent behavior and emergent functionality. The resulting unifying framework will be useful for any system of adaptive/learning agents, which we believe will be the dominant kind ofagents on the World Wide Web, due to the ubiquitous as well as mutable nature of the Web and its users' needs and interests.On the theory side, developing a formal framework for multi-agent learning can lead to an improved understanding of the general principle underlying learning in both computational and natural systems. Multi-agent learning is not just the sum of isolatedlearning activities of the constituent agents. The learning activities of an individual constituent agent may be significantly influenced by other agents.The building of a prototype to capitalize on its possibilities in Phase II would advance understanding of how to apply this novel architecture to complex, dynamic, and persistent problems. Other possible applications for the resulting framework includeagent systems for data mining on the Intranets or the Internet, agent systems for automatic construction of intelligence portals, and agent systems for automating and speeding up scientific discoveries on the Web or other complex information environments.