Emergent representations and reasoning in adaptive agents

Classically, cognition assumes that the underlying mechanisms of thinking are based on symbol manipulation processes. This assumption has several drawbacks, such as the issue of where symbols (representations) actually come from. To overcome this drawback, the interactivist approach proposes that representations emerge from the continuous interaction with the environment and subsequent anticipation of such interaction. This approach provides understanding of the nature of knowledge representation and reasoning in adaptive agents. We have used the interactivist approach as basis for an interaction-based computational model. Our model is embedded in an adaptive agent whose task it is to find food in a maze. Feedback about the agent's success is given through a reinforcement signal that marks the current interactions. With this computational model, we show the actual emergence of representation as well as the emergence of what we call 'simple reasoning'. Finally, we discus...

• Type: Conference
• Year: 2004
• Where: ICMLA
• Authors: Joost Broekens, Doug DeGroot
• Keywords: Adaptive Agent, ICMLA 2004, ICMLA 2007, Interactivist Approach, Symbol Manipulation Processes