Emergence of Meaning from a Stream of Experience – Grounding Symbols by Consequences and Intentions

At the heart of understanding the world is the ability to represent its relevant aspects in a meaningful manner. Trying to model this crucial power in an autonomous agent can contribute essentially to core problems of AI research. Some of these unanswered questions are: How is relevance to be integrated in a per se unambitious machine? What makes meaning from a systematics point of view? Can one single mechanism produce the plethora of mental phenomena we experience? Is high order cognition emergent from simple but yet numerous elementary processes? In this thesis we argue that relevancy and meaning can be represented by agents in direct relation to their past experiences and present intentions. From this base assumption a supervised multiple prototype learning algorithm is derived that receives examples and labels as feedback from its environment. Random optimisation allows for gradual increase in motor efficiency concerning local maxima of extrapolated sensor evaluations. A sensorimotor map of the agent's environment is generated that stores and optimises beneficial motor activations in evaluated sensor space by employing temporal Hebbian learning. Therefore this map is representing relevant, hence evaluated, physical activations that are pragmatically meaningful to the agent. Creating sensorimotor maps is understood as necessary precondition for forming more complex mental representations such as symbols or signs in general. A mechanism is outlined that generates higher level entities from basic sensorimotor representations. These structures represent domain specific competencies that differentiate contexts of beneficial activations in given states. Arbitrarily complex representations can be constructed by the same basic principles for different levels of abstraction. These representations are indisputably grounded in the most basic interaction between agent and world conceivable: sequential vectors in sensor and motor space.