Non-Symbolic Algorithms within the Context of Different Levels of Cognition Henning Veflingstad Computer Science Department Norwegian University of Science and Technology 7491 Trondheim NORWAY veflings@stud.ntnu.no Sule Yildirim Computer Science Department Hedmark University College 2451 Rena NORWAY suley@osir.hihm.no Abstract In Artificial Intelligence community, algorithms and symbols are accepted as the two sides of the same coin. On the other hand, there is a long lasting debate between the two significant approaches to AI, namely Symbolic AI and connectionism, on whether the human brain functions symbolically or not. In this work, we are proposing that algorithms and symbols are not necessarily the different sides of the same coin and that they appear separately. Thus, non-symbolic algorithms can exist. We further proceed to extend the idea of existence of non-symbolic algorithms to their existence in the human brain. We also present the representation of steps in an algorithm and the concepts on which those steps operate. The non-symbolic algorithms are high level and they can be part of either conscious or non- conscious thinking. We also elaborate on different levels of cognition and especially on what we call the conceptual level where high level human thinking happens and where the proposed non-symbolic algorithms reside. Introduction Reasoning and behaviour are the two aspects that are assigned to cognitive systems. Proving theorems, thinking, planning, language production are tasks which belong to the reasoning aspect of cognition. On the other hand, movement generation and coordination are tasks that belong to behaviour aspect of cognition. Parallel distributed processing or connectionism has been more successful with tasks relevant to the behavior aspect of cognition. In this paper, we investigate into their role relevant to the reasoning aspect of cognition within the scope of non-symbolic algorithms and non-symbolic concepts. More explicitly we can propose three levels of cognition: Stimulus-Response Level: This is the level where there is a direct functional mapping from the sensations of a situation to behavioural outcomes. For example, a robot might be wired up to avoid obstacles conforming to a particular pattern of activations across proximity sensors. Conceptual Level: This is the level where there is formation of concepts in parallel to a functional mapping from the sensations of a situation to behavioral outcomes. For example, a robot might be wired up to avoid obstacles conforming to a particular pattern of activations across proximity sensors and in the meantime it forms the concept of obstacle. The obtained concepts are employed in high level cognitive tasks e.g. thinking, planning, decision-making. The Language Level: This is the symbolic level and there is a mapping from the conceptual level to the symbolic language level. For example, a robot maps a concept of obstacle to the word “obstacle”. We believe that the current research in the field of cognitive science points to the above three levels in cognition. However, although we believe these levels are the major ones, the relation between levels is an issue under research in cognitive science including researchers from the fields of Artificial Intelligence, robotics, and complex systems. It is still vague how these levels are separated from each other. For example, if sensory-motor systems and conceptual representations are proven to be tightly related, the formation of concepts will be part of level 1, and level 2 will be left with the function of using formed concepts in higher level cognitive functioning. The research relevant to level 1 is grouped under sensory-motor actions or reactive behaviors. If the foot of a new born infant touches cold lake water, the infant will take its foot away in a reactive way. Since it is newly born, it has not yet formed concepts such as “lake water”, “cold”, “move away” etc. As it grows, it will form these concepts and will be able to utilize these concepts in forming thoughts such as “lakewater is cold” and this capacity is the subject matter of level 2. Relevant work that points to the possibility of level 2 is found in [Ziemke et al., 2005; Tani and Nolfi, 1999]. Rogers and McClelland (2006) suggest hidden layer representations in artificial neural networks as conceptual representations. Cangelosi (2004) uses similar representations to conceptualize world and also addresses the mapping from the conceptual layer to the language level. The relations between levels are formulated in the following questions: What is the relation between sensory-motor systems to conceptual representations? Are conceptual representations different in kind from those computed within the perceptual input systems and motor output systems that feed into and out of them? What is the relation of language to conceptual distinctions in thought? What is the dependency on language for the