Using neural networks for simplified discovery of some psychological phenomena

• Authors: Ryszard Tadeusiewicz
• Venue: ICAISC'10 Proceedings of the 10th international conference on Artifical intelligence and soft computing: Part II
• Year: 2010

Neural Networks are often used for solving practical problems and are known and appreciated as an effective soft computing tool. Nevertheless we cannot forget that neural networks are models of parts of the biological brain. Very simplified models, but similar in general behavior to some psychological phenomena. Therefore sometimes we can obtain interesting observations during study of neural network behavior and we can discover on this basis some new psychological ideas. Paper presents an example of such approach. Self-learning process (called also unsupervised learning) is an interesting form of neural network application, slightly different from other neural networks issues. During this process the network ought to discover new knowledge instead of registration of existing knowledge performed by the network during normal supervised learning. This process is described and discussed in many papers, but all of them are goal oriented ones: the main goal of the research is how to obtain the best self learning result e.g., in term of input data classification or similarity measurement. Meanwhile, during the self-learning process some phenomena can be encountered, very interesting from the psychological point of view, when the neural network self-learning process is considered as a model of cognitive processes occurring in our mind during self-learning or thinking. Such phenomena, observed during neural networks self-learning processes are described in the paper. Their psychological interpretation lead to hypothesis of artificial dreams, which can be discovered in neural networks and first time reported in this paper. In the paper, examples of such artificial dreams are presented and discussed. The problem under consideration is definitely controversial one, but the phenomena itself is interesting as a new interpretation of processes observed in neural networks.