Abstract for AISB 2000: How to Design a Functioning Mind
AUTHORS: Stevo Bozinovski and Liljana Bozinovska,
Laboratory of Intelligent Machines and Bioinformation Systems
Electrical Engineering Department
University of Skopje, Skopje, Macedonia
and
Physiology Institute, Medical Department,
University of Skopje, Skopje, Macedonia
TITLE: Integrated Biologically Induced Architecture:
Architecture of Mind Considering Integration of Genetic,
Neural, and Hormonal System
ABSTRACT
Biological minds are integration of genetic, neural and hormonal
system. By contrast, biologically inspired AI architectures are
typically neural, or neural and genetic. This discusses an
integration of genetic, neural, and hormonal concepts for
architectures of artificial minds.
An architecture will be discussed which shows integration of
those systems. Several concepts will be used in the analysis and
synthesis of such mind architecture, including the concept of
emotion-based learning. The basic idea is that a mind inherits
basic emotions from a genome vector. Secondary emotions are
learned through emotion backpropagation process carried by a
hormonal signal to the whole neural net of the mind, where only
some of the neural synapses are updated, due to the a neural
ability to remember previous actions toward the environment. A
crossbar neural architecture is used and the crossbar learning
rule is implemented in this mind architecture.
================================================================
SHORT CV: Stevo Bozinovski
Email bozinovs@rea.etf.ukim.edu.mk
Stevo Bozinovski made his BSEE, MSc, and PhD at the Electrical
Engineering Department, University of Zagreb all on teaching neural
networks based agents. As Fullbright fellow during 1980-81 he joined
the Adaptive Networks group, Computer Science Department, University
of Massachusetts at Amherst, where he worked on neural based agent
teaching and reinforcement learning. The ANW Group (Spinelli, Arbib,
Klopf, Barto, Sutton, Anderson, Bozinovski, Porterfield) worked in
that time on solving the Minsky's assignment of credit problem with
a neural network. Two approaches has been undertaken: the
Actor-Critic Architecture (Barto and Sutton) and Crossbar Adaptive
Array architecture (Bozinovski). It turned out that the CAA solved
the problem at least a year before AC architecture did. CAA
introduced several notions in learning agents, one of them being
self-reinforcement learning or emotion learning. After the Sloman
and Croucher (1981) work on emotion architectures, the CAA
(Bozinovski 1982) is the first (neural network based) architecture
that uses the concept of feelings and genetic bootstrapping.
Today Bozinovski is a professor of Artificial Intelligence and
Cybernetics at the Electrical Engineering department, University of
Skopje. His main research direction is beings, biological and
artificial. Currently he is active in emotion based architectures,
and his work on emotion (1996) is an example of the new wave of the
research in emotion based agent architectures catalyzed by
Canamero's initiative (Canamero, 1998). Some recent papers are
Bozinovski S. 1995. Consequence Driven Systems. Gocmar Press
Bozinovski, S., Stojanov G., Bozinovska, L. 1996. Emotion,
embodiment, and consequence driven systems. Proc AAAI Fall Symposium
on Embodied Cognition and Action, p. 12-17, AAAI Press
Bozinovski S. 1999. Crossbar Adaptive array: The first connectionist
network that solved the delayed reinforcement learning problem. In
A. Dobnikar, N. Steele, D. Pearson, R. Albrecht (Eds.) Artificial
Neural nets and Genetic Algorithms pp. 320-325, Springer Verlag
Bozinovski S. 1999. Training a goalkeeper robot using emotion-based
learning architecture, In A. Paiva, C. Martinho (Eds.) Proc.
Workshop on Affect in Interactions, pp. 67-76, Siena
Bozinovski S., Bozinovska L. A connectionist theory of emotion based
on value judgment. Journal of Cybernetics and Systems, (in press)