L. Rutkowski et al. (Eds.): ICAISC 2012, Part II, LNCS 7268, pp. 706–715, 2012.
© Springer-Verlag Berlin Heidelberg 2012
Hybrid Anticipatory Networks
Andrzej M.J. Skulimowski
AGH University of Science and Technology, Chair of Automatic Control,
Decision Sciences Laboratory, al. Mickiewicza 30, 30-050 Kraków, Poland
ams@agh.edu.pl
Abstract. This paper presents a theory of hybrid anticipatory networks that ge-
neralizes earlier models of consequence anticipation in multicriteria decision
problems. We assume that the decision maker takes into account the anticipated
outcomes of future decision problems linked by the causal relations with the
present decision problem. This can be represented by a multigraph, where deci-
sion problems are modeled as nodes linked causally and by one or more addi-
tional anticipation relations. These types of multigraphs are termed anticipatory
networks. Hybrid anticipatory systems may contain additional models of ran-
dom and non-cooperative game decisions. Constructive solution methods for
decision problems modeled by anticipatory networks are discussed as well.
Further, we present a generalization of hybrid anticipatory networks, known as
superanticipatory systems. In the final section we discuss some of their applica-
tions in the design of decision-making rules in autonomous robotic systems and
in filtering technology development scenarios.
Keywords: Anticipatory networks, decision theory, multicriteria optimization,
analysis of consequences, superanticipatory systems.
1 Introduction
The introduction of anticipatory networks as models of future consequences in a deci-
sion-making process was inspired by the idea formulated in [6] as “To use anticipated
future consequences of a decision as a source of additional preference information in
multicriteria decision problems”. The exploration of such anticipatory feedback is
possible owing to the following assumptions:
1. A decision maker is responsible for solving a decision problem which
corresponds to the starting node of the anticipatory network.
2. There exist estimates (forecasts or foresight scenarios) of future decision
problem formulations, their solution rules, decision makers’ preferences and of
the relations binding their anticipated outcomes with the current problem.
3. The decision maker knows the causal structure of future decision problems that
are modeled by the other nodes of the network, in particular the way in which
problem parameters are influenced by solutions to preceding decision problems.
The first and third assumptions allow us to model the impact of a decision to-be-made
on any subsequent problem in the network. The second assumption is a basis for