Despite the fact that the origin of software agent systems has been rooted in autonomous artificial intelligence and cognitive psychology, their implementations are still based on conventional imperative computing techniques rather than autonomous computational intelligence. This paper presents a cognitive informatics perspective on autonomous agent systems (AAS’s). A hierarchical reference model of AAS’s is developed, which reveals that an autonomous agent possesses intelligent behaviors at three layers known as those of imperative, autonomic, and autonomous from the bottom up. The theoretical framework of AAS’s is described from the facets of cognitive informatics, computational intelligence, and denotational mathematics. According to Wang’s abstract intelligence theory, an autonomous software agent is supposed to be called as an intelligent-ware, shortly, an intelware, parallel to hardware and software in computing, information science, and artificial intelligence.
TopIntroduction
A software agent is an intelligent software system that autonomously carries out robotic and interactive applications based on goal-driven cognitive mechanisms. The studies on software agent are rooted in the essences of computing science and cognitive science such as automata theory (von Neumann, 1946, 1958, 1963, 1966; Shannon, 1956), Turing machines (Turing, 1950), cognitive psychology (Newell, 1990; Sternberg, 1997; Anderson and Rosenfeld, 1998; Matlin, 1998), artificial intelligence (McCarthy, 1955, 1963; McCulloch, 1943, 1965; Barr and Feigenbaum, 1981), computational intelligence (Poole et al., 1997; Wang, 2008a), and decision theories (Wald, 1950; Newell and Simon, 1972; Berger et al., 1990; Bronson and Naadimuthu, 1997; Wang and Ruhe, 2007; Wang, 2008b).
The history towards software agents may be traced back to the work as early as in the 1940s. J. McCarthy, W. McCulloch, M.L. Minsky, N. Rochester, and C.E. Shannon proposed the term Artificial Intelligence (AI) (McCarthy, 1955, 1963; McCulloch, 1943, 1965). S.C. Kleene analyzed the relations of automata and nerve nets (Kleene, 1956). Then, Bernard Widrow developed the technology of artificial neural networks in the 1950s (Widrow and Lehr, 1990). The concepts of robotics (Brooks, 1970) and expert systems (Giarrantans and Riley, 1989) were developed in the 1970s and 1980s, respectively. In 1992, the notion of genetic algorithms was proposed by J.H. Holland (Holland, 1992). Then, distributed artificial intelligence and intelligent system technologies emerged since late 1980s (Bond and Gasser, 1988; Kurzweil, 1990; Chaib-Draa et al., 1992; Meystel and Albus, 2002, Meystel and Albus, 2002).