Artificial Cognition

Abstract

Abstract This chapter illustrates tools that can be used in order to endow artificial agents with cognitive abilities that go beyond behavioural algorithms—notably, Shafer’s evidence theory, Atkin’s Q-analysis, and Kohonen’s unsupervised neural networks. These tools have been selected with the aim of allowing modellers to reproduce states of mind that deviate even substantially from either pure or bounded means-ends rationality. In particular, evidence theory provides a framework to represent the sort of uncertainty that originates from the fear that something may happen, that one is not even able to figure out. According to several economic theories, such states of mind can have a substantial impact on investment decisions and eventually generate underemployment equilibria. Q-analysis of hypergraphs allows holistic set representations where interactions among components can generate qualitatively novel features. Applications may concern work teams and boards of directors, as well as intertwined opportunities in scenario analysis. Unsupervised neural networks construct their categories around the items to which they have been exposed most often. Their capability to discover novel and unexpected patterns makes them useful in domains ranging from time series to text analysis whenever one does not know what regularities one may expect. Finally, since cognition concerns the beehive just as much as the bee, this chapter concludes by hinting at the possibility of an agent-based model displaying its own cognitive capabilities. In particular, connectionist tools and results can be adapted to organizational contexts by exploiting the analogy between information loops and organizational routines.