A generative decision theory. An agent-based computational approach for modelling, studying and making decisions in organizations

In recent years organizations have improved their abilities to study and predict possible situations thanks to new management techniques and to the development of technological instruments capable of capturing and analyzing large volumes of data. In a system where organizational variables are becoming more and more con-trolled and where the main actor still remains the most important unknown factor: humans and their behavior, their decision-making processes are more and more crucial for organizations and their stability. Several business and social science fields such as management, economics and psychology in the last decades have identified better how human habits and decision-making processes are connected. Human’s behavior has been studied through the development of decision science. In particular, two approaches have been used for studying, modeling and making decisions. Which are: the normative and the descriptive approach. The normative approach is based on the analysis of decisions related to simple rules and norms used by a hypothetical human that make decisions. The descriptive approach, ex-amines individual decisions in the context of a set of needs, preferences, beliefs and values that an individual has. Considering the first approach, most human characteristics have been smooth out to create sophisticated simulations. Individuals are programmed to choose the best rational strategy and capable of maximizing the utility. Basically, this approach considers an ideal decision maker whom is fully informed, completely rational, and able to compute with perfect accuracy. The second approach named descriptive, aims to study people’s behavior starting from a basic decision level, in a very accurate way but often without computing a theory. Because many variables are required to predict people’s behavior. Although, the descriptive decision theory has demonstrated how the normative theory has a lack of understanding for some traits of real human behavior. The Prospect theory of Daniel Kahneman and Amos Tversky is probably the most well-known example. Kahneman and Tversky identified three regularities in human decision-making, which are that: people put more emphasis on adjustments in their utility-states than focusing on absolute utilities; losses are perceived as bigger than profits; and the evaluation of subjective odds is biased by a sort of anchoring effect when choosing. As for these empirical evidences, descriptive approach presents several other decision phenomena named heuristics and biases, which are human strategies and departures from classic normative rationality, that have a relevant impact on human decision making processes. These studies have proved that humans make decisions differently compare to the normative theory. This demonstrates the limits of this approach. On the other hand, studies developed by the descriptive approach, difficulty lead to new models that are able to predict human behavior because of the high level of analysis of the research. Limits of both approaches helped from one side to develop some normative models which consider more human habits. On the other side, the descriptive studies started by analyzing trend’s choices in order to model decision processes. Evolution in decision study approaches has allowed the development of some applicative studies, for example in the medical sector, engineering and especially in Economics and Organizational studies. This was the major factor in the emergence of behavioral economics, earning Kahneman a Nobel Prize in 2002. At the proof of interest for the heuristic and bias program, over time, studies from cited disciplines have demonstrated that subjects infringe on many other axioms of rationality in different applied fields, by detecting the presence of numerous biases. On the other hand, productivity in heuristics and biases has become a double-edged sword. The program in biases and heuristics led to a sort of rush in discovering new biases and heuristics, in several cases very similar to each other. Some researchers thought that it was possible to reorganize this fragmented research program by introducing classifications and taxonomies of cognitive phenomena. The scientific literature presents some examples of classifications and taxonomies of heuristics and biases based on different theoretical approaches. Currently, the absence of common criteria of categorizations makes the process of comparison between biases very difficult. Recent studies suggest an empirical approach to realize a taxonomy, based on experimental research in decision-making which has shown the presence of variability among different individuals in their abilities to solve problems that were created to identify cognitive fallacies. The first part of the dissertation will present a series of evidence on relations between several biases and heuristics, to highlight the presence of underling factors and dimensions of origin of these cognitive departures. Based on these results, researchers thought that departures from normative standards could be due to more random performance errors, rather than consistent fallacies across decision-making skills. The idea that heuristics and biases are driven by different mental strategies is also known by evolutionary psychologists but they con-sider the “heuristics and biases program” denigrating for human reasoning because it is described as fallacious instead of being an evolutionary resource. In particular, Gerd Gigerenzer has described heuristics as processes that help us make better choices and economize cognitive resources rather than just departures from rationality. He considers heuristics as efficient cognitive processes that ignore information in order to make better decisions in a limited amount of time. Finding a common definition for different approaches for biases and heuristics, some researchers consider them as all that can differentiate us as humans, in the decision making processes, compared to a normative rational agent, like a computer. This definition could be acceptable if these differences are considered as a process exclusively present in human nature, and impossible to be emulated by a ration-al agent. Like a computer for example. It might seem bizarre as a response, but during the same period when the descriptive approach was rising and showing all the limits of the normative approach, a new paradigm of studying and doing science was emerging: it is called the generative approach. As seen the descriptive approach is based on phenomena observation and of the deduction thinking process based on it. The observation method is related to experimental work and laboratory studies of individual decision-making processes. It has the merit of improving a real picture of how humans make decisions, compared to the normative method. However, even a perfect knowledge of individual decision-making rules does not guaranty the possibility of predicting the macroscopic structure of human behavior. This possibility has been lately explored in recent years, starting from some solid scientific studies on individual behavior and using advance computing techniques capable of “growing up" phenomena at the macro level, making it possible to obtain counterintuitive hypothesis about behavior and implications in organizations. This method is definitively powerful for testing, with generative sufficiency and some unexpected rules given by behavioral research. The method can be considered as a scientific revolution, according to some researchers, it is a third way of doing science, after deduction and induction methods. The generative method consists of generating and growing a phenomenon for explaining it. In recent years, computer simulation of social phenomena has produced a new scientific paradigm, which is the science that generates events and that would be impossible to recreate them or observe them. Computer simulation based on virtual agents (ABM) has become an essential tool to generate observable facts, an instrument of revolutionary development for the decision science. ABM offers an innovative and effective manner to conduct empirical research. ABM purposes to create considerable social phenomena qualitatively and quantitatively. Literature provides numerous models of prosocial behavior, cooperation, punishment, organization dynamics, and social phenomena. In fact, one of the possibilities offered by this tool is to explore behavioral effects at a macro level, such as the organizational level. Several organizations have already started using this approach to study the consequences of policies of the behavior of the individuals who com-pose them. The approach has given excellent results in terms of simulations enabling the study of effects of decisions and helping to make good organizational decisions. In particular a characteristic of the ABM instrument, and more generally the generative science, is the use of information from different approaches (descriptive and normative) to create dynamics that would not be possible otherwise. In this dissertation, in order to show the importance of this instrument, ABMs studies have been developed based on results obtained from the empirical results of experiments developed in the first part of the research. These studies have been realized through an implementation into agents of cognitive fallacies, biases and heuristics in order to study decisional processes and effects on organizations and artificial societies of belonging. The present dissertation will present step by step this innovative method that has revolutionized various sciences, in this case the decision sciences applied to organizational contexts. Because several disciplines are involved in this dissertation, such as: decisional sciences, organizational studies and artificial intelligence, it will present some theoretical parts regarding the scope of the study. First, the study of decision-making related to decisions within the organization will be introduced, and then more general theories of decisions will be presented. Then, it will introduce the generative method applied to the decisions and the ABM as the instrument for excellence for studying decisions in organizations.