| Abstract: The biopsychosocial model acknowledges that high-level phenotypes (e.g., impulsivity, good at math, etc.) are multiply-caused by a plethora of contributing, interacting factors, such as specific genetic configurations, socio-economic status, culture, ethnicity, gender, and so on. However, the model does not explain how these variables interact with each other or how, exactly, they contribute to a specific outcome. I submit to the reader a potential explanation. Drawing on theoretical advancements from the field of complex systems and on computational research on the dynamics of behavioral repertoires (McDowell, 2013; Popa, 2013; Popa & McDowell, 2016), the theory discussed here proposes that the interaction between agents and their environment consists of a continuous-choice process during which agents adapt to environmental changes. This process molds an individuals context (e.g., income, culture, etc.) into robust collections of cognitive, emotional, and behavioral manifestations like impulsivity, authoritative parent, etc. The factors typically associated with impulsivity (for example) contribute to its emergence indirectly, by altering the relative value of existing options, and, by extension, the moment-to-moment probability of choosing one course of action over another. High-level phenotypes, therefore, cannot be directly explained by the contributing factors themselves, but by the moment-to-moment changes they produce in cognition and action. |
| Abstract: Computational simulations of behavior are becoming increasingly common and complex within the quantitative analysis of behavior, but there has been very little discussion regarding how these models should be evaluated. For example, models of the unified theory of reinforcement (Donahoe, Burgos, & Palmer, 1993) and the evolutionary theory of behavior dynamics (McDowell, 2004) are designed to simulate the internal processes that lead to observable behaviors, but these models are relatively abstract, their predictions are best viewed qualitatively, and there is some randomness built into them. These models also have characteristics of genetic algorithms and neural networks, which are powerful problem solving techniques. This type of approach is very different form functional equations such as the matching law. How should we be evaluating simulations and what do they tell us about behavior? Other fields like biology and sociology have struggled with evaluating simulations, but behavior analysis has not had this discussion yet. In this talk I will review approaches to this issue and provide guidelines for how we should be evaluating simulations in behavior analysis. |