Psychologically realistic cognitive agents: taking human cognition seriously

Ron Sun, Sébastien Hélie
<span title="">2013</span> <i title="Informa UK Limited"> <a target="_blank" rel="noopener" href="" style="color: black;">Journal of experimental and theoretical artificial intelligence (Print)</a> </i> &nbsp;
Cognitive architectures may serve as a good basis for building mind/brain-inspired, psychologically realistic cognitive agents for various applications that require or prefer human-like behavior and performance. This article explores a well-established cognitive architecture CLARION and shows how its behavior and performance capture human psychology at a detailed level. The model captures many psychological quasi-laws concerning categorization, induction, uncertain reasoning, decision-making,
more &raquo; ... d so on, which indicates human-like characteristics beyond what other models have been shown capable of. Thus, CLARION constitutes an advance in developing more psychologically realistic cognitive agents. Keywords: psychology, agent, cognitive architecture, CLARION. theories/models could be put to test against many well-known and stable regularities of human cognition that have been observed in psychology (i.e., psychological quasi-laws). So far, these integrative theories/models have taken the form of "cognitive architectures", and some of them have been successful in explaining a wide range of data (e.g., Anderson and Lebiere, 1998; Sun, 2002 Sun, , 2004 . Specifically, a cognitive architecture is a broadly-scoped domain-generic computational cognitive model describing the essential structures and processes of the 5 mind used for multi-domain analysis of behavior (Sun, 2004) . Newell (1990) proposed the Soar (Laird, Newell, and Rosenbloom, 1987 ) cognitive architecture as an example unified theory. Several other cognitive architectures have been proposed since then (e.g., . CLARION assumes the distinction between explicit and implicit knowledge, as well as the distinction of actioncentered and non-action-centered knowledge. Like a number of other cognitive architectures, CLARION is focused on the explanation of human behavior and has been successful in capturing a wide range of psychological data and phenomena (e. It is worth noting that cognitive architectures are the antithesis of "expert systems": Instead of focusing on capturing performance in narrow domains, they are aimed at providing a broad coverage of a wide variety of domains (Langley and Laird, 2003). Applications of intelligent systems (especially intelligent agents) increasingly require broadly scoped systems that are capable of a wide range of behaviors, not just isolated systems of narrow functionalities. For example, one application may require the inclusion of capabilities for raw image processing, pattern recognition (categorization), reasoning, decision-making, and natural language communications. It may even require planning, control of robotic devices, and interactions with other systems and devices. Such requirements accentuate the importance of research on broadly scoped cognitive architectures that perform a wide range of cognitive functionalities across a variety of task domains. 6 Cognitive architectures tend to be complex, including multiple modules and many free parameters (in their computational implementations). Because each module in a cognitive architecture usually includes its share of free parameters, increasing the number of modules in a cognitive architecture usually increases the number of free parameters in its computational implementation. Increasing the number of (independent) free parameters in a model adds to model complexity (Pitt and Myung, 2002; Roberts and Pashler, 2000) . In cognitive architectures, this can result not only from the number of modules (as explained above), but also from the complexity of within-module processing. The number of free parameters within a module can be reduced by making the modules highly specialized, but this can only be achieved at the cost of adding more modules. Likewise, the number of modules can be reduced by making very general modules, which, however, usually have more free parameters. Is it possible to create a cognitive architecture that can act as a unified theory and yet constrain the model complexity? The problem has been discussed in Sun (2004) , which argued that a cognitive architecture should be minimal. In this paper, we study how the CLARION cognitive architecture can be minimal, exploring its core theory as the basis for building better cognitive agents. The remainder of this paper is organized as follow. First, a general discussion of CLARION is presented. Second, the core theory of CLARION is expressed as a mathematical model. Third, a number of cognitive/psychological regularities are reviewed, along with mathematical and/or conceptual explanations of the phenomena using the core theory of CLARION (e.g., concerning categorization, uncertain reasoning, and decision-making). This presentation is followed by a general discussion.
<span class="external-identifiers"> <a target="_blank" rel="external noopener noreferrer" href="">doi:10.1080/0952813x.2012.661236</a> <a target="_blank" rel="external noopener" href="">fatcat:x5dsbwnfnffi7ifbv2naxxq664</a> </span>
<a target="_blank" rel="noopener" href=";context=psychpubs" title="fulltext PDF download" data-goatcounter-click="serp-fulltext" data-goatcounter-title="serp-fulltext"> <button class="ui simple right pointing dropdown compact black labeled icon button serp-button"> <i class="icon ia-icon"></i> Web Archive [PDF] <div class="menu fulltext-thumbnail"> <img src="" alt="fulltext thumbnail" loading="lazy"> </div> </button> </a> <a target="_blank" rel="external noopener noreferrer" href=""> <button class="ui left aligned compact blue labeled icon button serp-button"> <i class="external alternate icon"></i> </button> </a>