Seeing is as Good as Doing

Given the privileged status claimed for active learning in a variety of domains (visuomotor learning, causal induction, problem solving, education, skill learning), the present study examines whether action-based learning is a necessary, or a suffi cient, means of acquiring the relevant skills needed to perform a task typically described as requiring active learning. To achieve this, the present study compares the effects of action-based and observation-based learning when controlling a complex dynamic task environment (N = 96). Both actionand observation-based individuals learn either by describing the changes in the environment in the form of a conditional statement, or not. The study reveals that for both active and observational learners, advantages in performance (p < .05), accuracy in knowledge of the task (p < .05), and self-insight (p < .05) are found when learning is based on inducing rules from the task environment. Moreover, the study provides evidence suggesting that, given task instructions that encourage rule-based knowledge, both active and observation-based learning can lead to high levels of problem solving skills in a complex dynamic environment. Seeing is as Good as Doing Who has better knowledge and skill: the back seat driver, who is learning to drive, or the actual driver, who is also learning to drive; the person watching their friend play a new game on the Sony play station, or the friend who is actually playing the game? Our daily lives frequently involve learning to control complex dynamic environments like those referred to in the question, but how we come to form the relevant skills needed to master such environments remains much debated. Laboratory versions of these tasks, referred to as Complex dynamic control tasks (CDC-tasks; see Figure 1: water purifi cation system) typically include several inputs (salt, carbon, lime) that are connected via a complex structure or rule to several outputs (chlorine concentration, temperature, oxygenation). 1 University College London Address correspondence to: Magda Osman, Department of Psychology, University College London, Gower Street, London WC1E 6BT, England Telephone: +4420 7679 7572; Fax: +4420 7436 4276; Email: [email protected] http://dx.doi.org/10.7771/1932-6246.1029 The Journal of Problem Solving • 30 Magda Osman In such environments, people are required to make a series of decisions, often in real time, that each depend on the other, and in an environment that changes autonomously as a consequence of a person’s actions (Brehmer, 1992). Implicit theorists (Berry, 1991; Berry & Broadbent, 1984, 1987, 1988; Dienes & Berry, 1997; Lee, 1995; Stanley, Mathews, Buss, & Kotler-Cope, 1989) claim that the learning process involved in dynamic task environments is procedural. The knowledge that is acquired is “knowing how” to perform actions that are tied to specifi c goals. In contrast, declarative knowledge, which is “knowing that” of particular facts about the underlying actions, involves structural knowledge concerned with the goal itself (Anderson, 1982). By extension, implicit learning theorists (Berry, 1991; Berry & Broadbent, 1988; Dienes & Berry, 1997; Lee, 1995; Sun, Merrill, & Peterson, 2001) have proposed that knowledge acquired in CDC-tasks and experience in controlling them is procedural and embedded within the interactions problem solvers have with the system. This form of learning produces instance-based procedural knowledge, that is, specifi c actions undertaken while interacting with the system become associated with the specifi c effects that they generate. Thus, mastering a control task requires successfully matching the goal and the current situation to previously encountered instances in order to determine the next appropriate response. Knowledge is conscious only to the extent that the response appropriate to a given situation can be stated, but what led them to make this response is unavailable to consciousness (Buchner, Funke, & Berry, 1995; Dienes & Berry, 1997; Dienes & Fahey, 1995, 1998). The empirical foundation of this position is the phenomenon showing that declarative knowledge is dissociated from procedural knowledge. This is because the learning systems understood to support declarative and procedural knowledge are distinct. In support of this, people have been shown to successfully control a CDC-task independently of any reportable declarative knowledge of the rule or causal structure that determines its operation, and without self-insight as to how they are able to perform it (Berry & Broadbent, 1984, 1987, 1988; Stanley et al., 1989). Another compelling demonstration of dissociations between ruleand instance-based knowledge is found in the contrasting effects of observation-based and procedural-based learning (Berry, 1991; Lee, 1995). Because declarative knowledge cannot easily be used to execute tasks suited to procedural knowledge (Berry, 1991; Berry & Broadbent, 1984, 1987, 1988; Dienes & Berry, 1997; Lee, 1995), it has also been claimed that possessing declarative knowledge will impair one’s later ability to perform a procedural task. Observation-based learning involves problem solvers watching the actions taken by another problem solver attempting to learn a CDC-task. This encourages the observer to focus on understanding how the system operates (i.e., rule-based knowledge), and is claimed to engage the explicit learning system (Berry, 1991). In contrast, conventional methods of training to control a CDC-task are procedural based and are thought to encourage knowledge of how to operate the system (i.e., instancebased knowledge). When compared, observers show better rule-based knowledge than Seeing is as Good as Doing 31 • volume 2, no. 1 (Summer 2008) procedural learners, but poorer control performance, further demonstrating dissociations between procedural and declarative learning. Others, however, have suggested that successful skill acquisition depends on a combination of rule-based and instance-based knowledge, which develops through hypothesis testing (Burns & Vollmeyer, 2002; Osman, 2008a, 2008b; Sweller, 1988; Vollmeyer, Burns, & Holyoak, 1996). By exploring the task and formulating rules about how it operates, learners are able to update their rule-based knowledge through the instances that they have generated to test them. Through practice, a wider range of instances are experienced, and these enable learners to form generalizable knowledge that they can transfer to other similar tasks. Evidence for this comes from studies that compare different types of goal instructions during learning. For instance, instructions like “explore the system,” an example of a nonspecifi c goal (NSG), are contrasted with “learn about the system while trying to reach and maintain specifi c outcomes,” an example of a specifi c goal (SG). The former instruction is assumed to encourage hypothesis testing, because rules can be generated and tested, whereas in the latter instruction, learning is constrained by generating instances that fulfi ll specifi c criteria. When tested on their ability to control the system to previously trained goals, SG-learners’ performance is equal to that of NSG-learners that have not learned to perform the task to any criteria. Furthermore, for untrained goals, NSG-learners outperform SG-learners. Taken together with evidence that NSG-learners also have superior structural knowledge about the system, this suggests that ruleand instance-based knowledge combined is more effective than instance-based knowledge alone (Burns & Vollmeyer, 2002; Osman, 2008a; Vollmeyer et al., 1996). Given these confl icting views, this article asks: Is procedural learning necessary to ensure skill acquisition in a complex dynamic environment? To address this, the present study included six conditions (Active [generate]-Instance, Active [generate]-Instance + Rule, Active [replicate]-Instance, Active [replicate]-Instance + Rule, Observe-Instance, ObserveInstance + Rule), across which the involvement of procedural learning was gradually attenuated. If procedural learning is necessary for skill development in CDC-tasks, then active-learning conditions will consistently show superior control performance compared with observation-based learning conditions, but poorer rule-based knowledge. If instead, instanceand rule-based knowledge combined are necessary for control skills to develop, then, regardless of actionor observation-based learning, both kinds of learners will show superior knowledge relative to those acquiring only instance-based knowledge.