A Review of Recent Research in Metareasoning and Metalearning
Anderson, Michael L., Oates, Tim, AI Magazine
Recent years have seen a resurgence of interest in the use of metacognition in intelligent systems. This article is part of a small section meant to give interested researchers an overview and sampling of the kinds of work currently being pursued in this broad area. The current article offers a review of recent research in two main topic areas: the monitoring and control of reasoning (metareasoning) and the monitoring and control of learning (metalearning).
What Is Metacognition in Computation?
Rosie (the robot maid from the TV show The fetsons) spends her days cooking, cleaning, ironing, and attending to the usual household tasks of late 21st century life. Because of a bug in one of her memory chips, however, she almost always forgets to buy dog food when she goes out. She has an adequate recovery plan for this: she simply feeds Astro some of the Jetson's dinner. But 21st century human food is expensive, so this strategy is wasteful. Realizing this, and recognizing that she has forgotten several times, Rosie adopts a special strategy to help her remember: she sticks the spare dog collar in her apron, where she will see it next time she is at the store. Rosie's special strategy is an instance of metacognition: Rosie monitored her performance in this cognitive task (remembering the grocery list), recognized a deficiency, and applied knowledge of her own operation (knowing she would see the collar in her apron) to take specific steps to address the deficiency.
Later that same day, Rosie consults her list of things to do. One of them, making the arrangements for the Jetsons' vacation, is going to involve a great deal of computationally intensive planning on her part, and Rosie's processor is old and slow. Knowing that doing this planning will take resources away from other tasks, and interfere with the other things she has to do that day, she schedules the computationally intensive task for the late evening, when her overall workload is less. This, too, is an instance of metacognition: knowing about her own capacities and scheduling tasks to make the best use of her limited resources. Essentially, metacognition is any such strategy that involves the monitoring, modeling, and control of cognition.
To put the matter more generally and formally, imagine two components, X and Y (where X and Y could be the same), related in such a way that state information flows from Y to X, and control information flows from X to Y. Component X is in a monitoring and control relationship with component Y, and when y is a cognitive component, we call this relationship metacognitive monitoring and control. Put formally, then, the research question for the subject of metacognition in computation is: what are the sets (X, Y, S, E)-where Y is a cognitive component of a computational system S, and E is its environment-such that having some X in such a relationship with Y provides benefits to the system (and what are these benefits)?
Recent years have seen a resurgence of interest in the topic of metacognition in computation. Metacognitive architectures and approaches have found application in diverse areas, from computer security (Caleiro, Vigan, and Basin, 2005; Welch and Stroud 2002; Kennedy 2003; Garfinkel and Rosenblum 2003) to cognitive modeling-including the modeling of decision making (Cohen and Thompson 2005, Oehlmann, 2003), commonsense psychology (Hobbs and Gordon 2005, Swanson and Gordon, 2005), the relations between emotion and judgement (Hudlicka 2005), and rhetorical force (Lundstrôm, Hamfelt, and Nilsson 2005)-to computer gaming applications such as adversary generation (Zachary and Le Mentec 2000), to human-computer interaction (Kirn 2005) and automated tutoring (Muldner and Conati 2005). Perhaps the most widely publicized metacognitive initiative has been IBM's autonomie computing project (Ganek and Corbi 2003), intended to use self-monitoring to improve the ability to build and manage both …
Questia, a part of Gale, Cengage Learning. www.questia.com
Publication information: Article title: A Review of Recent Research in Metareasoning and Metalearning. Contributors: Anderson, Michael L. - Author, Oates, Tim - Author. Magazine title: AI Magazine. Volume: 28. Issue: 1 Publication date: Spring 2007. Page number: 7+. © 2009 American Association for Artificial Intelligence. Provided by ProQuest LLC. All Rights Reserved.
This material is protected by copyright and, with the exception of fair use, may not be further copied, distributed or transmitted in any form or by any means.