Recent years have seen a resurgence of interest in the use of metacognition in intelligent systems. This essay 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 essay 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 Jetsons) spends her days cooking, cleaning, ironing, and attending to the usual household tasks of late 21 st 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 21 st 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

We present the basic principles and possible applications of systems capable of meta-reasoning and reflection. After a discussion of the seminal approaches, we outline our own perception of the state of the art, mainly but not only in computational logic and logic programming. We review relevat successful...

In previous work, we have proposed a multi-level agent model with (at least) a meta-level aimed at meta-reasoning and meta-control. In agents, these aspects are strongly related with time and therefore we retain that they can be expressed by means of temporal-logic-like rules. In this paper, we propose an “interval” temporal logic inspired by METATEM, that allows properties to be verified in specific time interval situated either in the past or in the future. We adopt this logic for definition and run-time verification of properties which can imply modifications to the agent’s knowledge base.

We illustrate a general agent model which includes a base level BA and a meta-level MA. The MA performs various forms of meta-reasoning including meta-control, which has the role of making meta-level decisions effective on the BA. As, in our view, meta-reasoning and meta-control are often concerned with time, we introduce the possibility of expressing temporal meta-rules. A very important meta-level activity in evolving agents is learning: we propose a general vision for interacting agents, where agents learn their patterns of behavior not only by observing and generalizing their observations, but also by “imitating ” other agents, after being told by them. The process of learning by imitation is based on meta-reasoning about various aspects, from self-monitoring to knowledge evaluation. We propose an operational model for knowledge exchange assuming an agent society which is based on concepts of reputation and trust.

Abstract. This paper seeks to extend notions of monitoring in metareasoning to include symbolic and linguistic expressions of self for purposes of communication and learning. The essay is intended to present a synthesis in plain language that challenges the agent community interested in metareasoning to consider what it means for a system to understand itself in any meaningful way. The basic claim is that if an agent truly knows what it is doing and why, it should be able explain itself to others using natural language or some other interactive mechanism with humans. To perform self-explanation it must be able to understand itself, and for this to occur it must monitor its own metareasoning and have an episodic memory that forms the basis of self. A further challenge is to incorporate self-explanation into an evaluation function that complements criteria based solely on action performance. 1

In previous work, we have proposed a multi-level agent model with at least a meta-level aimed at meta-reasoning and meta-control. In agents, these aspects are strongly related with time and therefore we retain that they can be expressed by means of temporal-logic-like rules. In this paper, we propose an interval temporal logic inspired by METATEM, that allows properties to be verified in specific time interval situated either in the past or in the future. We adopt this logic for definition and run-time verification of properties which can imply modifications to the agent’s knowledge base.

.... a conscious mental and usually purposive process aimed at the self-observation and reporting of conscious inner thoughts. Introspection relies upon thinking, reasoning, and examining one's own thoughts, and is contrasted with extrospection, the observation of things external to one's self. Introspection is in general used synonymously with selfreflection. In Computer Science, reflection is usually understood as the process by which a computer program can observe and modify its own structure and behavior. Reflection, as discussed in (Costantini, 2002) can take a variety of forms according to the underlying programming paradigm, but in essence it can be seen as the act of introspecting (upward reflection) and, symmetrically, of resuming normal operation (downward reflection). In Computer Science there is instead no commonly-agreed definition of introspection. It is interesting to notice that philosophy provides, since Locke, an “observational model ” of introspection where introspective capacity enables us to observe the inner world just as perceptual capacity enables us to observe the outer world. In analogy to philosophy, in this paper we shall understand introspection as the ability to create and access arepresentation of the object-level knowledge and reasoning processes and to be able to perform metareasoning based on this representation that can be considered as a (partial)