Barbara Hayes-Roth

A blackboard architecture for control

Abstract The control problem—which of its potential actions should an AI system perform at each point in the problem-solving process?—is fundamental to all cognitive processes. This paper proposes eight behavioral goals for intelligent control and a ‘blackboard control architecture’ to achieve them. The architecture distinguishes domain and control problems, knowledge, and solutions. It enables AI systems to operate upon their own knowledge and behavior and to adapt to unanticipated problem-solving situations. The paper shows how opm , a blackboard control system for multiple-task planning, exploits these capabilities. It also shows how the architecture would replicate the control behavior of hearsay-ii and hasp . The paper contrasts the blackboard control architecture with three alternatives and shows how it continues an evolutionary progression of control architectures. The paper concludes with a summary of the blackboard control architectures strengths and weaknesses.

A Cognitive Model of Planning

This paper presents a cognitive model of the planning process. The model generalizes the theoretical architecture of the Hearsay-II system. Thus, it assumes that planning comprises the activities of a variety of cognitive “specialists.” Each specialist can suggest certain kinds of decisions for incorporation into the plan in progress. These include decisions about: (a) how to approach the planning problem; (b) what knowledge bears on the problem; (c) what kinds of actions to try to plan; (d) what specific actions to plan; and (e) how to allocate cognitive resources during planning. Within each of these categories, different specialists suggest decisions at different levels of abstraction. The activities of the various specialists are not coordinated in any systematic way. Instead, the specialists operate opportunistically, suggesting decisions whenever promising opportunities arise. The paper presents a detailed account of the model and illustrates its assumptions with a “thinking aloud” protocol. It also describes the performance of a computer simulation of the model. The paper contrasts the proposed model with successive refinement models and attempts to resolve apparent differences between the two points of view.

Differences in Spatial Knowledge Acquired from Maps and Navigation

Abstract Models of the spatial knowledge people acquire from maps and navigation and the procedures required for spatial judgments using this knowledge are proposed. From a map, people acquire survey knowledge encoding global spatial relations. This knowledge resides in memory in images that can be scanned and measured like a physical map. From navigation, people acquire procedural knowledge of the routes connecting diverse locations. People combine mental simulation of travel through the environment and informal algebra to compute spatial judgments. An experiment in which subjects learned an environment from navigation or from a map evaluates predictions of these models. With moderate exposure, map learning is superior for judgments of relative location and straight-line distances among objects. Learning from navigation is superior for orienting oneself with respect to unseen objects and estimating route distances. With extensive exposure, the performance superiority of maps over navigation vanishes. These and other results are consonant with the proposed mechanisms.

The use of schemata in the acquisition and transfer of knowledge

Abstract A learning model based on “memory schemata” is presented. The model assumes that knowledge substructures in memory are shared by multiple representations of information from diverse contexts. These substructures, or schemata, are collections of concepts and associations that occur together repeatedly and act as unitary, higher-order concepts. When knowledge is represented in terms of a schema, associations from the schema to additional concepts specify the detailed information for that context. This organization of knowledge entails both costs and benefits for the acquisition and retention of new information that utilizes a schema. The use of a familiar encoding structure facilitates memory access at storage and retrieval time. Multiple uses of the shared structure produce interference among concepts from the various contexts. The predictions of the model were tested in two transfer experiments in prose learning. Experiment l demonstrated the simultaneous effects of both facilitation and interference in the learning of diverse information conforming to a single schema. Recall of information from the schema was a non-monotonic function of schema strength. In Experiment 2 facilitation was preserved while interference was eliminated by increasing the discriminability among competing contexts. Results from both experiments confirmed the predictions of the model.

Concept learning and the recognition and classification of exemplars

A model is proposed for concept learning and subsequent recognition and classification of OLD and NEW exemplars. The model, called the “property-set model,” assumes that a learned exemplar is encoded in memory as a set of the component properties and combinations of properties of the exemplar. Recognition of a presented exemplar is assumed to be an increasing function of the memory strengths of its component property-sets, while classification of the exemplar is determined by its most diagnostic property-set. This model is contrasted with a number of alternative models, including prototype-plus-transformation, feature-frequency, and nearest-neighbor models. In an experimental evaluation of alternative models, subjects attempted to learn two concepts by classifying exemplars in an anticipation paradigm. They then performed recognition and classification tasks with particular exemplars. On a within-subject basis, the property-set model was the best predictor of both recognition and classification performance.

Architectural foundations for real-time performance in intelligent agents

Intelligent agents perform multiple concurrent taks requiring both knowledge-based reasoning and interaction with dynamic entities in the environment, under real-time constraints. Because an agents opportunities to perceive, reason about, and act upon the environment typically exceed its computational resources, it must determine which operations to perform and when to perform them so as to achieve its most important objectives in a timely manner. Accordingly, we view the problem of real-time performance as a problem in intelligent real-time control. We propose and define several important control requirements and present an agent architecture that is designed to address those requirements. The proposed architecture is a blackboard architecture, whose key features include: distribution of perception, action, and cognition among parallel processes, limited-capacity I/O buffers with best-first retrieval and worst-first overflow, dynamic control planning, dynamic focus of attention, and a satisficing execution cycle. Together, these features allow an intelligent agent to trade quality for speed of response under dynamic goals, resource limitations, and peformance constraints. We illustrate application of the proposed architecture in the Guardian system for surgical intensive care monitoring and contrast it with alternative agent architectures.Intelligent agents perform multiple concurrent taks requiring both knowledge-based reasoning and interaction with dynamic entities in the environment, under real-time constraints. Because an agents opportunities to perceive, reason about, and act upon the environment typically exceed its computational resources, it must determine which operations to perform and when to perform them so as to achieve its most important objectives in a timely manner. Accordingly, we view the problem of real-time performance as a problem in intelligent real-time control. We propose and define several important control requirements and present an agent architecture that is designed to address those requirements. The proposed architecture is a blackboard architecture, whose key features include: distribution of perception, action, and cognition among parallel processes, limited-capacity I/O buffers with best-first retrieval and worst-first overflow, dynamic control planning, dynamic focus of attention, and a satisficing execution cycle. Together, these features allow an intelligent agent to trade quality for speed of response under dynamic goals, resource limitations, and peformance constraints. We illustrate application of the proposed architecture in the Guardian system for surgical intensive care monitoring and contrast it with alternative agent architectures.

A domain-specific software architecture for adaptive intelligent systems

A good software architecture facilitates application system development, promotes achievement of functional requirements, and supports system reconfiguration. We present a domain-specific software architecture (DSSA) that we have developed for a large application domain of adaptive intelligent systems (AISs). The DSSA provides: (a) an AIS reference architecture designed to meet the functional requirements shared by applications in this domain, (b) principles for decomposing expertise into highly reusable components, and (c) an application configuration method for selecting relevant components from a library and automatically configuring instances of those components in an instance of the architecture. The AIS reference architecture incorporates features of layered, pipe and filter, and blackboard style architectures. We describe three studies demonstrating the utility of our architecture in the subdomain of mobile office robots and identify software engineering principles embodied in the architecture. >

Acting in Character

In this context, it is not the personality of the actor that interests us, but the personality of the character he or she portrays. Thus, when we say that an actor is “in character,” we mean that the actor is behaving in accordance with a personality created by an author, shaped by a director, and assumed by an audience, for purposes of a particular performance. A good actor creates and communicates a consistent and compelling personality throughout a performance of a given role and creates different personalities for different roles.

A social-psychological model for synthetic actors

1. ABSTRACT WC provide synthetic actors that portray fictive characters by improvising their behavior in a muhimedia environment. Actors are either autonomous or avatars directed by users. Their improvisation is based on the directions they receive and the context, Directions can take different forms: high-level scenarios, user commands, and personality changes in the character portrayed. In this paper, WC look at this last form of direction. We propose a sociaLpsychological model, in which we can define personality traits that depend on the values of moods and attitudes. We show how synthetic actors can exploit such a mode1 to produce performances theatrically interesting, believable, and diverse. The Cybercaf6 is used to test those features.

Integration of Knowledge from Text.

Three experiments investigated factors influencing the integration of facts acquired from texts. Experiment 1 investigated integration of two constituent facts into a single higher-order knowledge structure. Integration was measured by (a) the ability to detect that case fillers from separately acquired facts were part of a single case frame and (b) the ability to retrieve case fillers from separate facts given only the common case frame for the two facts. Integration was facilitated by proximity of related facts and by identical, rather than paraphrased, wordings of the common case frames in related facts. Experiment 2 replicated the basic findings of Experiment 1 in a recognition paradigm and demonstrated that integrated memory representations preserve the identities of the original constituent facts. In Experiment 3, integration was measured as the ability to verify inferences deducible from separately acquired facts. This kind of integration also benefitted from identical, rather than paraphrased, wordings of the common information in related facts. In addition, Experiment 3 showed that the observed effects persisted after a 30-minute retention interval.