KSL Home Page

They are investigating adaptive intelligent systems -- AI systems that coordinate perception, reasoning and action to pursue multiple goals while functioning autonomously in dynamic environments. After spending many years investigating agents with deep knowledge and powerful reasoning capabilities, our more recent work focuses on more rounded agents that complement their intelligence with human qualities, such as identity, personality, affect, relationship, and embodiment.

Barbara Hayes-Roth, of the Knowledge System Laboratory at Stanford University, is the primary author of the Guardian system. The Guardian system is implemented an extended version of the dynamic control architecture, also designed by Dr. B. Hayes-Roth. The focus of this research was to create adaptive intelligent systems which can perform multiple related knowledge-based tasks while functioning autonomously in natural environments under real time constraints.

---

Общие сведения

Когнитивная архитектура AIS разработана в первоначальном варианте исследовательской группой под руководством Барбары Хейес-Рот (Barbara Hayes-Roth) в Лаборатории систем, основанных на знаниях (Knowledge System Laboratory - KSL) в Стенфордском университете, г. Пало-Альто, Калифорния, США.
Проект не поддерживается с 2003 года, но архитектура используется в серии других проектов, например:
Guardian -- An intelligent agent for real-time monitoring of intensive-care patients.
BB1 -- Software system for building intelligent agents и др.

Архитектура AIS [Hayes-Roth, Pfleger, Lalanda, Morignot, & Balabanovic, 1995] хранит процедурные знания в виде набора поведений, каждое со связанными условиями запуска, и планов управления, которые задают временные закономерности шагов плана. Они согласуются, изменяются и взаимодействуют через декларативную память, которая хранит фактические знания, намерения-деятельности и следы опыта агента. Во время каждого цикла мета-контроллер оценивает разрешенные поведения и выбирает, какие из них исполнить. Архитектура включает делиберативный когнитивный слой, который отвечает за оценивание ситуаций и планирование, и более быстрый физический слой, который обрабатывает восприятие и действия в окружающей среде.

---

Description

An Adaptive Intelligent System (AIS) is a knowledge-based system that reasons about and interacts with other dynamic entities in real time. For an AIS to be effective, it must be capable of perception, action, cognition, and attentional focus.

Adaptive Intelligent System Dynamic Control Architecture

http://ai.eecs.umich.edu/cogarch0/ais/arch.html

The Dynamic Control Architecture does all of the planning and other high-level processing for the AIS architecture. It has three components:

• Agenda Manager
• Scheduler
• Executor

These components share a global knowledge-base named BB1. BB1 stores its knowledge in a BB* conceptual network representation. This is a homogeneous, symbolic, declarative approach based on semantic networks.

The GUARDIAN Architecture

Hayes-Roth has developed an AIS called GUARDIAN for use in an intensive care patient monitoring application. GUARDIAN comprises a cognitive component, a set of asynchronous I/O subsystems, a set of dynamic I/O channels, and a satisficing reasoning cycle.
Implemented as a parallel blackboard system, the cognitive component performs general-purpose reasoning by engaging in dynamic control planning. This is a process for incrementally constructing and modifying control plans: temporally ordered patterns of control decisions, each of which describes a class of operations the agent intends to perform during a particular period of time. Operations within the cognitive component are cyclically processed in turn by an agenda manager, an operations scheduler, and an operations executor.

GUARDIAN is an example architecture built using the AIS generic architecture. It's task is the monitoring of patients in a surgical intensive-care unit (SICU). Patients in a SICU require the use of life-support devices wich must be constantly monitored since, changes in the readings offered by these devices could imply that certain treatments are required immediatly, or the patient might die.
Guardian was designed to monitor the patient's life signs and readings and determine if the situation is normal or if the patient is in some kind of danger. In the later case, Guardian also tries to diagnose the problem and suggest the best possible course of action. In situations where a quick answer is required (i.e. there is little time to deliberate) Guardian provides it by reducing the time it spends doing less critical monitoring tasks and by returning the most likely diagnosis, along with an explanation and recommendation for treament. After the treatment is applied Guardian continues normal monitoring of the patient to make sure that he returns to a stable condition (i.e. the treatment was successfull).

Skills

Guardian implements several skills as part of its knowledge base.
It has skills for:
- classification of perceptual observations into temporal episodes of known semantic categories (e.g. normal, high, very high);
- diagnosis of observed signs based on belief networks;
- diagnosis and explanation of observed signs based on generic system models.

Smart Interactive Characters
(Work at Extempo, Redwood City, CA, USA)

At Extempo Systems, Inc. Dr. Hayes-Roth and her colleagues have created sophisticated commercial technology for interactive characters, including a server-based run-time engine, development tools, and analysis tools. They have also created a variety of prototypes and commercial applications related to customer relationship management, online learning, and interactive entertainment.
Extempo's Imp Character Development (ICE) tools enable non-technical authors (e.g., writers, educators, marketing experts) to create Smart Interactive Characters that have distinctive personas, well-defined functional roles, multi-dimensional mood dynamics, natural language conversation, social and learning skills, and adaptation.
The R&D underlying this technology and tools was supported by competitive awards from the NIST ATP (Advanced Technology Program).

Dr. Hayes-Roth and her colleagues recently began offering open access to the ICE tools and other assets on www.learning-guides.com.

---

Selected publications
AIS publications
Hayes-Roth,B. - papers
Hayes-Roth, B., Pfleger, K., Lalanda, P., Morignot, P., & Balabanovic, M. (1995). A domain-specific software architecture for adaptive intelligent systems. IEEE Transactions on Software Engineering, 21, 288–301.
Barbara Hayes-Roth, Karl Pfleger, Philippe Lalanda, Philippe Morignot, Marko Balabanovic. A Domain-Specific Software Architecture for Adaptive Intelligent Systems. Journal: IEEE Transactions on Software Engineering - TSE , vol. 21, no. 4, pp. 288-301, 1995
B. Hayes-Roth. PRINCIPLES OF DESIGN FOR LIFE-LIKE CHARACTERS, 2003.