An Architecture for Collaborative Problem-Solving Control in Associate Systems

In the past, AI systems have strived to automate problem-solving processes completely. However, in recent years researchers have come to realize that it is not always possible or desirable to aim for total automation. Researchers are realizing the importance of human-computer collaborative systems in which the human and the computer work as a team in solving problems. This approach raises the question of how to design systems that support effective collaborative problem-solving between humans and computers. The primary contribution of this thesis is an architecture for coordination of collaborative problem-solving (CO-SOLVE) for an important class of human-computer collaborative systems called associate systems. Associate systems are knowledge-based systems which share the cognitive workload with their human partners. Designers of associate systems must deal with the complexities of integrating mixed-initiative (i.e. human and computer) control with the general issues of problem-solving control faced by traditional AI systems. CO-SOLVE provides mechanisms for attention synchronization and collaborative alternatives exploration. The Attention Synchronization Model (ASM) allows the system to track (rather than direct) the user's activities in order to provide advice relevant to the current user activities. The Collaborative Alternatives Exploration Model (CAEM) is a mixed-initiative approach to exploring large, complex solution spaces. In this collaborative framework the user serves as solution evaluator and system controller and the computer as solution alternative generator. System developers using CAEM explicitly lay out steps in the problem-solving process for a given task and define points of human-computer interaction within the sequence of process steps. The other contribution of this thesis is a proof-of-concept prototype of CO-SOLVE, called SEDAR, which is implemented for a real-world, complex domain (flat and low-slope roof design). Two evaluations were conducted on SEDAR. The first assessed the effectiveness and usability of the ASM and its critiquing strategies as implemented in SEDAR. The evaluation showed that SEDAR reduced the error rate of experienced architects and also which advising strategies they preferred. The second evaluation assessed the effectiveness of the CAEM as implemented in SEDAR and showed that SEDAR (1) helped experienced architects reduce the amount of time spent developing solutions, and (2) increased the number of alternatives searched in the solution space.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2015

Author dc:creator

• Fu, Michael Chin-Ming

Contributors dc:contributor

• Caroline C. Hayes

• Artificial Intelligence

Language dc:language

eng

Identifier

(MiAaPQ)AAI9737108

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/81878