A Brief History of the Soar Project

By Paul Rosenbloom

Soar started as a standard collaboration between a faculty member (Newell) and a graduate student (Laird). In '83 John and I finished our degrees and agreed to stay on at CMU and work with Allen on pushing Soar as a general cognitive architecture. In '84 John and I both left CMU and moved to the San Francisco Bay Area. We kept up the collaboration on Soar, but now at a distance. The first workshop occurred in '86 (at Stanford), and was instigated by the fact that we now had two semi-independent groups working on Soar (one at CMU and one at Stanford/Xerox), and we needed to increase the interactions among them to maintain a coherent project. We've continued to hold workshops in the US approximately every 8-10 months since then. The first few alternated between Stanford and CMU. Michigan was then added to the mix after John moved there in '86. USC/ISI replaced Stanford after I moved here in '87. Ohio St. later joined the rotation after a critical mass of Soar researchers grew up there (and they showed interest in holding workshops). We just recently attended the 13th Soar workshop at Ohio State. The next one is planned for Michigan later this year.

The community itself has grown over the years. The largest segment has traditionally been comprised of the groups that have grown up around Allen, John, and myself (whereever we have been). However, a range of other groups (and individuals) have also played important roles. In the US, Ohio St. has been the most significant such group. My understanding is that they got interested because of Soar's ability to combine problem solving methods, and issue that was then critical for them in the future development of their generic-task approach to expert system construction. Other groups and individuals have become interested because of the work on cognitive modeling, expert systems and robotics. These days we generally get around 60-80 people coming to the workshops.

My understanding is that the European Soar community originated primarily out of visits by several European researchers to CMU (in particular, John Michon of the University of Groningen in the Netherlands and Richard Young at the MRC in Cambridge, England). The first European Soar workshop was held in 1988 in England. The second one was held in Groningen, and included an intensive tutorial (more recent European workshops have also tended to include tutorials, as learning to use Soar has remained an issue, particularly for those who work far from the primary sources of expertise). More recent workshops have occurred in Cambridge, Regensburg (Germany), and Nottingham. The next one is planned for Leiden (Netherlands) later this year. My understanding is that the European Soar Workshops generally attract 20-30 people. The EuroSoar community has a strong cognitive flavor, whereas the US community combines cognitive and AI flavors.

There are now over 100 sites around the world that have a copy of Soar (though we don't know how many are actually using it). The heaviest concentrations are in North America and Europe, but there is a smattering of sites in Asia and elsewhere.

Some of the major research thrusts these days within the community include autonomous agents within simulated environments, human-computer interaction, and integrated natural language abilities. Various additional aspects of learning, planning/problem-solving, production system technology, expert systems, and cognitive modeling also remain strong.

@article[CarlW91,
author	=	"Carley, K. & Wendt, K.",
title	=	"Electronic Mail and Scientific Communication:
A study of the Soar extended research group",
Journal	=	"Knowledge: Creation, Diffusion, Utilization",
Volume	=	"12",
year	=	"1991",
BookPage=	"One: 563",
Pages	=	"406-440"
]