SIGHTPLAN: AN ARTIFICIAL INTELLIGENCE TOOL TO ASSIST CONSTRUCTION MANAGERS WITH SITE LAYOUT Iris D. Tommelein, Research Assistant Raymond E. Levitt, Associate Professor Department of Civil Engineering and Barbara Hayes-Roth, Senior Research Associate Department of Computer Science Stanford University Stanford, CA 94305 ABSTRACT SightPlan is an expert system that performs construction site layout. The system contains an explicit representation of the strategy it follows to construct a solution. Based on the assumption that the cognitive capabilities of an agent have an impact on the strategy that agent follows during problem-solving, we decided to experiment with alternate strategies. We modeled one strategy after the way in which a construction field manager would lay out a site (the Expert Strategy'), and we compared this model with a second one designed to make better use of the power provided by the computer (the 'Computational Strategy'). The results of this experiment, the comparison of both models, and the implications of this research are described in this paper. They allow us to conclude that Al- based computer programs assisting experienced managers in solving their task may result in better solutions than either the stand-alone machine or the individual person could obtain. 1 INTRODUCTION Site layout consists of identifying , sizing, and locating temporary facilities on a construction site . In practice , usually a field manager or a superintendent is assigned to draw up the layout plan at the beginning of construction . This preliminary layout plan describes the site when the project is started , displays the temporary facilities at that time, and shows facilities overlapping with the space allocated for major activities that will take place in the near - and long - term future. Research work in the 60's and 70's [1] formalized the layout process by identifying optimization criteria and by developing computer programs that could generate the best layout procedurally . When an optimum solution could not be guaranteed , criteria were relaxed and implementations would make use of heuristic methods and improvement procedures . Though these layout models were applied on some construction projects, the approach did not find general acceptance in industry , whereas for instance procedures for optimum location of single facilities [2; 3] proved to be more acceptable. We suggest that one of the reasons for their failure is that the black- box procedures for layout were too far removed from the way field practitioners do their job . Because they could not relate to it they would refuse the effort required to collect all needed data . Construction managers interested in disseminating their knowledge on site layout therefore had to resort to writing guidelines [ 4], checklists [ 5; 6; 7], or describing their experience from specific cases [8]. Though the shortcomings of algorithmic models were soon identified ( see, e . g., [9] for early ideas on the need for man - machine interactive systems in construction management ) it took several years for Al programming techniques to develop , mature [10; 340