1 Abstract—In this paper, we report on work that applies a form of artificial intelligence (AI) to autonomous underwater vehicle (AUV) operations. Called “language-centered intelligence” (LCI), this form of AI uses hypothetical reasoning to build contingency plans that enable AUVs to proactively anticipate changes in mission circumstance. After describing the control architecture we have used to embed LCI in our fleet of AUVs, we present an application of LCI to the problem of vehicle loss. The specific solution to this problem operates at two levels. It is assumed that a standard replacement approach will be used to direct vehicle replacement in normal circumstances, while a higher-level, so- called “imagination” replacement approach that uses LCI will direct replacement in more extraordinary circumstances. First, though, what differentiates normal from extraordinary circumstances must be identified. Preliminary results from an experiment designed to distinguish normal from extraordinary replacement circumstances are presented. I. INTRODUCTION HE United States Navy has recognized the need for functional autonomous underwater vehicles (AUVs) to perform a variety of tasks [1]. To meet this need, a number of different AUVs have been developed [2], [3], [4]. At the University of Idaho (UI), researchers have developed a low- cost mini-AUV based on the vehicle described by Stillwell in [4]. This vehicle has been employed in the development of language and logics for cooperative behavior [5] and has been used to field test these behaviors in mine countermeasure (MCM) missions [5], [6], [7]. In preparing for these field tests, the Autonomous Littoral Warfare Systems Evaluator-Monte- Carlo (ALWSE-MC) simulation environment has been used to evaluate collaborative strategies and mission design. When multiple vehicles are used in complex, potentially dangerous MCM missions, there is an increased risk of vehicular failure. Vehicle failure has a direct impact on mission operations, upsetting collaborative dynamics and reducing fleet efficiency; further, vehicular loss can result in the loss of unique, mission-relevant information gathered by the vehicle in question. Despite efforts to prevent information loss [8], the loss of a vehicle and its contribution to collaborative tasks remains a threat to overall mission success. One approach to mitigating the damage done by vehicle loss is to implement expanded, flexible formation maintenance logics that respond rapidly and efficiently to changes in the size of the fleet. In particular, the logics should generate replacement strategies when confronted with vehicle loss that fit the specific mission context and enable the fleet to continue collaborative activity and pursuit of mission objectives. In this paper, we adopt this approach, arguing for the superiority of strategies that are both reactive to changes in circumstance and proactive in anticipating changes to come. As we develop them, these strategies are instances of language-centered intelligence (LCI), an approach to artificial intelligence (AI) that involves the dual use of the languages and logics used by the AUVs for information transfer and communication. The language and logic structure that already exists for processing language is harnessed for use without further development in reasoning about future scenarios. The specific solution to the problem of vehicle loss pursued in this paper will operate at two levels. It is assumed that a standard replacement approach will be used to direct vehicle replacement in normal circumstances, while a higher-level, so- called “imagination” replacement approach that uses LCI in the form of hypothetical reasoning will direct replacement in more extraordinary circumstances. First, though, what differentiates normal from extraordinary circumstances must be identified. After describing in detail the LCI approach and its relation to the AI literature, LCI is applied to the problem of vehicle loss. Preliminary results from an experiment designed to distinguish normal from extraordinary replacement circumstances are presented. II. LCI: THEORETICAL BACKGROUND LCI is an approach to artificial intelligence developed by researchers at the UI for application to AUV operations. LCI is the use of language processes to enable autonomous, artificial systems to consider hypothetical mission contingencies and respond more rapidly and flexibly to changing circumstances [9], [10]. A “language process” includes any algorithm or information exchange medium present in the system. In the AUVs, these processes include Enabling Autonomous Underwater Vehicles to Reason Hypothetically Nicodemus J. Hallin, Henry Egbo, Patrick Ray, Terence Soule, Michael O’Rourke, Dean Edwards Center for Intelligent Systems Research University of Idaho PO Box 441024, 607 Urquhart Ave., BEL W3-1 Moscow, ID 83844-1024 1+208-885-6500 hall3095@vandals.uidaho.edu T