Toward Human-Agent Competition in TAC SCM Andrew Nelson * , Dickens Nyabuti, John Collins, Wolfgang Ketter † , and Maria Gini Dept of CSE, University of Minnesota, Minneapolis, USA † Rotterdam Sch. of Mgmt., Erasmus University, Rotterdam, NL Abstract We propose a variation of the TAC SCM supply-chain trad- ing competition, in which human decision-makers compete with fully-autonomous agents. Because of the complexity and time pressures of the competition environment, humans may be assisted by semi-autonomous agents, which could be modiﬁcations of existing agents. The research goal is to discover what kinds of decision support will make a hu- man decision-maker most effective in this environment. We show how an existing agent might be modiﬁed to operate in this new competition by updating our MinneTAC agent into a highly conﬁgurable, semi-autonomous agent that can sup- port human users playing a variety of roles in the modiﬁed competition environment. The agent’s decision processes are composed of networks of simple services that are described using an OWL ontology. The ontology describes the structure of the service network, along with the structure and seman- tics of the data elements that are produced and consumed by individual services. Introduction Organized competitions are an effective way to drive re- search and understanding in complex domains, free of the complexities and risks of operating in open, real-world en- vironments. Artiﬁcial economic environments typically ab- stract certain interesting features of the real world, such as markets and competitors, demand-based prices and cost of capital, and omit others, such as human resources, secondary markets, taxes, and seasonal demand. The Trading Agent Competition for Supply-Chain Management (Collins et al. 2005) (TAC SCM) is an economic simulation in which com- peting autonomous agents operate in a simple supply-chain scenario, purchasing components, managing a factory and warehouse, and selling ﬁnished products to customers. Supply chain management organizes the transfer of goods, information, and services between suppliers and buy- ers. Traditional supply chains are created and maintained through the interactions of human representatives of the var- ious companies involved. Much progress in business pro- cesses comes from automating the elements that do not re- quire human judgments, and from providing better and more * Work supported in part by National Science Foundation grant IIS-0414466 Copyright c  2009, American Association for Artiﬁcial Intelli- gence (www.aaai.org). All rights reserved. timely information for those that do. Recently many com- panies have adopted Business Intelligence (BI) (Eckerson 2005) techniques to support information needs, and a va- riety of systems to automate trivial tasks like data entry to decrease the margin of human error. Despite the many important research results arising from TAC SCM and the other Trading Agent Competitions, busi- ness people are not ready to trust fully autonomous agents with critical business decisions (Maes 1994), and existing BI systems are arguably not sufﬁciently ﬂexible to produce the kinds of ad hoc information and analysis that would truly leverage their skills and experience and maximize their ef- fectiveness (Collins, Ketter, & Gini 2008). One way to ad- dress both of these issues is through conﬁgurable, compos- able, and transparent decision processes that are fully de- scribed in terms that end users can understand. Our ap- proach to the needed ﬂexibility and transparency is called “ontology-driven decision support,” in which information, analyses, and decisions are composed of a large variety of data views and small, single-purpose analysis modules that can be composed into dataﬂow networks to produce results with well-deﬁned business meaning. What is missing at this point is a clear understanding of exactly what kinds of information, and what level of auto- mated assistance, will make human decision-makers most effective in complex, dynamic, and multi-faceted trading situations. It is also not clear whether an experienced hu- man decision-maker, with appropriate information and as- sistance, can outperform the current generation of fully- autonomous agents. We believe that a modiﬁcation to the Trading Agent Competition for Supply-Chain Management (TAC SCM), in combination with ﬂexible mixed-initiative agent technology such as we describe here, can help answer these important questions. The next section places our work in the context of related work. We then review the classiﬁcation of decisions into strategic, tactical, and operational levels in both a general business setting and in TAC SCM. These distinctions are im- portant to understand our approach to adjustable-autonomy decision making. We then discuss the ways in which the game and agents must be modiﬁed to operate in a human- agent competition environment. Next, we describe our ap- proach to ontology-driven decision support and to human- agent interaction, providing a few dashboard examples that