Cloning for Intelligent Adaptive Information Agents Keith Decker 1, Katia Sycara 2, and Mike Williamson 2 1 Dept. of Computer and Information Sciences University of Delaware, Newark, DE 19716 decker@cis.udel.edu 2 The Robotics Institute Carnegie Mellon University, Pittsburgh, PA 15213 (sycara,mikew) @cs.cmu.edu Abstract. Adaptation in open, multi-agent information gathering sys- tems is important for several reasons. These reasons include the inability to accurately predict future problem-solving workloads, future changes in existing information requests, future failures and additions of agents and data supply resources, and other future task environment charac- teristic changes that require system reorganization. We are developing a multi-agent financial portfolio management system that must deal with all of these problems. This paper will briefly describe our approaches and solutions at several different levels within tile agents: adaptation at the organizational, planning, scheduling, and execution levels. We discuss our solution for execution-level adaptation ("cloning") in detail, and present empirical evidence backing up the theory behind this execution-level so- lution. 1 Introduction Adaptation is behavior of an agent in response to unexpected (i.e., low proba- bility) events or dynamic environments. Examples of unexpected events include the unscheduled failure of an agent, an agent's computational platform, or un- derlying information sources. Examples of dynamic environments include the occurrence of events that are expected but it is not known when (e.g., an agent may reasonably expect to become overloaded), events whose importance fluc- tuates widely (e.g., price information on a stock is much more important while a transaction is in progress), the appearance of new information sources and agents, and finally underlying environmental uncertainty (e.g., not knowing the duration of a query). We have been involved in designing, building, and analyzing multi-agent systems that exist in these types of dynamic and partially unpredictable envi- ronments [18]. These agents handle adaptation at several different levels, from the high-level multi-agent organization down to the monitoring of individual method executions. In the next section we will discuss an agent's internal archi- tecture. Then we will discuss agent adaptation at the organizational, planning,