It is with a tinge of sadness but also with great pride that in December 2014—following 36 productive years—we closed the Multi-Agent Systems (MAS) Laboratory in the School of Computer Science at the University of Massachusetts Amherst. In November 2014, the Lab's last PhD student graduated and our sponsored research efforts were completed. It was time.

The Lab since its inception has been pursuing a distributed model of computation involving a network of cooperating, intelligent agents—potentially encompassing both people and computers. Our work has been strongly motivated by practical applications but hopefully never losing sight of the more general domain-independent implications of our solutions and wherever possible trying to construct formal models and analysis. The goal of this work is to create modular and scalable frameworks for building complex distributed problem-solving applications that operate in open environments where there is limited communication and a wide range of task and environmental uncertainties. Implicit in this work has been the desire to gain a deep understanding of the nature of coordination and cooperation, both from an empirical and theoretical perspective.

This led us to explore a number of larger research issues, many of which have been, in one way or another, an implicit motivator of the Lab's research since the beginning:

• How to create computationally practical models/theories/frameworks for cooperation/coordination that exploit the characteristics of the underlying domain to make them more useable in realistic distributed applications?
• How does the nature of local problem solving need to change when it is done in the context of other agents?
• How to scale up multi-agent systems to hundreds or thousands of agents? This includes issues involved in creation of agent organizations and the associated issue of how they are assembled in the marketplace of agents; how their structure is determined and evolves; how designed and emergent organizations can be integrated; and how to construct organizationally situated agents.
• How to formally characterize the interdependencies among agent activities and relate that characterization to the appropriate satisficing multi-agent coordination protocol, given specific resource constraints and utility criteria? More generally, is there a theory of distributed search that can explain and encompass all the different mechanisms that can be used to coordinate agent activities?
• How can agents adapt their activities from both a short- and long-term perspective—what are appropriate learning mechanisms in a multi-agent context? This also includes the question of how emergent behavior arises.
• How to develop an integrated view of "satisficing" that takes into account approximate problem solving, partial and abstract communication of problem-solving results, and approximate coordination?
• How is coordination of cooperative versus self-interested agents different? Is there a unifying theme that will bridge what are now perceived as very different subfields of MAS? What are mechanisms for agents to coordinate when balancing local (self-interested) and non-local concerns (social welfare – cooperative)?
• How can different representations of agent activities—for instance, at the network transport level and at the organization level—interact in order to adapt their local policies/strategies to the needs of other levels?

In pursuing our research, a number of large testbeds have been developed to evaluate ideas. These testbeds have involved applications such as sensor networks for vehicle tracking and weather monitoring, speech and sound understanding, information gathering on the internet, peer-to-peer information retrieval, intelligent user interfaces, distributed task allocation and scheduling, and virtual agent enterprises.

One of the key approaches that has been pursued by the Lab is the development of self-aware agents that reason about their own local state as well as the goals, plans, intentions, and knowledge of other agents in deciding how to interact with them. This reasoning can potentially be more complex than that required for domain problem solving. Agents, and the system as a whole, operate in a "satisficing" mode—doing the best they can with the current resource constraints. In such systems, managing uncertainty is as an integral part of network problem solving. Agents of necessity are highly adaptive and function in a way that leads to highly reliable systems. They are able to modify their problem-solving structure to respond to changing task and environmental situations in both short- and long-term ways. These systems may involve tens to hundreds (and more) of agents, requiring complex organizational relationships among agents. There is much work left for the full potential of this model to be realized. The challenges of this realization are as exciting and interesting as in the first days of the Lab's existence but we now must leave it to other researchers to continue this work.

When the Lab was formed, the MAS community consisted of only a handful of researchers, and there were no conferences or workshops focused on this area. It is very rewarding that the MAS community now consists of thousands of researchers, and there are a large number of workshops, conferences, and journals featuring autonomous agents and multi-agent research. We are pleased to have played a role in the creation and success of this community. In total, the Lab published over 400 papers (many highly cited), graduated 35 PhD students (many are in influential positions in academia and industry) with 94 PhD descendants, and hosted a large number of extended visits by international researchers.

In the final analysis, the longevity and success of the MAS Lab as an exciting and innovative place to investigate issues in autonomous agents and multi-agent systems stemmed from all the wonderful graduate students, colleagues, and visitors who worked with us over the years. We also must extend a special thanks to Michele Roberts, the Lab's business manager/grant administrator during the majority of its history, whose hard work, intelligence, and foresight kept things functioning smoothly.