Agents for the Masses?

• Authors: Jeffrey M. Bradshaw;Mark Greaves;Heather Holmback;Tom Karygiannis;Wayne Jansen;Barry G. Silverman;Niranjan Suri;Alex Wong
• Venue: IEEE Intelligent Systems
• Year: 1999

Agent technology is in a state of paradox. The field has neverenjoyed more energy and concomitant research progress, and yet therate of uptake of new research results in fielded systems has beenglacially slow. The few agents in the real world of everydayapplications generate more heat than enlightenment; most are easilyconfused, few collaborate except in trivial prearranged fashion,and all enjoy little freedom of movement. Significantly, thecurrent trapped state of our agents has less to do with lack ofmobility mechanisms than with their unpreparedness to work fully inthe open world of cyberspace and to interoperate outside of atightly circumscribed sphere of agent platforms and domains. Thekinds of agents that we want-citizens of the wired world, equippedwith stamped passports and Berlitz travelers guides explainingforeign phrases and places that allow them to hail, meet, and greetagents of any sort in the open landscape of the Internet and, ifnot able to team up on a project, at least able to ask intelligiblyfor directions-these kinds of agents, alas, exist today only in ourimaginations (and, of course, in the vision sections of ourresearch proposals).Actually building the sophisticated agent-based systems of thefuture will require research advances on at least three fronts: Wemust continue work on agent theory so that many currentlyunanswered questions about the scope and limitations of alternativeapproaches to agent design can be addressed. We must make agentframeworks and infrastructure powerful, interoperable, and secureenough to support robust large-scale coordinated problem-solvingactivity.Perhaps more importantly, we must develop new sorts of tools tohelp nonspecialists unlock the power of agent technology.The good news is that things are progressing well on the firsttwo fronts. Various initiatives are beginning to provide an earlypreview of the faster, more reliable, and more secure versions ofthe next-generation Internet that our large-scale visions require.Middleware and Internet technologies and standards are now maturingto the point that agent framework developers can rely onoff-the-shelf products as a ready substrate to their own work,rather than creating ad hoc alternatives from scratch. Advances inthe difficult theoretical issues of dynamic agent communication,coordination, and control are beginning to let us better understandhow to deploy large numbers of agents with confidence. Finally,recent work in theory and infrastructure has yielded exciting newkinds of blueprints for future systems that lie beyond theevolutionary development of current technologies. From grids toJini (http://java.sun.com/products/jini), these approaches aim toprovide a universal source of dynamically pluggable, pervasive, anddependable computing power, while guaranteeing levels of securityand quality of service that will make new classes of agentapplications possible.However, a large and ugly chasm still separates the world offormal theory and infrastructure from the world of practicalnuts-and-bolts agent-system development-this is where the thirdresearch front comes in. If agent technology is ever to become aswidely used as ordinary object technology is today, we must createnew sorts of tools to help non-guru developers bridge the gapsbetween theory, plumbing, and practice. Currently, fullappreciation of leading-edge developments in agent theory andframeworks requires sophisticated knowledge of speech-act theory,formal semantics, linguistic pragmatics, logic, security design,Internet and middleware technologies, distributed computing,planning, and other disciplines that are not likely to be presentin a typical developer's skill set. Without good tools, rapidadvances in theory and infrastructure might paradoxically attenuaterather than accelerate the adoption of agent technology as membersof the developer community spin their wheels or ultimately give upin disgust. Hence we must ask: Is it possible to make developmentof sophisticated agents simple enough to be practical?Fortunately, the agents community has not completely neglectedthe question of tools. The DARPA CoABS program and complementaryinitiatives in Europe and Asia are vigorously supporting researchto accelerate the development of scalable interoperable agenttheory and tools, and are promoting the eventual adoption ofstandards through bodies such as FIPA. As part of these efforts, weare working to extend theory and create tools in two areas: agentcommunication and agent management. This article discusses ourcurrent research directions and preliminary results in each ofthese areas.