Intelligent Software Agents and Multi-Agent Systems

Transcription

1 2126 &DWHJRU\$UWL FLDO,QWHOOLJHQFH Intelligent Software Agents and Multi-Agent Systems Milan Stankovic Uros Krcadinac Vitomir Kovanovic Jelena Jovanovic INTRODUCTION Agent-based systems are one of the most important and exciting areas of research and development that emerged in information technology (IT) in the past two decades. In a nutshell, an agent is a computer program that is capable RISHUIRUPLQJDÀH[LEOHDXWRQRPRXVDFWLRQLQW\SLFDOO\ dynamic and unpredictable domains (Luck, McBurney, Shehory, & Willmott, 2005). Agents emerged as a response of the IT research community to the new data-processing requirements that traditional computing models and paradigms were increasingly incapable to deal with (e.g., the huge and ever-increasing quantities of available data). Many IT researchers believe that agents represent one of the most important software paradigms that have emerged since the object orientation. From the historic point of view, the agent-oriented research and development (R&D) originates from different GLVFLSOLQHV8QGRXEWHGO\WKHPDLQFRQWULEXWLRQWRWKH HOG RIDXWRQRPRXVDJHQWVFDPHIURPDUWL FLDOLQWHOOLJHQFH$, Ultimately, AI is all about building intelligent artifacts and if these artifacts sense and act in some environment, then they can be considered agents (Russell & Norvig, 1995). Also, object-oriented programming (Booch, 2004), concurrent object-based systems (Agha, Wegner, & Yonezawa, 1993), and human-computer interaction (Maes, 1994) are HOGVWKDWFRQVWDQWO\GULYHIRUZDUGWKHDJHQW5 'LQWKH last few decades. In addition, the concept of an agent has become important in a diverse range of sub-disciplines of IT, including software engineering, computer networks, mobile systems, control systems, decision support, information retrieval and management, electronic commerce, and many others. Agents are being used in an increasingly wide variety of applications ranging from comparatively small systems such as SHUVRQDOL]HGHPDLO OWHUVWRODUJHFRPSOH[PLVVLRQFULWLFDO V\VWHPVVXFKDVDLUWUDI FFRQWURO BACKGROUND Even though there is no universal consensus over some key GH QLWLRQVLQWKH HOGLWLVLQWXLWLYHO\FOHDUZKDWDQ³DJHQW LV2QHRIWKHPRVWZLGHO\XVHGGH QLWLRQVVWDWHVWKDW³an agent is a computer system, situated in some environment, WKDWLVFDSDEOHRIÀH[LEOHDXWRQRPRXV DFWLRQLQRUGHUWR PHHWLWVGHVLJQREMHFWLYHV (Jennings, Sycara, & Wooldridge, 1998, p. 8). 7KHUHDUHWKUHHNH\FRQFHSWVLQWKLVGH QLWLRQsituatedness, autonomy, and ÀH[LELOLW\. Situatedness means that an agent receives sensory input from its environment and that it can perform actions which change the environment in some way. Autonomy is seen as the ability of an agent to act without the direct intervention of humans and that it has control over its own actions and internal state. In addition, the autonomy implies the capability of learning from experience. By ÀH[LELOLW\, we mean the agent s ability to perceive its environment and respond to changes in a timely fashion; it should be able to exhibit opportunistic, goal-directed behaviour and take the initiative whenever appropriate. Also, an agent should be able to interact with other agents and humans, thus to be social. Some authors emphasize Copyright 2009, IGI Global, distributing in print or electronic forms without written permission of IGI Global is prohibited.

2 the importance of the concept of rationality, which will be discussed in the next section. Moreover, agent technologies can be considered from three perspectives (Luck, McBurney, Shehory, & Willmott, 2005): As a design metaphor, agents offer designers a way of structuring an application around autonomous, communicative elements; As a source of technologies, agent-based computing VSDQVDUDQJHRIVSHFL FWHFKQLTXHVDLPHGDWVXSSRUWing interactions among entities in dynamic and open environments; and As a simulation tool, multi-agent systems offer robust models for representing complex and dynamic realworld environments, such as economies, societies and bio-systems. INTELLIGENT SOFTWARE AGENTS Agents and Environments Agents can be viewed as software entities that perceive their environment through sensors and act upon that environment through actuators (Russell & Norvig, 1995). There is an obvious analogy with a human agent who has ears, eyes, and other organs as sensors, and arms, legs, and other organs as actuators. When we refer to an agent s perceptual inputs, we refer to the agent s percepts. An agent typically collects its percepts during the time, so its action in any moment generally depends on the whole sequence of percepts up to that moment. If we could generate a decision tree for every possible SHUFHSWVHTXHQFHRIDQDJHQWZHFRXOGFRPSOHWHO\GH QH the agent s behavior. Strictly speaking, we would say that ZHKDYHGH QHGWKHagent function that maps any sequence RISHUFHSWVWRWKHFRQFUHWHDFWLRQ7KHSURJUDPWKDWGH QHV the agent function is called the agent program. These two Figure 1. Agent and environment S e n so rs A gent E n viro n m e n t A ctu a to rs concepts are different; the agent function is a formal description of the agent s behavior. The agent program is a concrete implementation of that formalism. As Russell and Norvig (1995) stipulate, one of the most desirable properties of an agent is its rationality. We say that an agent is rational if it always does the action that will cause the agent to be the most successful. The rationality of an agent depends on: 7KHSHUIRUPDQFHPHDVXUHWKDWGH QHVZKDWLVDJRRG action and what is a bad action; The agent s knowledge about the environment; The agent s available actions; The agent s percept history. 2QHRIWKHPRVWFLWHGGH QLWLRQVRIDUDWLRQDODJHQWLV for each possible percept sequence, a rational agent should select an action that is expected to maximize its performance measure, given the evidence provided by the percept sequence and whatever built-in knowledge the agent has. (Russell & Norvig, 1995, p. 36) 7KHPDLQFKDOOHQJHLQWKH HOGRIintelligent software agents is to develop an agent program that implements the desired functionalities. Since it is a computer program, we need to have some computing device with appropriate sensors and actuators on which the agent program will run. We call this agent architecture. So an agent is essentially made of two components: the agent architecture and the agent program. The Types of Agents When we deal with the structure of an agent, we consider various implementation models for agent development. There are several basic types of agents with respect to their structure (Russell & Norvig, 1995). The simplest kind of agents are the VLPSOHUHÀH[DJHQWV. Such an agent only reacts to its current percept, totally ignoring its percept history. When a new percept is received, a UXOHWKDWPDSVWKDWSHUFHSWWRDQDFWLRQLV UHG6XFKUXOHV are known as condition-action rules. 0RGHOEDVHGUHÀH[DJHQWV are more powerful agents, because they maintain some sort of internal state of the environment that depends on the percept history. For maintaining this sort of information, an agent must have two types of knowledge: (1) how the environment evolves, and (2) how its actions affect the environment. Goal-based agents have some sort of goal information that describes desirable states of the world. Such an agent s decision making process is fundamentally different, because when a goal-based agent is considering performing an action it is asking itself would this action make me happy? along I 2127

3 with the standard what this action will have as a result?. Utility-based agents use a utility function that maps each state to a number that represents the degree of happiness. They are able to perform rationally even in the situations when WKHUHDUHFRQÀLFWLQJJRDOVDVZHOODVZKHQWKHUHDUHVHYHUDO goals that can be achieved, but none with certainty. Learning agents do not have a priori knowledge of the environment, but learn about it. This is advantageous because these agents can operate in unknown environments and to a certain degree facilitates the job of developers because they do not need to specify their whole knowledge base. Multi-Agent Systems and Environments With an agent-oriented view of the world, it soon becomes FOHDUWKDWDVLQJOHDJHQWLVLQVXI FLHQW0RVWUHDOZRUOG problems require or involve multiple agents: to represent the decentralized nature of the problem, multiple perspectives, or competing interests. Systems composed of multiple autonomous components (agents) are considered multi-agent systems (MAS), and historically belong to GLVWULEXWHGDUWL FLDOLQWHOOLJHQFH (Bond & Gasser, 1998). 0$6FDQEHGH QHGDVDORRVHO\ coupled network of problem solvers that work together to solve problems that are beyond the individual capabilities or knowledge of each problem solver (Durfee & Lesser, 1989). In an MAS, each agent has incomplete information or capabilities for solving the problem and thus has a limited viewpoint. There is no global system control, the data is decentralized, and the computation is asynchronous. In addition to MAS, there is also the concept of a multiagent environment, which can be seen as an environment that includes more than one agent. Thus, it can be cooperative, or competitive, or a combined one. In an MAS and a multi-agent environment, the individual agents need to interact with one another (socialization) either to achieve their individual objectives, or to manage the dependencies that ensue from being situated in a common environment. These interactions range from simple semantic interoperation (exchanging comprehensible communications), through client-server interactions (the ability to request that a particular action is performed), to rich social interactions (the ability to cooperate, coordinate, and negotiate about a course of action). Agent communication is achieved by exchanging messages represented by mutually understandable syntax and containing mutually understandable semantics. In order WR QGDFRPPRQJURXQGIRUFRPPXQLFDWLRQDQ agent communication language (ACL) should be used to provide mechanisms for agents to negotiate, query, and inform each other. The most important such languages today are KQML (Knowledge Query and Manipulation Language) (ARPA Knowledge Sharing Initiative, 1993) and FIPA ACL (FIPA, 1997) proposed by FIPA (IEEE Foundation for Intelligent 3K\VLFDO$JHQWVKWWSZZZ SDRUJ Because of the obstacles stemming from heterogeneous QDWXUHRIDJHQWVLQYROYHGLQFRPPXQLFDWLRQHJ QGLQJRQH another), there is a need for middle-agents, which facilitate cooperation among agents and connect service providers with service requesters in the agent world. These agents are useful in various roles, such as matchmakers or yellow page agents that collect and process service offers ( advertisements ), blackboard agents that collect requests, and brokers that process both (Sycara, Decker, & Williamson, 1997). Usability Aspects The obvious advantage of agents is the clarity that the sole GH QLWLRQEULQJVZLWKUHJDUGVWRWKHUHSUHVHQWDWLRQRIIXQFtionality. When an end user is presented with certain software DVDQDJHQWIRUH[DPSOHDQDJHQWIRU OWHULQJVSDPHPDLO its functionalities are naturally obvious and the user experiences the system as a personal assistant that will work for him without much need for supervision. Concerning the agents that communicate directly with end-users, the concept of agents shifts the paradigm of user interaction from simple user computer manipulation to assigning tasks to the computer. Within this scenario, the user assigns the tasks and acquires perception about an agent as an anthropomorphic software entity. Controversies and Pitfalls of Agent Development It is not uncommon that some users, not having enough knowledge about intelligent agents express concern (even fear) about their usage. They tend to confuse them with software daemons that send spam messages, or with software viruses that can damage their systems. Others perceive danger in the amount of independency that agents possess. Lead by skepticism, they are likely to discard the concept of autonomous software agents in favor of software that allows complete control over every operation it executes. This point of view goes along with arguments that giving too much decision making autonomy to a software entity would lead to dependency on such an entity s will. Dangers arising from such autonomy are subject of various VFLHQFH FWLRQZRUNVHJ$6SDFH2G\VVH\,QWKH HOGRIVRIWZDUHHQJLQHHULQJDJHQWVDUHDOVRFULWLcized. The critiques are primarily concerned with traps in which a developer may fall when developing agents. Those pitfalls include applying agent technology in areas where other software engineering concepts are more suitable, the inappropriate use of other AI techniques, the inadequate number of agents in an agent-system, and so forth (Wooldridge & Jennings, 1998). 2128

4 APPLICATIONS Intelligent software agents are a suitable software engineering concept in a wide variety of application domains. Their DXWRQRP\TXDOL HVWKHPIRUVXFFHVVIXODSSOLFDWLRQLQWKH HOGVVXFKDVWUDI FDQGWUDQVSRUWDWLRQFRQWUROFRPSXWHU networks and telecommunication control, healthcare, and so forth. Process control software systems require an entity that can supervise a process (e.g., production process) and react ZKHQQHHGHG5HDFWLYHDQGUHVSRQVLYHDJHQWVSHUIHFWO\ WWKH QHHGVRIVXFKDWDVN([DPSOHGRPDLQVLQWKLV HOGLQFOXGH production process control, climate monitoring, spacecraft control, and monitoring nuclear power plants. An important application domain is information gathering, where agents are used to search trough heterogeneous information sources (e.g., World Wide Web) and acquire relevant information for their users. One of the most common domains is Web browsing and search, where agents are used to adapt the content (e.g., search results) to the users preferences and offer relevant help in browsing. Agent cooperation opens possibilities for applying multiagent systems to solving constraint satisfaction problems. Auction negotiation model, as a form of communication, HQDEOHVDJURXSRIDJHQWVWR QGJRRGVROXWLRQVE\DFKLHYing agreement and making mutual compromises in case of FRQÀLFWLQJJRDOV6XFKDQDSSURDFKLVDSSOLFDEOHWRWUDGLQJ systems, where agents act on behalf of buyers and sellers. Financial markets, as well as meeting scheduling, travel arrangement composing, and fault diagnosing also represent SURPLQHQW HOGVIRUDJHQWDSSOLFDWLRQ Intelligent tutoring systems often include pedagogical agents, which represent software entities constructed to present the learning content in a user-friendly fashion and monitor the user s progress through the learning process. These agents are responsible for guiding the user and suggesting additional learning topics related to the user s needs (Devedzic, 2006). Mobility Agents for People with Cognitive Disabilities Mobility agents is an agent-based architecture that helps a person with cognitive disabilities to travel using public transportation. Agents are used to represent transportation SDUWLFLSDQWVEXVHVDQGWUDYHOHUVDQGWRHQDEOHQRWL FDWLRQ of bus approaching and arrival. Information is passed to the traveler using a rich multimedia interface, via a handheld GHYLFH&XVWRPL]DEOHXVHUSUR OHVGHWHUPLQHWKHPRVWDSSURpriate modality of interaction (voice, text, and pictures) based on the user s abilities (Repenning & Sullivan, 2003). This architecture actually imposes a personal agent to guide users with cognitive disabilities and take care that abstract goals as go home are translated into concrete directions. To achieve this, an agent needs to collect information DERXWXVHUVSHFL FORFDWLRQVDQGPXVWEHDEOHWRVXJJHVW the right bus for the particular user s current location and destination. BluScreen BluScreen is an intelligent public display, developed at the University of Southampton, in order to adapt the selection of adverts for display to the present audience detected by Bluetooth technology. Customization of content is achieved using history information of past users exposure to certain sets of adverts, in order to predict which advert is likely to gain the highest attention. The system is implemented as a multi-agent auction-based mechanism. Each agent represents a stakeholder wishing to advertise, and it is provided with a bidding strategy that utilizes heuristics to predict future advert exposure, based on the expected audience composition. These agents compete in an auction to gain advertising space, ensuring that the most suitable advertising content is selected (Payne, David, -HQQLQJV 6KDUL BluScreen employs the concept of agents socialization to achieve context aware, intelligent behavior of the system as a whole, relying on particular agents autonomy to act on behalf of stakeholders and maximize their satisfaction. Talaria Talaria System (The autonomous lookup and report internet agent system), named after the Greek Messenger God Hermes s winged sandals, is a multi-agent system, developed for academic purposes at the, School of Business Administration ( yu/talaria/). It was built as a solution to the common problem of gathering information from diverse Web sites that do not provide RSS feeds for news tracking. The system was implemented using the JADE modeling framework ( 7DODULDLQWHJUDWHVLQIRUPDWLRQJDWKHULQJDQG OWHULQJ in the context of supporting a user to manage her/his Web interests. The system provides each user with a personal agent, which periodically monitors the Web sites that the user expressed interest in. The agent informs its user about UHOHYDQWFKDQJHV OWHUHGE\DVVXPHGXVHUSUHIHUHQFHVDQG default relevance factors. Human-agent communication is implemented via , so that a user can converse with her/his agent in natural language, whereas the agent heuristically interprets concrete instructions from the mail text (e.g., change site list or kill yourself ). Human-like interaction, autonomy-related aspects of this system, and acting on behalf of the user emphasize the usability advantages of this agent-based software. I 2129

5 FUTURE TRENDS The development of agent technologies is taking place within a context of broader visions and trends in IT, which are DERXWWRGULYHIRUZDUGWKHZKROH HOGRILQWHOOLJHQWDJHQWV We especially emphasize the semantic Web, Web services, ambient intelligence and peer-to-peer computing. The semantic Web is the vision of the future Web based RQWKHLGHDWKDWWKHGDWDRQWKH:HEFDQEHGH QHGDQG linked in such a way that it can be used by machines for the automatic processing and integration (Berners-Lee, Hendler, & Lassila, 2001). The key to achieving this is by augmenting Web pages with descriptions of their content in such a way that it is possible for machines to reason automatically about that content. The semantic Web offers a solid ground for further development of the agent technologies as well as successful deployment of a variety of agent-based applications. Actually, we share the opinion that the semantic Web itself will be a form of intelligent infrastructure for agents, allowing them to understand the meaning of the data on the Web. The other important drivers of agent development are the Web services and service-oriented computing, which are likely to become the dominant base technology in the foreseeable future. The Web service technology provides standard means for establishing interoperability between heterogeneous software applications that run on a variety of different platforms. Accordingly, this technology is almost ideal for use in supporting agent interactions in a multi-agent system. Moreover, an agent-oriented view of Web services is gaining an increasing attention, since agent-based systems are naturally seen as provider and consumer Web services environments (Booth et al., 2004). The concept of ambient intelligence, being a vision that describes a shift away from PCs to a variety of devices which are unobtrusively embedded in our environment and which are accessed via intelligent interfaces, with no doubt require agent-like technologies in order to achieve autonomy, distribution, adaptation, and responsiveness (Booth et al., 2004). Peer-to-peer (P2P) computing, presenting networked applications in which every node is in some sense equivalent to all others, tends to become more complex in the future. Auction mechanism design, agent negotiation techniques, increasingly advanced approaches to trust and reputation, and the application of social norms, rules and structures presents some of the agent technologies that are about to become relevant in the context of P2P computing (Booth et al., 2004). Almost each of today s compelling IT visions and trends such as the abovementioned, but also grid computing, complex systems, and many more, will require agent technologies (or something similar to them), before being fully implemented. Agent technologies are upstream of these visions and mission-critical to them. However, to be able to support these visions, the agent-based computing needs further development and strengthening. Some considerable challenges have remained in the agent-based world, among which we emphasize the lack of sophisticated software tools, techniques and methodologies that would support the VSHFL FDWLRQGHYHORSPHQWLQWHJUDWLRQDQGPDQDJHPHQWRI agent systems. CONCLUSION 5HVHDUFKDQGGHYHORSPHQWLQWKH HOGRILQWHOOLJHQWDJHQWVDQG multi-agent systems is rapidly expanding. It can be viewed as a melting pot of different ideas originating from diverse DUHDVVXFKDVDUWL FLDOLQWHOOLJHQFHREMHFWRULHQWHGV\VWHPV software engineering, distributed computing, economics, and so forth. At its core is the concept of autonomous agents interacting with one another for their individual and/or col- OHFWLYHEHQH W $QXPEHURIVLJQL FDQWDGYDQFHVKDYHEHHQPDGHRYHU the past two decades in design and implementation of individual autonomous agents, and in the way in which they interact with one another. These concepts and technologies DUHQRZ QGLQJWKHLUZD\LQWRFRPPHUFLDOSURGXFWVDQG real-world software solutions. Future IT visions share the common need for agent technologies and prove that agent technologies will continue to be of vital importance. REFERENCES Agha, G., Wegner, P., & Yonezawa, A. (Eds.). (1993). Re- VHDUFKGLUHFWLRQVLQFRQFXUUHQWREMHFWRULHQWHGSURJUDPming. Cambridge, MA: The MIT Press. ARPA Knowledge Sharing Initiative. (1993). 6SHFL FDWLRQ of the KQML agent-communication language plus example agent policies and architectures. Retrieved January 30, 2007, from Berners-Lee, T., Hendler, J., & Lassila, O. (2001, May). The semantic Web. 6FLHQWL F$PHULFDQ, Bond, A. H., & Gasser, L. (Eds.). (1998). Readings in GLVWULEXWHGDUWL FLDOLQWHOOLJHQFH. San Mateo, CA: Morgan Kaufmann Publishers. Booch, G. (2004). 2EMHFWRULHQWHGDQDO\VLVDQGGHVLJQ nd ed.). MA: Addison-Wesley. Booth, D., Haas, H., McCabe, F., Newcomer, E., Champion, M., Ferris, C., & Orchard, D. (2004, February). Web services architecture. W3C working group note 11. Retrieved January 30, 2007, from

6 Devedzic, V. (2006). Semantic Web and education. Berlin, Heidelberg, New York: Springer. Durfee, E. H., & Lesser, V. (1989). Negotiating task decomposition and allocation using partial global planning. In L. Gasser, & M. Huhns (Eds.), 'LVWULEXWHGDUWL FLDOLQWHOOLJHQFH Volume II ( pp ). London: Pitman Publishing and San Mateo, CA: Morgan Kaufmann. FIPA. (1997). 3DUWRIWKH),3$VSHFL FDWLRQV$JHQW communication language. Retrieved January 30, 2007, from KWWSZZZ SDRUJVSHFV SD2&$KWPO Jennings, N. R., Sycara, K., & Wooldridge, M. (1998). A roadmap of agent research and development. Autonomous Agents and Multi-Agent Systems, 1(1), Luck, M., McBurney, P., Shehory, O., & Willmott, S. (2005). Agent technology: Computing as interaction. Retrieved January 30, 2007, from Maes, P. (1994) Agents that reduce work and information overload. Communications of the ACM, 37(7), D\QH75'DYLG(-HQQLQJV15 6KDUL 0 $XFWLRQPHFKDQLVPVIRUHI FLHQWDGYHUWLVHPHQWVHOHFWLRQ on public displays. In B. Dunin-Keplicz, A. Omicini, & J. Padget (Eds.), Proceedings of the Fourth European Workshop on Multi-Agent Systems (in press). Repenning, A., & Sullivan, J. (2003). The pragmatic Web: Agent-based multimodal Web interaction with no browser in sight. In G. W. M. Rauterberg, M. Menozzi, & J. Wesson (Eds.), Proceedings of the Ninth International Conference on Human-Computer Interaction (pp ). Amsterdam, The Netherlands: IOS Press. Russel, S. J., & Norvig, P. (1995). $UWL FLDOLQWHOOLJHQFH$ modern approach. New Jersey: Prentice-Hall. Sycara, K., Decker, K., & Williamson, M. (1997). Middleagents for the Internet. In M. E. Pollack (Ed.), Proceedings RIWKH)LIWHHQWK,QWHUQDWLRQDO-RLQW&RQIHUHQFHRQ$UWL FLDO Intelligence (pp ). Morgan Kaufmann Publishers. Wooldridge, M., & Jennings, N. R. (1995). Intelligent agents: Theory and practice. The Knowledge Engineering Review, 10(2), Wooldridge, M., & Jennings, N. R. (1998). Pitfalls of agentoriented development. In K. P. Sycara, & M. Wooldridge (Eds.), Proceedings of the Second International Conference on Autonomous Agents (pp ). Minneapolis: ACM Press. KEY TERMS Actuators: Software component and part of the agent used as a mean of performing actions in the agent environment. Agent Communication Language (ACL): Language XVHGE\DJHQWVLQH[FKDQJHRIPHVVDJHVGH QLQJFRPPRQ syntax for cooperation between heterogeneous agents. Agent Percepts: Every information that an agent receives through its sensors, about the state of the environment or any part of the environment. Intelligent Software Agent: An encapsulated computer system that is situated in some environment and that is FDSDEOHRIÀH[LEOHDXWRQRPRXVDFWLRQLQWKDWHQYLURQPHQW in order to meet its design objectives (Wooldridge & Jennings, 1995). Middle-Agents: Agents that facilitate cooperation among other agents and typically connect service providers with service requesters. Multi-Agent System (MAS): A software system com- SRVHGRIVHYHUDODJHQWVWKDWLQWHUDFWLQRUGHUWR QGVROXWLRQV of complex problems. Sensors: Software component and part of the agent used as a mean of acquiring information about current state of the agent environment (i.e., agent percepts). I 2131