DSS AGENTS AND MULTI-AGENTS
13.8 DSS AGENTS AND MULTI-AGENTS
DSS AGENTS Some of the agents described earlier can be classified as problem-solving or DSS
agents (see Kvarroov.com). A framework for DSS agents has been proposed by Hess et al. (2000), who distinguish five types: data monitoring, data gathering, modeling,
TABLE 13.1 Example Agents IJtili/ed in the Extension of the Holsapplc and Whinston (19%) Manufacturing Firm DSS. Agent Agent Essential Characteristics Reference Point
Autonomous Homeostatic
Employer/
Domain Data monitoring
Report when any
Monitor the current Vendor site on an price change
Stay at supplier's
Capable of
User
rates of the three extranet crosses given
site "forever" or
detecting vendor
types of resources threshold values
as long as the
price changes
vendor supplies
and report on
parts
them
Data gathering Report discovery
Look for alternate Travel to directory of potential
Lifetime of the DSS Capable of
User
examining
vendors of
suppliers of
specific part; if manufactured
directory sites
and
found, send
message back reasonable prices
parts at
understanding
language used
with name and
location of source Modeling
there
Maintain "optimal" Lifetime of the DSS Capable of receiving Domain manager When notified by Model base price and resource
D M A , formulate management policies; report
inputs from the
agent ( D M A )
an LP model, system (MBMS) significant dollar
domain manager
of DSS consequences
agent ( D M A ) and
solve it using
passing results
Excel's solver,
back to the D M A
and report solution to D M A
Domain managing Monitor location
Monitor all other Database (say, in the
Lifetime of the DSS Capable of
User
agents (both local management DBMS)
and tasks of both
communicating
local and remote
system (DBMS) agents functioning
with agents (even
and remote)
acting on behalf of DSS (similar on behalf of
at a distance) and
of the domain; agents exist in domain activities;
keeping track of
the MBMS and respond to all
their whereabouts
trigger
DGMS) messages.
appropriate
actions on hearing from them
Preference Learn a specific
Record whether Dialog generation learning
"Lifetime" of a user
Capable of
User
and based on the
user's preferences
of the DSS, even
observing user
specific user
takes modeling management actual history of
across different
actions and
agent's advice or system (DGMS) user/DSS
sessions
storing them
proceeds on own of D S S interactions
C H A P T E R 13 INTELLIGENT SYSTEMS OVER THE INTERNET
three major characteristics and three reference points. This table, based on the work of Holsapple and Whinston (1996), presents examples from a manufacturing firm DSS.
Furthermore, Hess et al. have proposed a general framework in which they map the five types of agents against the three major components of DSS (data, modeling, user interface). This framework is shown in Figure 13.6, and it can be used as a guide in agent development and research.
MULTI-AGENTS
Multi-agent systems are a computer-based environment that contains multiple soft- ware agents to perform certain tasks. The theoretical basis for multiple agents started with research in a field called distributed artificial intelligence (DAI), which basically represents the intelligent part of distributed problem-solving. D A I is the study of dis-
tributed but centrally designed Al systems (Avouris and Gasser, 1992) and involves the design of a multiple-agent distributed system with a problem to solve or a task to accomplish. The issue is how to perform in an effective and efficient manner.
FIGURE
1 3 . 6 T H E AGENT-ENHANCED GENERAL D S S FRAMEWORK
p S L Stand- i*
M BM S Modeling Packages
Alone
/DMA\
Metalmodeling
C T f c > Modeling AgunLs L j Proxy Agent
' •••' Proxy Agent Prfjferenc.fi Preference
Learning Agent f , L e a r n i n g Agent
Contact Agent
P A R T IV INTELLIGENT DECISION SUPPORT SYSTEMS
The D A I approach decomposes the task into subtasks, each of which is addressed by an agent. Therefore, in distributed problem-solving it is assumed that there is a sin- gle body that is able, at design time, to directly influence the preferences of all the agents in the system ( O ' H a r e and Jennings, 1996). The infrastructure of D A I can be constructed with an architecture known as a blackboard (Avouris and Gasser, 1992). Nute et al. (1995) provide an example of how to perform forest management with a blackboard architecture written in P R O L O G . Shih and Srihari (1995) describe D A I in manufacturing-system control.
However, distributed artificial intelligence systems differ f r o m multi-agent sys- tems. There is no single designer standing behind all of the agents in a multi-agent sys- tem. The agents can be working toward different goals, even contradictory goals, and sometimes in parallel; they can cooperate or compete with each other (Decker et al.,
1999). In a D A I system, an agent acting in a particular way is good for the system as a whole, which is not necessarily the case in a multi-agent system. However, by using incentives, it is possible to influence the agents in a multi-agent system. For example, Chi and Turban (1995) proposed a DAI system for an EIS. Wang et al. (1996) define a model of an autonomous agent in a multi-agent environment, focusing on belief-state models of the agents and the changes that communication forces on their belief states. Chau et al. (2002) used the multi-agent approach to Web mining. O ' H a r e and O ' G r a d y (2003) proposed a multi-agent system for intelligent content delivery. This should lead to better communication so that the agents can solve a problem cooperatively in a dis- tributed open system. The agent environment is called a multi-agent processing envi- ronment (MAPE). Figure 13.7 shows an example of a multi-agent system architecture called Genie, in which the identification agent identifies p r o p e r user, the calendar agent schedules events, and the Web agent interacts with the user through Web-based user interfaces (Riekki et al., 2003).
In a multi-agent system, for example, a customer may want to place a long-distance call. O n c e this information is known, agents representing the carriers submit bids simultaneously. The bids are collected, and the best bid wins. In a complex system, the customer's agent may take the process one step f u r t h e r by showing all bidders the offers, allowing them to rebid or negotiate. This process is currently accomplished man- ually by increasing the number of companies that place projects and subassemblies up for bids in business-to-business electronic commerce (Turban et al., 2000).
F I G U R E 1 3 . 7 GENIE THE N E T : A SAMPLE MULTI-AGENT SYSTEM
C H A P T E R 13 INTELLIGENT SYSTEMS OVER THE INTERNET
A complex solution is decomposed into subproblems, each of which is assigned to an agent that works on the problem independently of others and is supported by a knowledge base. Information is acquired and interpreted by knowledge-processing
agents that use deductive and inductive methods as well as computations. The data are refined, interpreted, and sent to the coordinator, who transfers to the user interface whatever is relevant to the user's specific inquiry or need. Multimedia agents can orga- nize the presentation to fit individual executives. If no existing knowledge is available to answer an inquiry, knowledge creating and collecting agents of various types are triggered.