Software Agents

1
Software Agents

• Catherine L. Hansen
• Jessica C. Baker
• CPS 001

2
What is an agent?

• There are many possible definitions. There are
  also several characteristics that most agents
  possess.
• -Autonomous
• -Adaptive/Learning
• -Mobile
• -Persistent
• -Goal-oriented
• -Communicative/Collaborative
• -Flexible -Active/Proactive
• -Small in size
• But what do these words mean, and more
  importantly, how do they come together to define
  the concept of an AGENT?

3
Agent Characteristics

• Here is how a few computer scientists have
  defined agent -The AIMA Agent AIMA stands for
  Artificial Intelligence A Modern Approach by
  Russell and Norvig (1995.) They define an agent
  as Anything that can be viewed as perceiving its
  environment through sensors, and acting upon that
  environment through effectors. The problem with
  this definition its too general. It can apply
  to any organisms (including humans and most
  animals) that interact with their environment
  through sensation and perception. How about
  something more specific?
• -The Maes Agent Maes (1995) describes autonomous
  agents as Computational systems that inhabit
  some complex, dynamic environment, sense and act
  autonomously in this environment, and by doing so
  realize a set of goals or tasks for which they
  are designed.

4
How is a software agent different from a program?

• Using the Maes definition as a starting point,
  the difference between a software agent and a
  program is that an agent is designed to take
  action that is not necessarily anticipated by the
  programmer, in a complex and changing
  environment. It also interacts with its
  environment, and often learns from it. A program,
  in contrast, is designed to take a predictable
  action with known inputs, and to stop running
  after this action is completed.
• An agent, on the other hand, is persistent.
• Smith, Cypher, and Spohrer (1994) defines a
  software agent in this manner a persistent
  software entity dedicated to a specific purpose.
  Persistent distinguishes agents from
  subroutines agents have their own ideas about
  how to accomplish tasks, their own agendas.

5
More Definitions

• As to learning from a dynamic environment,
  Hayes-Roth (1995) specifies three functions that
  an intelligent software agent must continuously
  perform perception of dynamic conditions in the
  environment action to effect conditions in the
  environment and reasoning to interpret
  perceptions, solve problems, draw inferences, and
  determine actions.
• Next question How are different kinds of agents
  classified?

6
A Working Taxonomy of Agents

• It may be easier to understand the concept of a
  software agent by keeping in mind that while all
  agents are programs, all programs are not
  necessarily agents.
• Although by definition, all agents must be
  reactive, autonomous, goal-oriented and
  temporally continuous, they can also be
  sub-classified by other properties specifically
  assigned to a particular agent to help it reach
  its goals.
• The computer scientist Brustolini (1991) designed
  a taxonomy of software agents based on three
  properties
• Regulation Agents these agents only react to and
  regulate input, and neither plan nor learn. They
  can be said to come the closest to basic software
  programs, although their autonomous function
  prevents them from being classified as such.

7
A Working Taxonomy Continued

• Planning Agents to put it simply, they plan.
  There are several methods of planning perhaps
  the most well-known of these is through
  artificial intelligence.
• Adaptive Agents the highest level of agents
  they not only plan, but learn.
• In addition, agents with multiple functions may
  be classified under more than one of these
  definitions.
• Other methods of agent classification may be
  defined via control structures, applications,
  language employed, and environment. Keep in mind
  that none of these taxonomies will necessarily
  cover the waterfront there are as many
  possible ways to classify software agents as
  there are to classify biological organisms.
• Next question In which types of environments do
  software agents function and communicate?

8
Dialogs and Multi-User Domains

• Agents communicate with each other in specially
  designed forums.
• Michael Cohen, the creator of SodaBot, a
  communication environment for software agents,
  defines agents as programs that engage in
  dialogs and negotiate and coordinate transfer of
  information.
• Multi-user domains are another example of a
  communication environment.

9
MUD?

• The first MUDs (multi-user domains) were
  interactive role-playing Dungeons and
  Dragons-type games online. Now, they are forums
  for communication and collaboration, a kind of
  virtual reality in real time, based on
  exchanges of texts. Agents use social
  environments similar to MUDs in order to interact
  with each other.
• Next question how do agents communicate?

10
Communication Mechanisms

• There are three commonly used communication
  mechanisms employed by software agents to
  interact and share information with each other
  within their environment. Communication is
  necessary because on their own, agents are
  relatively small and weak part of their
  usefulness is derived from the ability to share
  information over a network.
• Keep in mind that these three mechanisms are
  employed by threads, programs and subroutines as
  well, but they will be discussed specifically as
  the basic building blocks of agent communication.

11
Procedure Call Mechanism

• 1) A calls B and waits.
• 2) B calls C and waits.
• 3) C returns information for B.
• 4) B returns information for A.
• This method is fast and efficient, and allows the
  user to easily track the flow of execution.
• Example Javas remote method invocation (RMI)
  system.

12
Callback Mechanism

• 1) A calls B and returns.
• 2) B calls C and returns.
• 3) C calls back B with result.
• 4) B calls back A with result.
• Pro permits asynchronous processing agents can
  continue on with their tasks while waiting for
  callback.
• Con agents must halt tasks when information
  arrives complicated and therefore harder for
  user to track flow of execution.
• Examples Javas GUI toolkits (AWT and Swing)

13
Mailbox Mechanism

• 1) A calls B and returns with mailbox.
• 2) B calls C and returns with mailbox.
• 3) C updates the mailbox.
• 4) B updates the mailbox.
• Cons more difficult to implement than other two.
  
• Pros allows for asynchronous programming with
  less confusion (best of both worlds!)

14
Mobile Agents

• Mobile agents do not merely pass information
  around like RMI communication does. They are able
  to navigate through networks and MUD-like
  environments, carrying state information with
  them and gathering new information from other
  agents, the environment and human users. This
  information will often affect the course of their
  future state and behaviorthus, they are
  artificial intelligence agents.
• A successful mobile agent should be able to
  execute on every machine in a network using a
  mobile code system like Java.

15
Benefits of Mobile Agents

• Network bandwidth is a valuable resource. If
  traditional communication mechanisms are employed
  and there are too many clients and servers
  executing transactions over the network,
  bandwidth may be exceeded.
• Mobile network agents help solve this problem by
  handling client/server transactions the agent
  will carry the information as it moves and
  consume less bandwidth than if pieces of the
  information itself were sent.

16
Mobile Agent Architecture

• An agent host allows multiple agents to co-exist,
  execute and communicate with one another
  simultaneously. It also prevents agents from
  interfering with the execution of other agents so
  that the network runs smoothly.
• The system (think of it as the environment in
  which agents exist and interact) includes four
  classes
• Agent class, which defines the agent and exists
  for each agent executing on the system.
• AgentHost class, which defines the host and keeps
  track of all agents in the system.
• AgentInterface class, the main conduit for
  communication between individual agents.
• AgentIdentity class, which allows agents to
  identify other agents to collaborate with.

17
Java and Mobile Agents

• So far, the Java system provides the best
  technology base for a mobile agent communication
  system. Java virtual machines are so ubiquitous
  that agent code will not have to be installed on
  every machine agents may visit the agent classes
  listed above need only be loaded at runtime over
  the network.
• One major related issue being tackled by
  programmers is the problem of agent security in
  the face of such an extensive network. Is it
  possible to protect agents from servers in the
  system that deliberately mis-execute their code?

18
Agent Context

• Agents live and operate within agent hosts. The
  agent hosts move individual agents around and
  enable them to operate on the system. However,
  they do not allow any one agent access to the
  entire application. Instead, they refer each
  agent to a more limited, specific class in which
  to operate, called the agent context.
• The following code both shows what an agent
  context would look like on a Java machine, and
  provides a concrete (albeit small) example of
  actual agent code.

19
An Abbreviated Look at the Agent Class

• Something like this would be used for an agent
  communication system.
• Code source (http//www.javaworld.com/javaworld/j
  w-09-1998/jw-09-howto.html)
• Abstract public class Agent
• /
• The following agent context class will provide
  the agents interface with its environment. The
  agent context is set by the agent host when the
  agent is transferred into a new environment.
• /
• Private transient AgentContext agentcontext
  null
• / This gets the agent context. The reason the
  field is called transient is that a new context
  specific to the agent is created whenever an
  agent is transferred to a new environment.

20
Code for Implementing Agent Class

• It looks something like this
• class AgentContext
• private AgentHost agenthost null
• AgentContext (AgentHost agenthost )
• this.agenthost agenthost
• / The following will send a message and return
  the agents response to the message. The agent
  will call the sendMessage method in order to send
  a message to an agent specified by AgentIdentity.
  After this, the Message class will encapsulate
  the actual message that is sent.
• /
• public MessageResponse
• sendMessage(AgentIdentity agentidentity, Message
  message)
• return new MessageResponse(agenthost,
  agentidentity, message)

21
Conclusion/Summary

• Almost anything that functions autonomously,
  interacting with its environment and other
  similar entities, can be classified as an agent.
  This includes humans, animals and some robots,
  all of which live in the real world. In
  contrast, software agents live on networks,
  computer operating systems, databases, MUDs and
  other virtual environments. Mobile agents, which
  we studied more closely, have a primarily
  communicative function, and make the process of
  network transactions much quicker and more
  efficient.

22
Sources Consulted

• Franklin, Stan, and Art Graesser. Is it an
  Agent, or Just a Program? A Taxonomy for
  Autonomous Agents. University of Memphis
  Institute for Intelligent Systems, 1996.
  http//www.msci.memphis.edu/franklin/AgentProg.ht
  ml
• Groves, Paul. MUDs, MOOs, WOOs and IRC.
  http//www.oucs.ox.ac.uk/ltg/projects/jtap/reports
  /teaching/chapter5.html
• Minar, Nelson. Java as Technology Base for
  Mobile Agents. http//xenia.media.mit.edu/nelson
  /research/dc/node3.html
• Sundsted, Todd. Agents Talking to Agents.
  http//www.javaworld.com/javaworld/jw-09-1998/jw-0
  9-howto.html