Spring Term 200203 ismailtk'unilinz'ac'at

1
Spezielle Kapitel aus TelekooperationIntelligent
Agents

• Ismail Khalil Ibrahim
• Summer Semester 2003

2
Course Unit 4

• Intelligent Agents Architectures ...

3
What is an Agent Architecture?

• Maes
• An architecture proposes a particular methodology
  for building an autonomous agent. It specifies
  how the overall problem can be decomposed into
  sub-problems, i.e. how the construction of the
  agent can be decomposed into the construction of
  a set of component modules and how these modules
  should be made to interact. The total set of
  modules and their interactions has to provide an
  answer to the question of how the sensor data and
  the current internal state of the agent determine
  the actions (effectors outputs) and future
  internal state of the agent. An architecture
  encompasses techniques and algorithms that
  support this methodology.

4
What is an Agent Architecture?

• Kaelbling
• A specific collection of software (or hardware)
  modules, typically designated by boxes with
  arrows indicating the data and control flow among
  the modules. A more abstract view of an
  architecture is as a general methodology for
  designing particular modular decompositions for
  particular tasks

5
What is an Agent Architecture?

• Software Architecture
• describes the high level configuration of a
  systems constituent components and the
  connections that coordinates the activities among
  those components
• Agent Architecture
• is a model of an intelligent information-processin
  g system defining its major subsystems, their
  functional roles, and the flow of information and
  control among them.

6
Abstract Agent Architecture

• introduce a simple formal model of an agent
• Two sets
• Environment states
• S s1, s2,
• Actions
• A a1, a2,
• An agent is a function
• action S ? A
• that maps sequences of environment states to
  actions

7
Properties of environments

• Accessible
• The agent can obtain complete, accurate,
  up-to-date information about the environment
  state
• Deterministic
• Any action has a single guaranteed effect
  there is no uncertainty about the state that will
  result from performing an action
• Episodic
• The performance of an agent is dependent on a
  number of discrete episodes, with no link between
  the performance of an agent in different
  scenarios ? Do not need to think ahead.
• Static
• can be assumed to remain unchanged except by the
  performance of actions by the agent.
• Discrete
• There are a fixed, finite number of actions and
  percepts in it.

8
Non-deterministic Environment

• (Non-deterministic) environment modeled as a
  function
• env S X A ? ? (S)
• which takes the current state of the
  environment s ? S and an action a ? A
  (performed by the agent), and maps them to a
  set of environment states env (s, a)
• those that could result from performing
  action a in state s.

9
Agent Environment

• Interaction of agent and environment is a history
  or run
• a0 a1 a2 an-1 an
• h s0 ? s1 ? s2 ? ? sn ?
• where
• s0 is the initial state of the environment
• an is the nth action that the agent performs
• sn is the nth environment state (which is one
  of the possible results of executing action an-1
  in state sn-1

10
Purely Reactive Agents

• Some agents decide what to do without reference
  to their history
• they base their decision-making entirely on the
  present, with no reference at all to the past
• We call such agents purely reactive
• action S ? A
• Example thermostat
• off if s temperature OK
• action(s)
• on otherwise.

11
Agents with Perception

• The see function is the agents ability to
  observe its environment, whereas the action
  function represents the agents decision making
  process.
• Output of the see function is a percept
• see S ? P
• which maps environment states to percepts, and
  action is now a function
• action P ? A
• which maps sequences of percepts to actions.

12
Agents with State
see
next
environment
state
action
13
Agents with State

• These agents have some internal data structure,
  which is typically used to record information
  about the environment state and history.
• Let I be the set of all internal states of the
  agent.
• The perception function see for a state-based
  agent is unchanged
• see S ? P
• The action-selection function action is now
  defined as a mapping
• action I ? A
• from internal states to actions. An additional
  function next is
• introduced, which maps an internal state and
  percept to an
• internal state
• next I X P ? I

14
Control Loop

• Agent starts in some initial internal state i0
• Observes its environment state s, and generates a
  percept see(s)
• Internal state of the agent is then updated via
  next function, becoming next(i0, see (s))
• The action selected by the agent is action
  (next(i0,see(s))). This action is then performed.
• Goto (2).

15
Utility based Agents

• We build agents in order to carry out tasks for
  us.
• The task must be specified by us
• but we want to tell agents what to do without
  telling them how to do it.
• One possibility associate utilities with
  individual states the task of the agent is then
  to bring about states that maximize utility.
• A task specification is a function V S ? R,
  which associated a real value with every
  environment state.
• Disadvantage difficult to specify a long term
  view when assigning utilities to individual
  states.

16
Utility based Agents

• Another possibility assigns a utility not to
  individual states, but to runs themselves
• V H ? R
• Such an approach takes an inherently long term
  view.
• Other variations incorporate probabilities of
  different states emerging.
• Difficulties with utility-based approaches
• where do the numbers come from?
• we dont think in terms of utilities!
• hard to formulate tasks in these terms.

17
Predicate Task Specification

• A special case of assigning utilities to
  histories is to assign
• 0 (false) or 1 (true) to a run
• If a run is assigned 1, then the agent succeeds
  on that run, otherwise it fails.
• Two special cases
• Achievement tasks achieve state of affairs P
• Maintenance tasks maintain state of affairs Q

18
Predicate Task Specification

• An achievement task is specified by a set G of
  good or goal states G S.
• The agent succeeds if it is guaranteed to bring
  about at least one of these states.
• A maintenance goal is specified by a set B of
  bad states B S. The agent succeeds in a
  particular environment if it manages to avoid all
  states in B if it never performs actions which
  result in any state in B occurring.

19
Deliberative Agent
Sensors

• Explicit symbolic model of the world in which
  decisions are made via logical reasoning, based
  on pattern matching and symbolic manipulation
• sense-plan-act problem-solving paradigm of
  classical AI planning systems

World Model
Planner
Effectors
Plan Executer
20
Disadvantages

• Performance problems
• transduction problem
• time consuming to translate all of the needed
  information into the symbolic representation,
  especially if the environment is changing
  rapidly.
• representation problem
• how the world-model is represented symbolically
  and how to get agents to reason with the
  information in time for the results to be useful.
• Does not scale to real-world scenarios
• Examples of deliberative architectures
• BDI

21
Reactive Agents
Stimulus-response behavior

• Reactive agents have
• at most a very simple internal representation of
  the world, but provide tight coupling of
  perception and action
• Behavior-based paradigm
• Intelligence is a product of interaction between
  an agent and its environment
• Examples
• Brooks subsumption architecture

Sensors
s1
s2
sn
an
a2
a1
Effectors
22
Hybrid Agent
Deliberative Component

• Combination of deliberative and reactive behavior
• An agent consists of several subsystems
• Subsystems that develop plans and make decisions
  using symbolic reasoning (deliberative component)
• Reactive subsystems that are able to react
  quickly to events without complex reasoning
  (reactive component)
• Layered architectures

Sensors
Observations
Modification
Effectors
Reactive Component
23
Agent Behaviour
World State

• Orientation
• sensing, modelling
• evaluating current situation in relation to
  external parameters
• Planning
• specify plans describing the approach towards
  reaching the goals
• Scheduling
• select appropriate plan and allocate
  corresponding resources
• identify and assign tasks
• Acting
• execute / perform tasks
• with effect on external parameters and world state

Orientation
define/modify
Knowledge
Goals
Planning
plan
Capabilities
Plans
Scheduling
decide
Tasks
Resources
Acting
effect
24
Components of an Agent
communicator
message manager
other agents
UI manager
user
sensor
environment
actuator
decision finder
knowledge base
inference component
local data storage
task-specific functions
wrapper
new implementations
25
(No Transcript)
26
Exam ??

• Web search to collect as many relevant web links
  as possible to create your Agents Portal
• Good place to start with http//agents.umbc.edu/
• General agents resources and portals
• Agent related conferences, seminars, people,
  research groups, courses, etc.,
• Examples of agents systems
• Agent implementation architectures
• APIs and frameworks for agent programming
• Agents security
• Agents applications

27
Exam ??

• Seminar Agents in Action
• Identify the class of agents you are interested
  in and search for examples of this kind of agents
  on the web.
• Agents in eCommerce
• User interface agents
• Network agents
• Information retrieval agents
• Filtering and profiling agents
• etc.,

28
Exam ??

• Your first agent
• What is the purpose of the agent you created
• Which means (information sources, input, etc.,)
  the agent uses to accomplish its goals
• How well does the agent succeed in its task
• What is the high level architecture of the agent
• etc.,