Expert Systems and KBS

1
Expert Systems and KBS
Intelligent Agents
Lecture 6 Abdennour El Rhalibi
2
Introduction

• What is an agent?
• How do we build agents?
• What is an Agent Environment?
• What is a Multi-Agent System?
• What I expect you to know!

3
What is an agent?

• An agent provides a service.
• An agent is situated in an environment.
• An agent perceives its environment
• sensors
• An agent acts upon its environment
• effectors
• An environment inhabited by agents may be
  physical or virtual.

4
The Basic Idea
Percepts Actions Goal Environment
5
PAGE Descriptions
Must first specify the setting for agent design
6
PAGE Descriptions
7
The Properties of an Environment
Must then specify the properties of the
environment

• Accessible vs. Inaccessible
• The sensors detect all the aspects of the
  environment
• Deterministic vs. NonDeterministic
• The next state can completely be determined by
  the current state and the agent action
• Episodic vs. NonEpisodic
• The experience of the agent is divided in
  episodes
• Static vs. Dynamic
• The environment doesnt change when the agent is
  delibirating
• Discrete vs. Continous
• There is a finite number of percepts and actions

8
Environment types
The environment type largely determines the agent
design The real world is (of course)
inaccessible, stochastic, sequential, dynamic,
continuous
9
Rational Agency
Must finally specify the rationality of agent

• A rational agent is an agent that chooses its
  actions in such a way that the outcomes for the
  agent are the most successful that the agent
  could achieve.
• We use the term performance measure to determine
  the success of an agent.

10
Rationality at any Given Time Depends on

• The performance measure.
• The percept sequence.
• What the agent knows about the environment.
• The actions that the agent can perform.

11
Performance Measure Heuristic

• The initial state the agent knows itself to be in
• Operators that, applied to states, yield the
  state that will result in the relevant action
  being taken
• The goal test
• A path cost function assigning a cost to a path

12
Partial Search Tree for Tic-Tac-Toe
13
Example Minimax Algorithm

• Generate the whole minimax game tree.
• Apply the utility function to each terminal state
  to get its value.
• Use the utility of the terminal states to
  determine the utility of the nodes one level
  higher up the search tree.
• Continue backing up the values from the leaf
  nodes toward the root, one layer at a time.
• When the backed-up values reach the top, Max
  chooses the move that leads to the highest value.

14
Evaluation Function

• Replaces utility function applied to entire tree
  - We cant always develop the whole tree
• An example in chess is material value
  (Heuristics)
• Of course, masters and grandmasters have some
  rather more sophisticated evaluation functions at
  work
• E.g. Deep Blue eval functions
• Problem solving depends on Evaluation Function

15
Autonomy in an Agent

• A system is autonomous to the extent that its
  behaviour is determined by its own experience.
• E.g. the mail services (both physical and
  virtual) display autonomous behaviour.
• A thermostat displays autonomous behaviour

16
Agent types

• Four basic types in order of increasing
  generality
• Simple reflex agents
• Reflex agents with state
• Goal-based agents
• Utility-based agents

17
A SKELETON AGENT
An agent is completely specified by the agent
function, mapping percept sequences to
actions (in principle one can supply each
possible sequence to see what it does. Obviously,
a lookup table would usually be immense)
18
Overview of Reflex Agent with Internal State
19
A Reflex Agent With Internal State
20
Utility-Based Agent
21
A Simple Problem-Solving Agent
function simple-problem-solving-agent(p) returns
an action inputs p, a percept static s, an
action sequence, initially empty state, some
description of the current world state g, a
goal, initially null problem, a problem
formulation state ? Update-State(state, p) if s
is empty then g ? Formulate-Goal(state) problem
? Formulate-Problem(state, g) s ?
Search(problem) action ? Recommendation(s,
state) s ? Remainder(s, state) return action
22
Multi-Agent Systems

• Many systems require more than one agent to
  operate within the environment in parallel. Such
  systems raise issues of
• Communication
• Cooperation
• Coordination
• In addition to the agent architecture we must
  also consider the system architecture which
  depends on the Application

23
Multi-Agent SystemsApplications

• Workflow management
• Network management
• Air traffic control
• Tranport planning
• smart databases
• Factory control
• Online search engine
• Road traffic

• Digitales library
• personal digital assistants (PDA)
• e-mail filtring
• Information management
• data mining
• Electronic commerce

24
What I Expect You to Know

• I expect you to be able to
• describe rationality, autonomy, situatedness
  with respect to agents in an environment
• describe performance measures and percept
  sequences.
• define a rational agent
• describe and provide examples of reflex agents
• describe a multi-agent system in terms of
  relations.

25
Blackboard Architecture I

• Imagine a group of human experts seated next to a
  classroom blackboard trying to solve some problem
• each expert is a specialist in some area relevant
  to the problems solution
• no expert has a global view of the overall
  problem
• The problem and initial data are written on the
  blackboard

26
Blackboard Architecture II

• The experts watch the blackboard to see if they
  can make a contribution to the solving of the
  problem
• When an expert can contribute, the expert will
  record the result on the blackboard
• This new information may make it possible for
  another expert to contribute
• The process halts when the problem has been solved

27
Blackboard Architecture Example

• We have three agents
• reader - can read numbers and convert them into
  blocks and vice-versa
• grouper - can add up piles of blocks by pushing
  the blocks together
• swapper - knows how to swap a ten block for ten
  one blocks and back again

28
Blackboard Architecture Example

• The initial problem is posed in terms of numbers
• add 13 to 8
• and the answer is also required as a written
  number.

29
Blackboard Architecture Example

• The three agents solve the problem in the
  following steps
• The reader converts the numbers thirteen and
  eight into the equivalent blocks
• The pusher clumps the blocks together to give an
  answer in blocks
• the swapper changes ten one blocks into one ten
  block
• the reader converts the pile of blocks into the
  number 21

30
Blackboard Architecture Example

• Comments
• None of the agents have any idea of an overall
  solution or an explicit strategy
• cooperation is implicit and benign, agents
  always try to fulfill their responsibilities to
  the group
• there is no coordination in this pure example of
  a simple blackboard