1. Agent Technology for e-Commerce
Chapter 2: Software Agents
Maria Fasli
Maria Fasli, University of Essex

2. What is an agent?
A piece software (and/or hardware) that acts on behalf of the user
Unfortunately there is no unique and universally accepted definition of what constitutes an agent
Different characteristics are important for different domains of application

3. Defining agents
“An agent is anything that can be viewed as perceiving its environment through sensors and acting upon that environment through effectors” (Russell and Norvig 2003)
“Agents are active, persistent (software) components that perceive, reason, act and communicate” (Huhns and Singh 1997)
“an entity that functions continuously in an environment in which other processes take place and other agents exist” (Shoham 1997)
“Autonomous agents are computational systems that inhabit some complex environment, sense and act autonomously in this environment, and by doing so realize a set of goals or tasks that they are designed for” (Maes 1995)

4. Characteristics of agents
Although there is no agreement regarding the definitive list of characteristics for agents, among the most important seem to be:
Autonomy
Proactiveness
Reactiveness
Social ability

5. Autonomy
Difficult to pin down exactly: how self-ruled the agent really is
An autonomous agent is one that can interact with its environment without the direct intervention of other agents and has control over its own actions and internal states
The less predictable an agent is the more autonomous it appears to be to an external observer
Absolute autonomy (complete unpredictability) may not be desirable; travel agent may exceed the allocated budget
Restrictions on autonomy via social norms

6. Proactiveness
Proactive (goal-directed) behaviour: an agent actively seeks to satisfy its goals and further its objectives
Simplest form is writing a procedure or method which involves:
Preconditions that need to be satisfied for the procedure/method to be executed
Postconditions which are the effects of the correct execution of the procedure
If the preconditions are met and the procedure executes correctly, then the postconditions will be true

7. Reactiveness
Goal-directed behaviour as epitomised via the execution of procedures makes two limiting assumptions:
while the procedure executes the preconditions remain valid
the goal and the conditions for pursuing such a goal, remain valid at least until the procedure terminates
Not realistic in dynamic, complex and uncertain environments
Agents must not blindly attempt to achieve their goals, but should perceive their environment and any changes that affect their goals and respond accordingly
Building an agent that achieves a balance between proactive and reactive behaviour is difficult

8. Social ability
Agents are rarely isolated entities, they usually live and act in an environment with other agents, human or software
Social ability means being able to operate in a multi-agent environment and coordinate, cooperate, negotiation and even compete with others
This social dimension of the notion of agency must address many difficult situations which are not yet fully understood, even in the context of human behaviour

9. Agents as intentional systems
Trying to understand and analyze the behaviour of complex agents in a natural, intuitive and efficient way is a nontrivial task
Methods that abstract us away from the mechanistic and design details of a system may be more convenient
Intentional stance: ascribing to a system human mental attitudes (anthropomorphism), e.g. beliefs, desires, knowledge, wishes

10. How useful/legitimate is this approach? McCarthy explains:
To ascribe certain ‘beliefs’, ‘knowledge’, ‘free will’, ‘intentions’, consciousness’, ‘abilities’ or ‘wants’ to a machine or computer program is legitimate when such an ascription expresses the same information about the machine that it expresses about a person. It is useful when the ascription helps us understand the structure of the machine, its past or future behaviour, or how to repair or improve it.

11. The intentional stance provides us with a powerful abstraction tool: the behaviour of systems whose structure is unknown can be explained
Computer systems or programs are treated as rational agents.
An agent possesses knowledge or beliefs about the world it inhabits, it has desires and intentions and it is capable of performing a set of actions
An agent uses practical reasoning and based on its information about the world and its chosen desires and intentions will select an action which will lead it to the achievement of one of its goals

12. Making decisions
We require software agents to whom complex tasks and goals can be delegated
Agents should be smart so that they can make decisions and take actions to successfully complete tasks and goals
Endowing the agent with the capability to make good decisions is a nontrivial issue
Environment
Sensory information
Action

13. A simple view of an agent
Environment states S={s1, s2, …}
Perception see:S→P
An agent has an internal state (IS) which is updated by percepts:
next:IS P →IS
An agent can choose an action from a set A={a1, a2, …}:
action:IS →A
The effects of an agent’s actions are captured via the function do:
do:A  S →S

14. The control loop of such an agent would look as follows:

15. Characteristics of the environment
The nature of the environment has a direct impact on the design of an agent and its decision-making process
It can be characterized as:
Fully or partially observable
Deterministic or stochastic
Static or dynamic
Episodic or sequential
Discrete or continuous
Single-agent or multi-agent

16. Fully vs partially observable environments
Observability describes access to information about the world
In a fully observable environment an agent has complete access to its state and can observe any changes as they occur in it
Most realistic environments are only partially observable
Partial observability can be attributed to noise in the agent’s sensors or perceptual aliasing
Partial observability is important in multi-agent systems as it also affects what the agent knows about the other agents
The more information an agent has about its world, the easier it is to choose and perform the best action

17. Deterministic vs stochastic environments
Deterministic environments
The next state is completely determined by the current state and the actions performed by the agent
The outcome of an agent’s actions is uniquely defined, no need to stop and reconsider
Most environments are stochastic
There is a random element that decides how the world changes
Limited sphere of influence: the effects of an agent’s actions are not known in advance
An agent’s actions may even fail
Stochasticity complicates agent design

18. Static vs dynamic environments
Static environments
The world only changes by the performance of actions by the agent itself
If an agent perceives the world at time t0 and the agent performs no action until t1, the world will not change
Dynamic environments
The world constantly changes
Even when an agent is executing an action a with a precondition p which holds true before the execution, p may not be true at some point during the execution
The outcome of an agent’s action cannot be guaranteed as other agents and the environment itself may interfere

19. It is more difficult to build agents for dynamic environments
Issue: the agent needs to do information gathering often enough in order to have up-to-date information about the world; this depends on the rate of change of the environment
An agent also needs to take into account the other agents and synchronize and coordinate its actions with theirs in order to avoid interference and conflicts

20. Episodic vs sequential environments
In an episodic environment
A cycle of perception and then action is considered to be a separate episode
The agent’s performance depends only on the current episode
The agent need not worry about the effects of its actions on subsequent episodes and need not think ahead
In sequential environments each decision made affects the next one
How an environment is characterized depends on the level of abstraction

21. Discrete vs continuous environments
In a discrete environment
There is a fixed, finite number of actions and percepts
In principle, one can enumerate all possible states and the best action to perform in each of these – not practical though
In continuous environments
The number of states may be infinitely long

22. Single-agent vs multi-agent environments
In a single-agent environment there is one agent operating whereas in multi-agent environments there are many agents that interact with each other
But, at times objects or entities that we would not normally consider as agents may have to be modelled as such
Nature may be modelled as an agent
Usually any entity/object that affects or influences the behaviour of the agent under consideration needs to be regarded as an agent

23. Open environments
The most complex class of environments are those that are partially observable, stochastic, dynamic, sequential, continuous and multi-agent
Known as open environments

24. Performance measure
Objective: develop agents that perform well in their environments
A performance measure indicates how successful an agent is
Two aspects: how and when

25. How is performance assessed:
Different performance measures will be suitable for different types of agents and environments
Contrast a trading agent with a vacuum cleaning agent
Objective performance measures are defined by us as external observers of a system
When is performance assessed:
Important
Continually, periodically or one-shot

26. Goal states
One possible way to measure how well an agent is doing is to check that it has achieved its goal
There may be a number of different action sequences that will enable an agent to satisfy its goal
A good performance measure should allow the comparison of different world states or sequences of states

27. Preferences and utilities
Agents need to be able to express preferences over different goal states
Each state s can be associated with a utility u(s) for each agent
The utility is a real number which indicates the desirability of the state for the agent
For two states s and s’
agent i prefers s to s’ if and only if u(s)>u(s’)
is indifferent between the two states if and only if u(s)=u(s’)
The agent’s objective is to bring about states of the environment that maximise its utility

28. Maximum Expected Utility
In a stochastic environment the performance of an action may bring about any one of a number of different outcomes
The expected utility of an action a is:
The agent then chooses to perform action a* which has the maximum expected utility (MEU):

29. A complete specification of the utility function allows rational decisions when:
there are conflicting goals, only some of which can be accomplished; the utility function indicates the appropriate tradeoff;
there are several goals that the agent can endeavour to achieve, but none of which can be achieved with certainty; the utility provides a way in which the likelihood of success can be evaluated against the importance of the goals

30. Rationality
What is rational at any given time depends on:
the performance measure that determines the degree of success
everything that the agent has perceived so far
what the agent expects to perceive and happen in the future
what the agent knows about the environment
the actions that the agent can perform
Definition: An ideal rational agent performs actions that are expected to maximize its performance measure

31. Bounded rationality
Making a decision requires computational power, memory and computation takes time
Agents are resource-bounded and this has an impact on their decision-making process: optimal decision making may not be possible
Ideal rationality may be difficult to achieve
Bounded rationality
restrictions on the types of options may be imposed
the time/computation for option consideration may be limited
the search space may be pruned
the option selected will be strategically inferior to the optimal one

32. Rational decision making and optimal policies
An agent is situated in an environment where there may be other agents present
Time can be measured in discrete time points T={1,2,…}
The environment is in a state st at time t and the set of world states is indicated by S
The agent perceives its environment through a percept pt
The agent affects its environment by performing an action at
A policy  is a complete mapping from states to actions
Given a state, a policy tells an agent what action to perform
An agent may take into account information about the past and the future

33. Taking into account the past
Information about the past: percept and action pairs (pt,at)
An agent’s policy:
((p1,a1),(p2,a2),…,pt)=at
This can be problematic
(i) history may be too large, and
(ii) computing an optimal policy  from a computational complexity point of view would be nontrivial

34. Markov environments
In some environments, the state of the world at time t provides a complete description of the history before t, hence pt=st
All necessary information to decide on an optimal action is in pt
Such a world state is said to be Markov or have the Markov property
An agent’s policy is:
(pt)=at or (st)=at
Such an agent that can ignore the past is called a reactive agent and its policy reactive or memoryless

35. Taking into account the future
In a discrete world the agent performs an action a at each time point t, and the world changes as a result of this in t+1
A transition model T(s,a,s’) describes how the world s changes as a result of an action a being performed
In a deterministic environment
the transition model maps (s,a) to a single resulting state s’
a number of approaches exist to plan ahead
In a stochastic environment:
the transition model is T(s,a,s)= P(s’|s,a)
Graph search is not applicable as there is uncertainty about the transitions between states

36. MDPs and POMDPs
The problem of calculating an optimal policy in a fully observable, stochastic environment with a transition model that satisfies the Markov property, is called a Markov decision problem (MDP)
In a partially observable environment, pt provides limited information and the agent cannot determine in which state the world really is
The problem of calculating an optimal policy in a partially observable environment is called a partially observable Markov decision problem (POMDP)
Methods for solving MDPs are not directly applicable to POMDPs

37. Optimal policies in MDPs
Given a MDP, one can calculate a policy from the transition model (probabilities) and the utility function
Assume a stochastic, single-agent world with transition model P(s’|s,a). The agent should choose an optimal action a*:
The optimal policy is
But to be able to calculate the policy we need to know the utilities of all states

38. Example The agent can move North, South, East and West
Bumping onto a wall leaves the position unchanged
Fully observable and stochastic: every action to the intended direction succeeds with p=0.8, but with probability p=0.2 the agent moves at right angles towards the intended direction
Only the utilities of the terminal states are known
Start +1 -1 3 2 1

39. The utility function needs to be based on a state sequence u([s0,…,sn]) instead of a single state
To use the MEU rule, the utility function needs to be separable
u([s0,…,sn]) =R(s0,)+ u([s1,…,sn])
where R(s0,) is called the reward function

40. The immediate reward for each of the non-terminal states -0.04
The utility u(s) of a state is defined as
The rewards of the terminal states are propagated out through all the other states
Known as the Bellman equation forms the basis of dynamic programming
Calculating the utilities in dynamic programming is an n-step decision problem

41. Two other methods of calculating optimal policies in MDPs are:
Value iteration: starting with a transition model and a reward function, calculate a utility for each state, and then use this to create a policy
Policy iteration: start with some policy, and then repeatedly calculate the utility function for that policy, and then use it to calculate a new policy, and so on. A policy usually converges long before the utility function does.

42. The utilities have been calculated using the value iteration method+1 -1 3 2 1
(a)
(b)
0.87
0.82
0.76
0.72
0.49
0.78
0.93

43. Planning
A key ability for intelligent behaviour as it increases an agent’s flexibility enabling it to construct sequences of actions to achieve its goals
Planning is deciding what to do; closely related to scheduling which is deciding when to do an action
The agent needs to figure out the list of steps (subtasks) that need to be carried out in order to achieve its goal
Different approaches to planning

44. Approaches to planning
Logic-based approaches: such as Situation Calculus
Case-based approaches: previously generated plans are stored in a library and can be reused to solve similar problems in the future
Operator-based approaches: such as STRIPS

45. Planning as search:
Situation space search: find a path from the initial state to the goal state
Progression: forward search
Regression: backward search from the goal state
Plan space search: search through the space of possible plans
Total order planning (totally-ordered sequence of actions)
Partial order planning or least-commitment planning (partially-ordered set of actions)

46. Learning
Learning: the process of acquiring knowledge, skills, or attitudes through experience, imitation, or teaching, which then causes changes in the behaviour
Important in open, dynamic and constantly evolving environments that electronic marketplaces are

47. Why is learning important
If we succeed in endowing agents with the ability to learn, this may also offer important insights into how humans learn
Impractical or even impossible to specify agent systems correctly and completely at the time of design and implementation
It may not be practical to build all intelligence into the system in advance
There may be hidden relationships and correlations among huge amounts of data which may not be easy to discern
Produced systems may not work as well as desired or expected in the environments in which they are used

48. Knowledge about certain tasks may simply be too large to be explicitly encoded by humans
Agents that operate in dynamic environments have to be able to cope with the constant changes
Any initial knowledge and skills encoded at the development stage with time it may become useless or out-of-date as the environment changes
The nature of the task may change
The environment may change in such a way so that the agent’s goals need to be changed as well

49. Learning methods
Learning takes place as a result of the interaction between an agent and the environment and from observing one’s own decision-making
Isolated (centralized) learning
Learning with others (decentralized or interactive learning)
Through knowledge acquisition (rote, memorization)
From instruction and through advice taking
From examples and practice
By analogy
Through problem-solving
By discovery – the most difficult type of learning to implement

50. To improve the learning process feedback may be provided:
Supervised learning: the feedback specifies the desired activity and the objective is to match this desired action as closely as possible
Reinforcement learning: the feedback only specifies the utility of the learner’s activity and the goal is to maximise this utility
Unsupervised learning: no explicit feedback is provided and the objective is to learn useful and desired skills and activities through a process of trial and error

51. Agent architectures
An agent architecture describes how an agent can be built
Logic-based
Reactive
Belief-Desire-Intention
Hybrid

52. Logic-based architecture
The ‘traditional’ symbolic artificial intelligence approach
The agent possesses a symbolic representation of its environment (logical formulas) and rules on how it should behave and what actions it can take
The behaviour of the system is generated by syntactic manipulation of the symbolic representations (logical deduction)
Agent execution as theorem proving: If there is a theory that explains how the agent behaves, how goals are generated and how the agent can take action to satisfy them, then this specification can be directly executed to produce behaviour

53. The environment is described by sentences in L, KB= (L)
At every moment in time t an agent’s internal state is KBtKB
Environment states S={s1, s2, …}
Perception see:S→P
The agent’s internal state is updated by percepts:
next:KB P →KB
An agent can choose an action from a set A={a1, a2, …}:
action:KB →A
The effects of an agent’s actions are captured via the function do:
do:A  S →S

54. The agent’s decision-making process is modelled through the rules of inference 
KB :  can be proven from the inference rules 
The agent programmer has to encode the inference rules  in a way that that enables the agent to decide what to do

55. The control loop for a logic-based agent takes the following form

56. If Do(a) cannot be proven, find an action that is consistent with KB

57. The Maze World
The agent’s objective is to discover the gold, pick it up and then get it to the exit (2,2)
(0,1)
(2,0)
(1,1)
(2,1)
(0,0)
(1,0)
(2,2)
(1,2)
(0,2)

58. Starting position (0,0) facing East
The state of the world is described by the following predicates
In(x,y) the agent is in square with coordinates (x,y)
Gold(x,y) there is gold in square (x,y)
Facing(d) the agent faces d{North, South, East, West}
Perception:
The agent can perceive the world by detecting whether or not there is gold in a square, gold or null respectively
It can also perceive its position on the grid and its direction
Possible actions A={pick-up, forward, turn}
When the agent turns, it turns 90 degrees clockwise

59. The rules of inference  determine the agent’s behaviour
Rule for picking up the gold when detected:
In(x,y)Gold(x,y)→Do(pick-up)
Rules to enable the agent to move around:
In(0,0)  Facing(East)  Gold(0,0) →Do(forward)
In(0,1)  Facing(East)  Gold (0,1) → Do(forward)
In(0,2)  Facing(East)  Gold(0,2) → Do(turn)
In(0,2)  Facing(South)  Gold(0,0) → Do(forward) …

60. Advantages of logic-based approach
If there is a theory  which describes the agent’s behaviour, all we have to do is execute this specification
Elegant, intuitive, clear semantics

61. Disadvantages of logic-based approach
Two issues in building agents with traditional AI approach
Transduction problem: how can the world be translated into a meaningful symbolic model at the right abstraction level and in time for that model to be useful (images, speech etc.)
Representation problem: how to represent information in a symbolic form suitable for the agents to reason with and in time for the results of the reasoning to be useful (knowledge representation, reasoning and planning)

62. Other issues
How to transform percepts into declarative statements that describe the environment precisely enough
Writing down all the rules that would allow agents to operate in complex environments is unrealistic
Assumes calculative rationality: the world does not change in a significant way while the agent is deliberating – not realistic
Computational complexity of theorem proving is a problem. Propositional logic is decidable, but first-order logic is only semi-decidable: even if there is a proof, the theorem prover may fail to terminate
Representing temporal information and changes is difficult

63. Reactive architecture
The use of an internal representation and decision-making based on it is rejected
‘Smart’ behaviour is linked directly to the environment that the agent inhabits and can be generated by responding to changes
The representation of the world is built into the agent’s sensory and effectory capabilities; perceptual input is mapped to actions

65. Brooks’ position
The first to reject the idea of a symbolic model was Brooks
‘Real’ intelligence is situated in the world and not in disembodied systems
Intelligence is an emergent property
Intelligent behaviour can be generated without an explicit internal representation and without explicit reasoning, but by the interaction of simple behaviours

66. Subsumption Architecture
Brooks proposed the Subsumption Architecture
Task Accomplishing Behaviours (TABs): a TAB can be a finite state machine or a rule of the form situation →action
Each behaviour achieves a task and can be considered as an individual action function which takes sensory input and maps it to an action to be performed
Many behaviours can ‘fire’ simultaneously
Behaviours are arranged in layers: the lower the layer, the higher the priority
Lower levels are usually associated with tasks/functions that are critical to the agent’s survival (obstacle avoidance etc.)

67. The Luc Steels scenario
A mission to a distant planet to collect samples of rocks and minerals
Spaceship
Vehicle

68. if detect crumb → pick up 1 and travel down the gradient
Each behaviour may encompass more than one situation-action rules, for example:
if detect crumb → pick up 1 and travel down the gradient
if carrying samples and not at the base → drop 2 crumbs and travel up the gradient
The behaviours are arranged in a subsumption hierarchy S E N I G A C T I N G
Random movement
Return movement
Exploration
Trail detection
Obstacle avoidance

69. Advantages of the reactive approach
Simple and elegant
An agent’s behaviour is computationally tractable
Very robust against failure
The power lies in numbers: complex tasks can be accomplished by a group of simple reactive agents
Complex behaviours emerge from the interaction of simple ones

70. Disadvantages of the reactive approach
With no model of the environment, the agents need sufficient information about their current state in order to determine an action
Short-sighted and with no planning capabilities
Learning is difficult to achieve
Emergence or emergent behaviour is not yet fully understood and it is even more difficult to engineer
Hence, difficult to build task-specific agents

71. Belief-Desire-Intention architecture
Based on Bratman’s (1987) philosophical theory of practical reasoning
Theoretical reasoning: the processes of deriving knowledge or reaching conclusions using one’s beliefs and knowledge
Practical reasoning: deciding what to do
Deliberation: deciding what to achieve
Means-end reasoning: deciding how to achieve it
Beliefs: an agent’s information about the world
Desires: an agent’s motivation or possible options
Intentions: an agent’s commitments

72. The role of intentions Intentions are key in practical reasoning:
They describe states of affairs that the agent has committed to bringing about and as a result they are action-inducing
They resist reconsideration, are volitional and reasoning-centered, and one intention leads to further being generated
Forming intentions is critical to an agent’s success

73. Intentions play three characteristic roles (Bratman 1987):
An agent needs to determine ways to achieve its intentions
An agent’s intentions lead it to adopt or refrain from adopting further intentions
An agent is interested in succeeding in its intentions and thus it keeps track of its attempts to do so

74. BDI components

75. A set of current beliefs (Beliefs)
A belief revision function
brf: (Beliefs) P→ (Beliefs)
A set of desires (Desires)
An option generation function:
options: (Beliefs)  (Intentions) → (Desires)
A set of current intentions (Intentions)
A filter function
filter: (Beliefs)  (Desires)  (Intentions) → (Intentions)
An action selection function
asf: (Intentions) → A

76. Control loop for a BDI agent would take the form:

77. Advantages of the BDI approach
Clear and intuitive
Clear functional decomposition of the agent’s subsystems
The formal properties can be studied (BDI logics)

78. Disadvantages of the BDI approach
Although the decomposition of the subsystems is clear, how to efficiently implement their functionality is not
Agents need to achieve a balance between commitment and reconsideration
This depends on the rate of the environment change
Bold and cautious agents
If the rate of the environment change is low, then bold agents do better
If the rate of the environment change is high, then cautious agents do better

79. Hybrid architectures
Combine reactive and deliberative components and form a hierarchy of interacting layers
Each layer reasons at a different level of abstraction
Two types of layering:
Horizontal layering
Vertical layering

80. Horizontal layering Each layer can act as an independent agent
For n different behaviours n layers are implemented
The layers compete with each other in order to take control of the agent; a mediator function can be introduced
Layer n
Layer 2
Layer 1
Action
output
Sensory
input

81. Problems
The layers’ competition for the agent’s control can cause incoherence
Consistency can be achieved by introducing a function which achieves mediation between the layers
Mediator function is exponentially complete: if there are n layers capable of suggesting m possible actions there are mn interactions
The mediator function or a central control system can introduce a bottleneck into the agent’s decision making

82. TouringMachines Modeling layer Sensory Action input output
Planning layer
Reactive layer
Sensory
input
Action
output
Control rules

83. Reactive layer
Acts as a reactive agent and responds to changes as they occur
Implemented through situation-action rules
There is no model of the environment in this layer
Planning layer
Achieves the agent’s pro-active behaviour via plans based on a library of plan skeletons or schemas

84. Modelling layer
Endows the agent with reflective and predictive capabilities
Entities are modelled as having a configuration, beliefs, desires and intentions
Generates goals to resolve conflicts which are then propagated to the planning layer
Control subsystem
Decides which of the layers has control over the agent
It is implemented via control rules which can either suppress sensor information between the control rules and the control layers or else censor action outputs from the control layers

85. Vertical layering One-pass Two-pass Layer n Layer 2 Layer 1
Sensory input
Action output
Sensory
input
Action
output
One-pass
Two-pass

86. Advantages
Low complexity. If there are n layers there are n-1 interfaces between them. If each layer is capable of suggesting m possible actions then there are at most m2(n-1) interactions
No central control, no bottleneck in the agent’s decision making
Problems
Less flexible
Not fault tolerant

87. InteRRaP Cooperation planning layer Local planning layer
Behaviour-based layer
World interface (WIF)
Sensory information
Action output
Social model
Mental model
World model KB CU

88. Each layer consists of two subprocesses
Situation recognition and goal activation process (SG)
Planning, scheduling and execution process (PS)
Two main types of interactions take place between the layers:
Activation requests (bottom up) which are issued when a lower layer passes control to a higher layer. The request is issued by the PS of layer i to the SG of layer i+1
Commitment postings (top down) are sent from layer i to i-1 in order to achieve its goals. These are communicated between the PSs of the two layers

89. Agents in perspective Two views: A new software engineering paradigm
What is the fuss all about? – nothing new

90. Agents vs objects In the object-oriented approach
objects have identity, state and behaviour and communicate via messages
A message is a procedure call which is executed in the context of the receiving object
In the agent-oriented approach
Agents have identity, state (knowledge, beliefs, desires, intentions) and behaviour (plans to achieve goals, actions, reactions to events and even roles)
So are objects agents?

91. Agents exhibit autonomy, they have control over their state, execution and ultimately behaviour
Agents exhibit goal-directed, reactive and social behaviour
They are persistent, self-aware and able to learn and adapt
Agents have their own thread of control in a multi-agent system
Control in multi-agent systems is also distributed
Objects have none of these characteristics

92. Agents vs expert systems
Expert systems which started appearing in the 1970s were heralded as the first intelligent pieces of software
Are they agents?
Although intelligent, they have no control over their actions
They only provide domain-specific information and usually do not adapt
No true social ability: limited interaction with humans or other (expert) systems
They are decoupled from their environment

93. Agents, Web Services and the Semantic Web
Semantic Web vision: enrich web pages with meaning so that information can be accessed by programs
To facilitate automated processing on the Web and enable applications to communicate and work with each other, we need programmatic interfaces, namely web services
A web service is a collection of functions that are packaged as a single entity and published on the network to be used by others
Web services can be combined to offer complex functionality
Users delegate their needs to agents who can then deploy web services

94. Users Agents Semantic Web Web Services Documents Emails Databases
Applications
Web sites
Ontologies

95. Both web services and agents encapsulate functionality, but:
Web services are static, unable to adapt
Services are passive until they are invoked – once invoked the service will execute given the correct parameters
Communication between services is at a low syntactic level; services can exchange data but they do not understand them; they do not ‘understand’ ontologies to resolve incompatibilities
Composed functionality of web services can only be achieved when orchestrated by a third party, e.g. an agent; no social ability to coordinate with other web services or entities

96. Methodologies and languages
The reluctance to accept agents and multi-agent systems as a distinct software engineering paradigm is mainly because:
The field is relatively new
Lack of standard definitions and well-established and accepted methodologies for designing and developing such systems

97. Methodologies
Establishing a systematic methodology for the analysis and design of agent-based applications is imperative
Current methodologies can be divided into
Agent-oriented methodologies (Gaia, Prometheus, Tropos, ROADMAP, MaSE)
Methodologies that directly extend or adapt object-oriented methodologies (AAAI, AUML)
Methodologies that adapt knowledge engineering models (MAS-CommonKADS, DESIRE) or other techniques (Z language)

98. Programming languages and environments
Currently, no languages support agent-oriented programming at a high level of abstraction
One of the first attempts to create a pure agent-oriented language based on a mentalistic view of agents was Shoham’s (1993) agent-oriented programming (AOP) proposal and AGENT0
Logic-based approaches to agent programming: METATEM, AgentSpeak(L), recently Jason
Toolkits and environments to support the agent developer mostly based on Java: ZEUS, JADE, JACK Intelligent Agents, JATLite
Other frameworks: OAA, FIPA-OS, Grasshopper, ABLE, AgentBuilder

99. From legacy systems to agents
Wrapper
Transducer
(a)
(b)
(c)
Multi-agent world
Agent
Agentification methods
(a) Transducer; (b) Wrapper; (c) Rewrite the system