Wooldridge and Jennings - 1995 - Intelligent agents theory and practice.pdf

The Knowledge Engineering Review, Vol. 10:2, 1995, 115-152
Intelligent agents: theory and practice
MICHAEL WOOLDRIDGE1 and NICHOLAS R. JENNINGS2
1 Department of Computing Manchester Metropolitan University Chester Street, Manchester MI 5GD, UK
(M. Wooldridge@doc.mmu.ac.uk)
Department
of Electronic F.ngineering, Queen Mary & Westfield College, Mile End Road, London El 4N.S, UK
( N. R.Jennings@gm w.ac.uk)
Abstract
The concept of an agent has become important in both artificial intelligence (AI) and mainstream
computer science. Our aim in this paper is to point the reader at what we perceive to be the most
important theoretical and practical issues associated with the design and construction of intelligent
agents. For convenience, we divide these issues into three areas (though as the reader will see, the
divisions arc at times somewhat arbitrary). Agent theory is concerned with the question of what an
agent is, and the use of mathematical formalisms for representing and reasoning about the
properties of agents. Agent architectures can he thought of as software engineering models of
agents; researchers in this area are primarily concerned with the problem of designing software or
hardware systems that will satisfy the properties specified by agent theorists. Finally, agent
languages are software systems for programming and experimenting with agents; these languages
may embody principles proposed by theorists. The paper is not intended to serve as a tutorial
introduction to all the issues mentioned; we hope instead simply to identify the most important
issues, and point to work that elaborates on them. The article includes a short review of current and
potential applications of agent technology.
1 Introduction
We begin our article with descriptions of three events that occur sometime in the future:
I. The key air-traffic control systems in the country of Ruritania suddenly fail, due to freak
weather conditions. Fortunately, computerised air-traffic control systems in neighbouring
countries negotiate between themselves to track and deal with all affected flights, and the
potentially disastrous situation passes without major incident.
2. Upon logging in to your computer, you are presented with a list of email messages, sorted into
order of importance by your personal digital assistant (PDA). You are then presented with a
similar list of news articles; the assistant draws your attention to one particular article, which
describes hitherto unknown work that is very close to your own. After an electronic discussion
with a number of other PD As, your PDA has already obtained a relevant technical report for
you from an FTPsite, in the anticipation that it will be of interest.
3. You are editing a file, when your PDA requests your attention: an email message has arrived,
that containsnotification about a paper you sent to an important conference, and the PDA
correctly predicted that you would want to sec it as soon as possible. The paper has been
accepted, and without prompting, the PDA begins to look into travel arrangements, by
consulting a number of databases and other networked information sources. A short time later,
you are presented with a summary of the cheapest and most convenient travel options.
We shall not claim that computer systems of the sophistication indicated in these scenarios are just
around the corner, but serious academic research is underway into similar applications: air-traffic

M. WOOLDRIDGE AND NICHOLAS JENNINGS 116
control has long been a research domain in distributed artificial intelligence (DAI) (Steeb et al.,
1988); various types of information manager, that filter and obtain information on behalf of their
users, have been prototyped (Maes, 1994a); and systems such as those that appear in the third
scenario are discussed in (Mc Gregor, 1992; Levy ct al., 1994). The key computer-based com
ponents that appear in each of the above scenarios are known as agents. It is interesting to note that
one way of defining Al is by saying that it is the subfield of computer science which aims to construct
agents that exhibit aspectsof intelligent behaviour. The notion of an "agent" is thus central to AI. It
is perhaps surprising, therefore, that until the mid to late 1980s, researchers from mainstream AI
gave relatively little consideration to the issues surrounding agent synthesis. Since then, however,
there has been an intense flowering of interest in the subject: agents are now widely discussed by
researchers in mainstream computer science, as well as those working in data communications and
concurrent systems research, robotics, and user interface design. A British national daily paper
recently predicted that:
"Agent-hased computing (ABC) is likely to be the next significant breakthrough in software development.··
(Sargent, 1992)
Moreover, the UK-based consultancy firm Ovum has predicted that the agent technology industry
would be worth some US$3.5 billion worldwide by the year 2000 (Houlder, 1994). Researchers
from both industry and academia arc thus taking agent technology seriously: our aim in this paper is
to survey what we perceive to be the most important issues in the design and construction of
intelligent agents, of the type that might ultimate appear in applications such as those suggested by
the fictional scenarios ahove. We begin our article, in the following sub-section, with a discussion
on the subject of exactly what an agent is.
I. I What is an agent?
Carl Hewitt recently remarked1 that the question what is an agent? is embarrassing for the agent
based computing community in just the same way that the question what is intelligence? is
embarrassing for the mainstream AI community. The problem is that although the term is widely
used, by many people working in closely related areas, it defies attempts to produce a single
universally accepted definition. This need not necessarily be a problem: after all, if many people
are successfully developing interesting and useful applications, then it hardly matters that they do
not agree on potentially trivial terminological details. However, there is also the danger that unless
the issue is discussed, "agent" might become a "noise'" term, subject to both abuse and misuse, to
the potential confusion of the research community. It is for this reason that we briefly consider the
question.
We distinguish two general usages of the term "agent": the first is weak, and relatively
uncontentious; the second is stronger, and potentially more contentious.
I. I. I A Weak Notion of Agency
Perhaps the most general way in which the term agent is used is to denote a hardware or (more
usually) software-based computer system that enjoys the following properties:
• autonomy: agents operate without the direct intervention of humans or others, and have some
kind of control over their actions and internal state (Castelfranchi, 1995);
• social ability: agents interact with other agents (and possibly humans) via some kind of agent
communication language (Genesereth & Ketchpel, 1994);
• reactivity: agents perceive their environment (which may be the physical world, a user via a
graphical user interface, a collection of other agents, the Internet, or perhaps all of these
combined), and respond in a timely fashion to changes that occur in it;
• pro-activeness: agents do not simply act in response to their environment, they are able to exhibit
goal-directed behaviour by taking the initiative.
1A t the Thirteenth International Workshop on Distributed Al.

Intelligent agents: theory and practice 117
A simple way of conceptualising an agent is thus as a kind of UNIX-like software process, that
exhibits the properties listed above. This weak notion of agency has found currency with a
surprisingly wide range of researchers. For example, in mainstream computer science, the notion
of an agent as a self-contained, concurrently executing software process, that encapsulates some
state and is able to communicate with other agents via message passing, is seen as a natural
development of the object-based concurrent programming paradigm (Agha, 1986; Agha ct al.,
1993).
This weak notion of agency is also that used in the emerging discipline of agent-based software
engineering:
"[Agents} communicate with their peers by exchanging messages in an expressive agent communication
Language. While agents can be as simple as subroutines, typically they are larger entities with some sort of
persistent control." (Gcncscrcth & Kctchpcl, 1994, p.48)
A softbot (software robot) is a kind of agent:
"A softbot is an agent that interacts with a software environment by issuing commands and interpreting the
environments feedback. A softbot's effectors arc commands (e.g. Unix shell commands such as mv or
compress) meant to change the external environments state. A softbot's sensors are commands (e.g. pwd
or ls in Unix) meant to provide . . information." (Etzioni et al., 1994, p.10)
1.1.2 A stronger notion of agency
For some researchers-particularly those working in AI- the term "agent" has a stronger and
more specific meaning than that sketched out above. These researchers generally mean an agent to
be a computer system that, in addition to having the properties identified above, is either
conceptualised or implemented using concepts that arc more usually applied to humans. For
example, it is quite common in AI to characterise an agent using mentalistic notions, such as
knowledge, belief, intention, and obligation (Shoham, 1993). Some AI researchers have gone
further, and considered emotional agents (Bates et al., 1992a; Bates, 1994). (Lest the reader
suppose that this is just pointless anthropomorphism, it should be noted that there are good
arguments in favour of designing and building agents in terms of human-like mental states-sec
section 2.) Another way of giving agents human-like attributes is to represent them visually,
perhaps by using a cartoon-like graphical icon or an animated face (Maes, 1994a, p. 36)-for
obvious reasons, such agents are of particular importance to those interested in human-computer
interfaces.
1.1.3 Other attributes of agency
Various other attributes arc sometimes discussed in the context of agency. For example:
• mobility is the ability of an agent to move around an electronic network (White, 1994);
• veracity is the assumption that an agent will not knowingly communicate false information
(Galliers, 1988b, pp. 159-164);
• benevolence is the assumption that agents do not have conflicting goals, and that every agent will
therefore always try to do what is asked of it (Rosenschein and Genesereth, 1985, p. 91); and
• rationality is (crudely) the assumption that an agent will act in order to achieve its goals, and will
not act in such a way as to prevent its goals being achieved-at least insofar as its beliefs permit
(Galliers, 1988b, pp. 49-54).
(A discussion of some of these notions is given below; various other attributes of agency are
formally defined in (Goodwin, 1993).)
1.2 The structure of this article
Now that we have at least a preliminary understanding of what an agent is, we can embark on a
more detailed look at their properties, and how we might go about constructing them. For

M. WOOLDRIDGE AND NICHOLAS JENNINGS 118
convenience, we identify three key issues, and structure our survey around these (cf. Seel, 1989,
p.1 ):
• Agent theories are essentially specifications. Agent theorists address such questions as: How are
we to conceptualise agents? What properties should agents have, and how are we to formally
represent and reason about these properties?
• Agent architectures represent the move from specification to implementation. Those working in
the area of agent architectures address such questions as: How are we to construct computer
systems that satisfy the properties specified by agent theorists? What software and/or hardware
structures are appropriate? What is an appropriated separation of concerns?
• Agent languages are programming languages that may embody the various principles proposed
by theorists. Those working in the area of agent languages address such questions as: How are
we to program agents? What are the right primitives for this task? How are we to effectively
compile or execute agent programs?
As we pointed out above, the distinctions between these three areas are occasionally unclear. The
issue of agent theories is discussed in the section 2. In section 3, we discuss architectures, and in
section 4, we discuss agent languages. A brief discussion of applications appears in section 5, and
some concluding remarks appear in section 6. Each of the three major sections closes with a
discussion, in which we give a brief critical review of current work and open problems, and a section
pointing the reader to further relevant reading.
Finally, some notes on the scope and aims of the article. First, it is important to realise that we
are writing very much from the point of view of AI, and the material we have chosen to review
clearly reflects this bias. Secondly, the article is not a intended as a review of Distributed AI,
although the material we discuss arguably falls under this banner. We have deliberately avoided
discussing what might be called the macro aspects of agent technology (i.e., those issues relating to
the agent society, rather than the individual (Gasser, 1991), as these issues are reviewed more
thoroughly elsewhere (see Bond and Gasser, 1988, pp. 1-56, and Chaibdraa et al., 1992). Thirdly,
we wish to reiterate that agent technology is, at the time of writing, one of the most active areas of
research in AI and computer science generally. Thus, work on agent theories, architectures, and
languages is very much ongoing. In particular, many ofthe fundamental problems associated with
agent technology can by no means be regarded as solved. This article therefore represents only a
snapshot of past and current work in the field, along with some tentative comments on open
problems and suggestions for future work areas. Our hope is that the article will introduce the
reader to some of the different ways that agency is treated in D(AI), and in particular to current
thinking on the theory and practice of such agents.
2 Agent theories
In the preceding section, we gave an informal overview of the notion of agency. In this section, we
turn our attention to the theory of such agents, and in particular, to formal theories. We regard an
agent theory as a specification for an agent; agent theorists develop formalisms for representing the
properties of agents, and using these formalisms, try to develop theories that capture desirable
properties of agents. Our starting point is the notion of an agent as an entity 'which appears to be
the subject of beliefs, desires, etc.' (Seel, 1989, p. 1). The philosopher Dennett has coined the term
intentional system to denote such systems.
2. 1 Agents as intentional systems
When explaining human activity, it is often useful to make statements such as the following:
Janine took her umbrella because she believed it was going to rain.
Michael worked hard because he wanted to possess a Ph D.

Intelligent agents: theory and practice 119
These statements make use of a folk psychology, by which human behaviour is predicted and
explained through the attribution of attitudes, such as believing and wanting (as in the above
examples), hoping, fearing and so on. This folk psychology is well established: most people reading
the above statements would say they found their meaning entirely clear, and would not give them a
second glance.
The attitudes employed in such folk psychological descriptions are called the intentional notions.
The philosopher Daniel Dennett has coined the term intentional system to describe entities "whose
behaviour can be predicted by the method of attributing belief, desires and rational acumen"
(Dennett, 1987, p. 49). Dennett identifies different "grades" of intentional system:
"A first-order intentional system has beliefs and desires (etc.) but no beliefs and desires (and no doubt other
intentional states) about beliefs and desires .... A second-order intentional system is more sophisticated; it
has beliefs and desires (and no doubt other intentional states) about beliefs and desires (and other
intentional state)-both those of others and its own" (Dennett, 1987, p. 243)
One can carry on this hierarchy of intentionality as far as required.
An obvious question is whether it is legitimate or useful to attribute beliefs, desires, and so on, to
artificial agents. Isn't this just anthropomorphism? Mc Carthy, among others, has argued that there
are occasions when the intentional stance is appropriate:
"To ascribe beliefs, free will, intentions, consciousness, abilities, or wants to a machine 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. It is perhaps never logically required even for humans, but expressing
reasonably briefly what is actually known about the state of the machine in a particular situation may
require mental qualities or qualities isomorphic to them. Theories of belief, knowledge and wanting can be
constructed for machines in a simpler setting than for humans, and later applied to humans. Ascription of
mental qualities is most straightforward for machines of known structure such as thermostats and computer
operating systems, but is most useful when applied to entities whose structure is incompletely known."
(Mc Carthy, 1978) (quoted in (Shoham, 1990))
What objects can be described by the intentional stance? As it turns out, more or less anything can.
In his doctoral thesis, Seel showed that even very simple, automata-like objects can be consistently
ascribed intentional descriptions (Seel 1989); similar work by Rosenschein and Kaelbling (albeit
with a different motivation), arrived at a similar conclusion (Rosenschein & Kaelbling, 1986). For
example, consider a light switch:
"It is perfectly coherent to treat a light switch as a (very cooperative) agent with the capability of
transmitting current at will, who invariably transmits current when it believes that we want it transmitted
and not otherwise; flicking the switch is simply our way of communicating our desires.'' (Shoham, 1990, p.
6)
And yet most adults would find such a description absurd-perhaps even infantile. Why is this?
The answer seems to be that while the intentional stance description is perfectly consistent with the
observed behaviour of a light switch, and is internally consistent,
.. it does not buy us anything, since we essentially understand the mechanism sufficiently to have a
simpler, mechanistic description of its behaviour." (Shoham, 1990, p. 6)
Put crudely, the more we know about a system, the less we need to rely on animistic, intentional
explanations of its behaviour. However, with very complex systems, even if a complete, accurate
picture of the system's architecture and working is available, a mechanistic, design stance
explanation of its behaviour may not be practicable. Consider a computer. Although we might
have a complete technical description of a computer available, it is hardly practicable to appeal to
such a description when explaining why a menu appears when we click a mouse on an icon. In such
situations, it may be more appropriate to adopt an intentional stance description, if that description
is consistent, and simpler than the alternatives. The intentional notions are thus abstraction tools,
which provide us with a convenient and familiar way of describing, explaining, and predicting the
behaviour of complex systems.

M. WOOLDRIDGE AND NICHOLAS JENNINGS 120
Being an intentional system seems to be a necessary condition for agenthood. but is it a sufficient
condition? In his Master's thesis, Shardlow trawled through the literature of cognitive science and
its component disciplines in an attempt to find a unifying concept that underlies the notion of
agenthood. He was forced to the following conclusion:
"Perhaps there is something more to an agent than its capacity for beliefs and desires, but whatever that
thing is, it admits no unified account within cognitive science." (Shardlow, 1990)
So, an agent is a system that is most conveniently described by the intentional stance; one whose
simplest consistent description requires the intentional stance. Before proceeding, it is worth
considering exactly which attitudes are appropriate for representing agents. For the purposes of
this survey, the two most important categories are information attitudes and pro-attitudes:
desire intention obligation commitment
intention
belief
obligation
information attitudes pro-attitudes l
{ commitment
knowledge
choice
Thus information attitudes are related to the information that an agent has about the world it
occupies, whereas pro-attitudes are those that in some way guide the agent's actions. Precisely
which combination of attitudes is most appropriate to characterise an agent is, as we shall sec later,
an issue of some debate. However, it seems reasonable to suggest that an agent must be
represented in terms of at least one information attitude, and at least one pro-attitude. Note that
pro- and information attitudes are closely !inked, as a rational agent will make choices and form
intentions, etc., on the basis of the information it has about the world. Much work in agent theory is
concerned with sorting out exactly what the relationship between the different attitudes is.
The next step is to investigate methods for representing and reasoning about intentional
notions.
2.2 Representing intentional notions
Suppose one wishes to reason about intentional notions in a logical framework. Consider the
following statement (after Genesereth & Nilsson, 1987, pp. 210-211):
Janine believes Cronos is the father of Zeus. (1)
A naive attempt to translate (1) into first-order logic might result in the following:
Bel(Janine, Father(Zeus, Cronos)) (2)
Unfortunately, this naive translation does not work, for two reasons. The first is syntactic: the
second argument to the Bel predicate is a formula of first-order logic, and is not, therefore, a term.
So (2) is not a well-formed formula of classical first-order logic. The second problem is semantic,
and is potentially more serious. The constants Zeus and Jupiter, by any reasonable interpretation,
denote the same individual: the supreme deity of the classical world. It is therefore acceptable to
write, in first-order logic:
(Zeus= Jupiter). (3)
Given (2) and (3), the standard rules of first-order logic would allow the derivation of the following:
Bel(Janine, Father(Jupiter, Cronos)) (4)
But intuition rejects this derivation as invalid: believing that the father of Zeus is Cronos is not the
same as believing that the father of Jupiter is Cronos. So what is the problem? Why does first-order

Intelligent agents: theory and practice 121
logic fail here? The problem is that the intentional notions-such as belief and desire-are
referentially opaque, in that they set up opaque contexts, in which the standard substitution rules of
first-order logic do not apply. In classical (propositional or first-order) logic, the denotation, or
semantic value, of an expression is dependent solely on the denotations of its sub-expressions. For
example, the denotation of the propositional logic formulap /\ q is a function of the truth-values of
p and q. The operators of classical logic are thus said to be truth functional. In contrast, intentional
notions such as belief are not truth functional. It is surely not the case that the truth value of the
sentence:
Janine believes p (5)
is dependent solely on the truth value of p 2 So substituting equivalents into opaque contexts is not
going to preserve meaning. This is what is meant by referential opacity. Clearly, classical logics are
not suitable in their standard form for reasoning about intentional notions: alternative formalisms
are required.
The number of basic techniques used for alternative formalisms is quite small. Recall, from the
discussion above, that there arc two problems to be addressed in developing a logical formalism for
intentional notions: a syntatic one, and a semantic one. It follows that any formalism can be
characterised in terms of two independent attributes: its language of formulation, and semantic
model (Konolige, 1986a, p. 83).
There are two fundamental approaches to the syntactic problem. The first is to use a modal
language, which contains non-truth-functional modal operators, which arc applied to formulae. An
alternative approach involves the use of a meta-language: a many-sorted first-order language
containing terms that denote formulae of some other object-language. Intentional notions can be
represented using a meta-language predicate, and given whatever axiomatisation is deemed
appropriate. Both of these approaches have their advantages and disadavantagcs, and will be
discussed in the sequel.
As with the syntactic problem, there arc two basic approaches to the semantic problem. The
first, best-known, and probably most widely used approach is to adopt a possible worlds semantics,
where an agent's beliefs, knowledge, goals, and so on, arc characterised as a set of so-called
possible worlds, with and accessibility relation holding between them. Possible worlds semantics
have an associated correspondence theory which makes them an attractive mathematical tool to
work with (Chellas, 1980). However, they also have many associated difficulties, notably the well
known logical omniscience problem, which implies that agents are perfect reasoners (we discuss
this problem in more detail below). A number of variations on the possible-worlds theme have
been proposed, in an attempt to retain the correspondence theory, but without logical omnis
cience. The commonest alternative to the possible worlds model for belief is to use a sentential, or
interpreted symbolic structures approach. In this scheme, beliefs are viewed as symbolic formulae
explicitly represented in a data structure associated with an agent. An agent then believes
sig
i
m
f asigmias
present in its belief data structure. Despite its simplicity, the sentential model works well under
certain circumstances (Konolige, 1986a).
In the subsections that follow, we discuss various approaches in some more detail. We begin
with a close look at the basic possible world:-, model for logics of knowledge (episremic logics) and
logics of belief (doxastic logics).
2.3 Possible worlds semantics
The possible worlds model for logics of knowledge and belief was originally proposed by Hintikka
(1962), and is now most commonly formulated in a normal modal logic using the techniques
2Note, however, that the sentence (5) is itself a proposition, in that its denotation is the value true or false.

M. WOOLDRIDGE AND NICHOLAS JENNINGS 122
developed by Kripke (1963).3 Hintikka's insight was to see that an agent's beliefs could be
characterised as a set of possible worlds, in the following way. Consider an agent playing a card
game such as poker.4 In this game, the more one knows about the cards possessed by one's
opponents, the better one is able to play. And yet complete knowledge of an opponent's cards is
generally impossible (if one excludes cheating). The ability to play poker well thus depends, at least
in part, on the ability to deduce what cards are held by an opponent, given the limited information
available. Now suppose our agent possessed the ace of spades. Assuming the agent's sensory
equipment was functioning normally, it would be rational of her to believe that she possessed this
card. Now suppose she were to try to deduce what cards were held by her opponents. This could be
done by first calculating all the various different ways that the cards in the pack could possibly have
been distributed among the various players. (This is not being proposed as an actual card playing
strategy, but for illustration!) For argument's sake, suppose that each possible configuration is
described on a separate piece of paper. Once the process is complete, our agent can then begin to
systematically eliminate from this large pile of paper all those configurations which are not possible,
given what she knows. For example, any configuration in which she did not possess the ace of spades
could be rejected immediately as impossible. Call each piece of paper remaining after this process a
world. Each world represents one state of affairs considered possible, given what she knows.
Hintikka coined the term epistemic alternatives to describe the worlds possible given one's beliefs.
Something true in all our agent's epistemic alternatives could be said to be believed by the agent.
For example, it will be true in all our agent's epistemic alternatives that she has the ace of spades.
On a first reading, this seems a peculiarly roundabout way of characterising belief. but it has two
advantages. First, it remains neutral on the subject of the cognitive structure of agents. It certainly
doesn't posit any internalised collection of possible worlds. It is just a convenient way of
characterising belief. Second, the mathematical theory associated with the formalisation of
possible worlds is extremely appealing (see below).
The next step is to show how possible worlds may be incorporated into the semantic framework
of a logic. Epistemic logics arc usually formulated as normal modal logics using the semantics
developed by Kripke (1963). Before moving on to explicitly epistemic logics, we consider a simple
normal modal logic. This logic is essentially classical propositional logic, extended by the addition
of two operators:'·□" (necessarily), and"." (possibly). Let Prop= {p, q, .. . } be a countable set
of atomic propositions. Then the syntax of the logic is defined by the following rules: (i) if p e Prop
then p isa formula; (ii) if sigmpahi, are formulae, then so are -sigmaand i:p V 1.j J; and (iii) if rp i~ a formula
then so arc □qi and ◊ff. The operators "-," (not) and "V" (o r) have their standard meanings. The
remaining connectives of classical propositional logic can be defined as abbreviations in the usual
way. The formula □q, is read: "necessarily cp" and the formula ◊ff is read: "possibly cp". The
semantics of the modal connectives arc given by introducing an accessibility relation into models for
the language. This relation defines what worlds are considered accessible from every other world.
The formula □qi is then true if q; is true in every world accessible from the current world; ◊rp is true
if rp is true in at least one world accessible from the current world. The two modal operators are
duals of each other, in the sense that the universal and existential quantifiers of first-order logic arc
duals:
It would thus have been possible to take either one as primitive, and introduce the other as a
derived operator. The two basic properties of this logic arc as follows. First, the following axiom
= =
schema is valid: □(q-, 1.j J) (Oq, -=O1.JJ). This axiom is called K, in honour of Kripkc. The second
property is as follows: if <pis valid, then □q; is valid. Now, since K is valid, it will be a theorem of any
3In Hintikka's original work. he used a technique based on "model sets which is equivalent to Kripke's
formalism, though less elegant. See Hughes and Cresswell (1968, pp. 351-352) for a comparison and
discussion of the two techniques.
4This example was adapted from Halpern (1987).

Intelligent agents: theory and practice 123
complete axiomatisation of normal modal logic. Similarly, the second property will appear as a rule
of inference in any axiomisation of normal modal logic; it is generally called the necessitation rule.
These two properties turn out to be the most problematic features of normal modal logics when
they are used as logics of knowledge/belief (this point will be examined later).
The most intriguing properties of normal modal logics follow from the properties of the
accessibility relation, R, in models. To illustrate these properties, consider the following axiom
schema: □<p =-<p. It turns out that this axiom is characteristic of the class of models with a reflexive
accessibility relation. (By characteristic, we mean that it is true in all and only those models in the
class.) There are a host of axioms which correspond to certain properties of R: the study of the way
that properties of R correspond to axioms is called correspondence theory. For our present
purposes, we identify just four axioms: the axiom called T (which corresponds to a reflexive
accessibility relation); D (serial accessibility relation); 4 (transitive accessibility relation); and 5
(euclidean accessibility relation):
T □q; => q; D □,p ~ ◊'P
4 □q; => □□q; 5 ◊'P ~ □◊'PThe results of correspondence theory make it straightforward to derive completeness results for a
range of simple normal modal logics. These results provide a useful point of comparison for normal
modal logics, and account in a large part for the popularity of this style of semantics.
To use the logic developed above as an epistemic logic, the formula □q: is read as: "it is known
that rp". The worlds in the model are interpreted as epistemic alternatives, the accessibility relation
defines what the alternatives arc from any given world.
The logic defined above deals with the knowledge of a single agent. To deal with multi-agent
knowledge, one adds to a model structure an indexed set of accessibility rehitions, one for each
agent. The language is then extended by replacing the single modal operator "O" by an indexed set
of unary modal operators { K1}, where i E {1 , ... , n }. The formula K;r:r is read: •'i knows that cp".
Each operator K, is given exactly the same properties as·'□".
The next step is to consider how well normal modal logic serves as a logic of knowledge/belief.
Consider first the necessitation rule and axiom K, since any normal modal system is committed to
these. The necessitation rule tells us that an agent knows all valid formulae. Amongst other things,
this means an agent knows all propositional tautologies. Since there is an infinite number of these,
an agent will have an infinite number of items of knowledge: immediately, one is faced with a
counter-intuitive property of the knowledge operator. Now consider the axiom K, which says that
an agent's knowledge is closed under implication. Together with the necessitation rule, this axiom
implies that an agent's knowledge is closed under logical consequence: an agent believes all the
logical consequences of its beliefs. This also seems counter intuitive. For example, suppose, like
every good logician, our agent knows Pcano's axioms. Now Fermat's last theorem follows from
Pean o's axioms-but it took the combined efforts of some of the best minds over the past century to
prove it. Yet if our agent's beliefs are closed under logical consequence, then our agent must know
it. So consequential closure, implied by necessitation and the K axiom, seems an overstrong
property for resource bounded reasoners.
These two problems-that of knowing all valid formulae, and that of knowledge/belief being
closed under logical consequence-together constitute the famous logical omniscience problem. It
has been widely argued that this problem makes the possible worlds model unsuitable for
representing resource bounded believers-and any real system is resource bounded.
2.3.1 Axioms for knowledge and belief
We now consider the appropriateness of the axioms D. T, 4, and 5 for logics of knowledge/
belief. The axiom D says that an agent's beliefs are non-contradictory; it can be re-written as:
K,cp => -.K, ..,<p, which is read: '•if i knows rp, then i doesn't know -irp'". This axiom seems a
reasonable property of knowledge/belief. The axiom Tis often called the knowledge axiom, since it
says that what is known is true. It is usually accepted as the axiom that distinguishes knowledge

M. WOOLDRIDGE AND NICHOLAS JENNINGS 124
from belief: it seems reasonable that one could believe something that is false, but one would
hesitate to say that one could know something false. Knowledge is thus often defined as true belief;
i knows cp if i believes <p and <pis true. So defined, knowledge satisfies T. Axiom 4 is called the
positive introspection axiom. Introspection is the process of examining one's own beliefs, and is
discussed in detail in (Konolige, 1986a, Chapter 5). The positive introspection axiom says that an
agent is aware of what it knows. Similarly, axiom 5 is the negative introspective axiom, which says
that an agent is aware of what it doesn't know. Positive and negative introspection together imply
an agent has perfect knowledge about what it does and doesn't know (cf. (Konolige, 1986a,
Equation (5.11), p. 79)). Whether or not the two types of introspection are appropriate properties
for knowledge/belief is the subject of sonie debate. However, it is generally accepted that positive
introspection is a less demanding property than negative introspection, and is thus a more
reasonable property for resource bounded reasoners.
Given the comments above, the axioms KTD45 are often chosen as a logic of (idealised)
knowledge, and KD45 as a logic of (idealised) belief.
2.4 Alternatives to the possible worlds model
As a result of the difficulties with logical omniscience, many researchers have attempted to develop
alternative formalisms for representing belief. Some of these are attempts to adapt the basic
possible worlds model; others represent significant departures from it. In the subsections that
follow, we examine some of these attempts.
2.4.1 Levesque-belief and awareness
In a 1984 paper, Levesque proposed a solution to the logical omniscience problem that involves
making a distinction between explicit and implicit belief (Levesque, 1984). Crudely, the idea is that
an agent has a relatively small set of explicit beliefs, and a very much larger (infinite) set of implicit
beliefs, which includes the logical consequences of the explicit beliefs. To fonnalise this idea,
Levesque developed a logic with two operators; one each for implicit and explicit belief. The
semantics of the explicit belief operator were given in terms of a weakened possible worlds
semantics, by borrowing some ideas from situation semantics (Barwise & Perry, 1983; Devlin,
1991). The semantics of the implicit belief operator were given in terms of a standard possible
worlds approach. A number of objections have been raised to Levesque's model (Reichgelt, 1989b.
p. 135): first, it does not allow quantification-this drawback has been rectified by Lakemeycr
(1991); second, it docs not seem to allow for nested beliefs; third, the notion of a situation, which
underlies Levesque's logic is, if anything, more mysterious than the notion of a world in possible
worlds; and fourth, under certain circumstances, Levesque's proposal still makes unrealistic
predictions about agent's reasoning capabilities.
In an effort to recover from this last negative result, Fagin and Halpern have developed a "logic
of general awareness" based on a similar idea to Levesque's but with a very much simpler semantics
(Fagin & Hapern, 1985). However, this proposal has itself been criticised by some (Konolige,
1986b).
2.4.2 Konolige-the deduction model
A more radical approach to modelling resource bounded believers was proposed by Konolige
(Konolige, 1986a). His deduction model of belief is, in essence, a direct attempt to model the
"beliefs" of :.ymbolic Al systems. Konolige observed that a typical knowledge-based system has
two key components: a database of symbolically represented "beliefs" (which may take the form of
rules. frames, semantic nets, or, more generally, formulae in some logical language), and some
logically incomplete inference mechanism. Konolige modelled such systems in terms of deduction
structures. A deduction structure is a pair d = (ii, p), where ~ is a base set of formulae in some
logical language, and pis a set of inference rules (which may be logically incomplete), representing
the agent's reasoning mechanism. To simplify the formalism, Konolige assumed that an agent

Intelligent agents: theory and practice 125
would apply its inference rules wherever possible, in order to generate the deductive closure of its
base beliefs under its deduction rules. We model deductive closure in a function close:
where 6.1--,, rp means that rpcan be proved from 6. using only the rules in p. A belief logic can then be
defined, with the semantics to a modal belief connective [i], where i is an agent, given in terms of the
deduction structured; modelling i's belief system: [i]qi iff rp e c!ose(d;).
Konolige went on to examine the properties of the deduction model at some length, and
developed a variety of proof methods for his logics, including resolution and tableau systems
(Geissler & Konolige, 1986). The deduction model is undoubtedly simple; however, as a direct
model of the belief systems of AI agents, it has much to commend it.
2.4.3 Meta-languages and syntactic modalities
A meta-language is one in which it is possible to represent the properties of another language. A
first-order meta-language is a first-order logic, with the standard predicates, quantifier, terms, and
so on, whose domain contains formulae of some other language, called the object language. Using a
meta-language, it is possible to represent a relationship between a meta-language term denoting an
agent, and an object language term denoting some formula. For example, the meta-language
formula Bel(Janine,[Father(Zeus, Cronos)]) might be used to represent the example (1) that we
saw earlier. The quote marks, [ ... ], are used to indicate that their contents are a meta-language
term denoting the corresponding object-language formula.
Unfortunately, meta-language formalisms have their own package of problems, not the least of
which is that they tend to fall prey to inconsistency (Montague, 1963; Thomason, 1980). However,
there have been some fairly successful meta-language formalisms, including those by Konolige
(1982), Haas (1986), Morgenstern (1987), and Davies (1993). Some results on retrieving consist
ency appeared in the late 1980s (Pcrlis, 1985, 1988; des Rivieres & Levesque, 1986; Turner, 1990).
2.5 Pro-attitudes: goals and desires
An obvious approach to developing a logic of goals or desires is to adapt possible worlds
semantics-sec, e.g .. Cohen and Levesque (1990a), Wooldridge (1994). In this view, each goal
accessible world represents one way the world might be if the agent's goals were realised. However,
this approach falls prey to the side effect problem, in that it predicts that agents have a goal of the
logical consequences of their goals (cf. the logical omniscience problem, discussed above). This is
not a desirable property: one might have a goal of going to the dentist, with the necessary
consequence of suffering pain, without having a goal of suffering pain. The problem is discussed (in
the context of intentions), in Bratman ( 1990). The basic possible worlds model has been adapted by
some researchers in an attempt to overcome this problem (Wainer, 1994). Other, related semantics
for goals have been proposed (Doyle et al., 1991; Kiss & Reichgelt, 1992; Rao& Georgeff, 1991b).
2.6 Theories of agency
All of the formalisms considered so far have focused on just one aspect of agency. However, it is to
be expected that a realistic agent theory will be represented in a logical framework that combines
these various components. Additionally, we expect an agent logic to be capable of representing the
dynamic aspects of agency. A complete agent theory, expressed in a logic with these properties,
must define how the attributes of agency are related. For example, it will need to show how an
agent's information and pro-attitudes are related; how an agent's cognitive state changes over time;
how the environment affects an agent's cognitive state; and how an agent's information and pro
attitudes lead it to perform actions. Giving a good account of these relationships is the most
significant problem faced by agent theorists.

M. WOOLDRIDGE AND NICHOLAS JENNINGS 126
An all-embracing agent theory is some time off, and yet signficant steps have been taken towards
it. In the following subsections, we briefly review some of this work.
2.6. I Moore-knowledge and action
Moore was in many ways a pioneer of the use of logics for capturing aspects of agency (Moore,
1990). His main concern was the study of knowledge pre-conditions for actions-the question of
what an agent needs to know in order to be able to perform some action. He formalised a model of
ability in a logic containing a modality for knowledge, and a dynamic logic-like apparatus for
modelling action (cf. Hare!, 1984). This formalism allowed for the possibility of an agent having
incomplete information about how to achieve some goal, and performing actions in order to find
out how to achieve it. Critiques of the formalism (and attempts to improve on it) may be found in
Morgenstern (1987) and Lesperance (1989).
2.6.2 Cohen and Levesque-intention
One of the best-known and most influential contributions to the area of agent theory is due to
Cohen and Levesque (1990a). Their formalism was originally used to develop a theory of intention
(as in "I intend to ... "), which the authors required as a pre-requisite for a theory of speech acts
(Cohen & Levesque, 1990b). However, the logic has subsequently proved to be so useful for
reasoning about agents that it has been used in an analysis of conflict and cooperation in multi
agent dialogue (Galliers, 1988a,b), as well as several studies in the theoretical foundations of
cooperative problem solving (Levesque ct al., 1990; Jennings, 1992; Castelfranchi, 1990; Castel
franchi et al., 1992). Here, we shall review its use in developing a theory of intention.
Following Bratman (1990), Cohen and Levesque identify seven properties that must be satisfied
by a reasonable theory of intention:
1. Intentions pose problems for agents, who need to determine ways of achieving them.
2. Intentions provide a "filter" for adopting other intentions, which must not conflict.
3. Agents track the success of their intentions, and arc inclined to try again if their attempts fail.
4. Agents believe their intentions are possible.
5. Agents do not believe they will not bring about their intentions.
6. Under certain circumstances, agents believe they will bring about their intentions.
7. Agents need not intend ail the expected side effects of their intentions.
Given these criteria, Cohen and Levesque adopt a two-tiered approach to the problem of
formalising intention. First, they construct a logic of rational agency, "being careful to sort out the
relationships among the basic modal operators" (Cohen & Levesque, 1990a, p. 221). Over this
framework, they introduce a number of derived constructs, which constitute a "partial theory of
rational action" (Cohen & Levesque, 1990a, p. 221); intention is one of these constructs.
The first major derived construct is the persistent goal. An agent has a persistent goal of rp iff:
l. It has a goal that q; eventually becomes true, and believes that rp is not currently true.
2. Before it drops the goal cp, one of the following conditions must hold: i the agent believes cp has
been satisfied; or ii the agent believes cp will never be satisfied.
It is a small step from persistent goals to a first definition of intention, as in '•intending to act'': an
agent intends to do action a iff it has a persistent goal to have brought about a state wherein it
believed it was about to do (1, and then did a. Cohen and Levesque go on to show how such a
definition meets manyof Bratman'scritcria for a theory of intention (outlined above). A critique of
Cohen and Levesque's theory of intention may be found in Singh (1992).
2.6.3 Rao and Georgeff-belief, desire, intention architectures
As we observed earlier, there is no clear consensus in either the Al or philosophy communities
about precisely which combination of information and pro-attitudes are best suited to characteris
ing rational agents. In the work of Cohen and Levesque, described above, just two basic attitudes

Intelligent agents: theory and practice 127
were used: beliefs and goals. Further attitudes, such as intention, were defined in terms of these. In
related work, Rao and Georgeff have developed a logical framework for agent theory based on
three primitive modalities: beliefs, desires and intentions (Rao & Georgeff, 1991a,b, 1993). Their
formalism is based on a branching model of time (cf. Emerson & Halpern, 1986), in which belief-,
desire- and intention-accessible worlds are themselves branching time structures.
They are particularly concerned with the notion of realism-the question of how an agent's
beliefs about the future affect its desires and intentions. In other work, they also consider the
potential for adding (social) plans to their formalism (Rao & Georgcff, 1992b; Kinny et al., 1992).
2.6.4 Singh
A quite different approach to modelling agents was taken by Singh, who has developed an
interesting family of logics for representing intentions, beliefs, knowledge, know-how, and
communication in a branching-time framework (Singh, 1990, 199la,b; Singh & Asher, 1991); these
articles are collected and expanded in Singh (1994). Singh's formalism is extremely rich, and
considerable effort has been devoted to establishing its properties. However, its complexity
prevents a detailed discussion here.
2.6.5 Werner
In an extensive sequence of papers, Werner has laid the foundations of a general model of agency,
which draws upon work in economics, game theory, situated automata theory, situation semantics,
and philosophy (Werner, 1988, 1989, 1990, 1991). At the time of writing, however, the properties
of this model have not been investigated in depth.
2.6.6 Wooldridge-modelling multi-agent systems
For his 1992 doctoral thesis, Wooldridge developed a family of logics for representing the
properties of multi-agent systems (Wooldridge, 1992; Wooldridge & Fisher, 1992). Unlike the
approaches cited above, Wooldridge's aim was not to develop a general framework for agent
theory. Rather, he hoped to construct formalisms that might be used in the specification and
verification of realistic multi-agent systems. To this end, he developed a simple, and in some sense
general, model of multi-agent systems, and showed how the histories traced out in the execution of
such a system could be used as the semantic foundation for a family of both linear and branching
time temporal belief logics. He then gave examples of how these logics could be used in the
specification and verification of protocols for cooperative action.
2. 7 Communication
Formalisms for representing communication in agent theory have tended to be based on speech act
theory, as originated by Austin (1962), and further developed by Searle (1969) and others (Cohen
& Perrault, 1979; Cohen & Levesque, 1990a). Briefly, the key axiom of speech act theory is that
communicative utterances arc actions, in just the sense that physical actions arc. They are
performed by a speaker with the intention of bringing about a desired change in the world:
typically, the speaker intends to bring about some particular mental state in a listener. Speech acts
may fail in the same way that physical actions may fail: a listener generally has control over her
mental state, and cannot be guaranteed to react in the way that the speaker intends, Much work in
speech act theory has been devoted to classifying the various different types of speech acts. Perhaps
the two most widely recognised categories of speech acts are representatives (o f which informing is
the paradigm example), and directives (of which requesting is the paradigm example).
Although not directly based on work in speech acts (and arguably more to do with architectures
than theories), we shall here mention work on agent communication languages (Genesereth &
Ketchpel, 1994). The best known work on agent communication languages is that by the ARPA
knowledge sharing effort (Patil et al,, 1992). This work has been largely devoted to developing two
related languages: the knowledge query and manipulation language (KQML) and the knowledge

M. WOOLDRIDGE AND NICHOLAS JENNINGS 128
interchange format (KIF). KOML provides the agent designer with a standard syntax for
messages, and a number of performatives that define the force of a message. Example performatives include tell, perform, and reply; the inspiration for these message types comes largely from
speech act theory. KIF provides a syntax for message content-KIF is essentially the first-order
predicate calculus, recast in a LISP-like syntax.
2.8 Discussion
Formalisms for reasoning about agents have come a long way since Hintikka's pioneering work on
logics of knowledge and belief (Hintikka, 1962). Within AI, perhaps the main emphasis of
subsequent work has been on attempting to develop formalisms that capture the relationship
between the various elements that comprise an agent's cognitive state; the paradigm example of
this work is the well-known theory of intention developed by Cohen and Levesque (1990a).
Despite the very real progress that has been made, there still remain many fairly fundamental
problems and issues still outstanding.
On a technical level, we can identify a number of issues that remain open. First, the problems
associated with possible worlds semantics (notably, logical omniscience) cannot be regarded as
solved. As we observed above, possible worlds remain the semantics of choice for many
researchers, and yet they do not in general represent a realistic model of agents with limited
resources-and of course all real agents are resource-bounded. One solution is to ground possible
worlds semantics, giving them a precise interpretation in terms of the world. This was the approach
taken in Rosenschein and Kaelbling's situated automata paradigm, and can be very successful.
However, it is not clear how such a grounding could be given to proattitudes such as desires or
intentions (although some attempts have been made (Singh, 1990a; Wooldridge, 1992; Werner,
1990)). There is obviously much work remaining to be done on formalisms for knowledge and
belief, in particular in the area of modelling resource bounded reasoners.
With respect to logics that combine different attitudes, perhaps the most important problems
still outstanding relate to intention. In particular, the relationship between intention and action has
not been formally represented in a satisfactory way The problem seems to be that having an
intention to act makes it more likely that an agent will act, but does not generally guarantee it.
While it seems straightforward to build systems that appear to have intentions (Wooldridge, 1995),
it seems much harder to capture this relationship formally. Other problems that have not yet really
been addressed in the literature include the management of multiple, possibly conflicting
intentions, and the formation, scheduling, and reconsideration of intentions.
The question of exactly which combination of attitudes is required to characterise an agent is
also the subject of some debate. As we observed above, a currently popular approach is to use a
combination of beliefs, desires, and intentions (hence BDI architectures (Rao and Georgeff,
199lb)). However, there are alternatives: Shoham, for example, suggests that the notion of choice
is more fundamental (Shoham, 1990). Comparatively little work has yet been done on formally
comparing the suitability of these various combinations. One might draw a parallel with the use of
temporal logics in mainstream computer science, where the expressiveness of specification
languages is by now a well-understood research area (Emerson & Halpern, 1986). Perhaps the
obvious requirement for the short term is experimentation with real agent specifications, in order
to gain a better understanding of the relative merits of different formalisms.
More general!y, the kinds of logics used in agent theory tend to be rather elaborate, typically
containing many modalities which interact with each other in subtle ways. Very little work has yet
been carried out on the theory underlying such logics (perhaps the only notable exception is
Catach, 1988). Until the general principles and limitations of such multi-modal logics become
understood, we might expect that progress with using such logics will be slow. One area in which
work is likely to be done in the near future is theorem proving techniques for multi-modal logics.
Finally, there is often some confusion about the role played by a theory of agency. The view we
take is that such theories represent specifications for agents. The advantage of treating agent

Intelligent agents: theory and practice 129
theories as specifications, and agent logics as specification languages, is that the problems and
issues we then face are familiar from the discipline of software engineering: How useful or
expressive is the specification language? How concise are agent specifications? How does one
refine or otherwise transform a specification into an implementation? However, the view of agent
theories as specifications is not shared by all researchers. Some intend their agent theories to be
used as knowledge representation formalisms, which raises the difficult problem of algorithms to
reason with such theories. Still others intend their work to formalise a concept of interest in
cognitive science or philosophy (this is, of course, what Hintikka intended in his early work on
logics of knowledge of belief). What is clear is that it is important to be precise about the role one
expects an agent theory to play.
2. 9 Further reading
For a recent discussion on the role of logic and agency, which lays out in more detail some
contrasting views on the subject, see Israel (1993, pp. 17-24). For a detailed discussion of
intentionality and the intentional stance, see Dennett (1978, 1987). A number of papers on AI
treatments of agency may be found in Allen et al. ( 1990). For an introduction to modal logic, sec
Chellas (1980); a slightly older, though more wide ranging introduction, may be found in Hughes
and Cresswell (1968). As for the use of modal logics to model knowledge and belief, see Halpern
and Moses (1992), which includes complexity results and proof procedures. Related work on
modelling knowledge has been done by the distributed systems community, who give the worlds in
possible worlds semantics a precise interpretation; for an introduction and further references, see
Halpern (1987) and Fagin et al. (1992). Overviews of formalisms for modelling belief and
knowledge may be found in Halpern (1986), Konolige (1986a), Reichgelt (1989a) and Wooldridge
(1992). A variant of the possible worlds framework, called the recursive modelling method, is
described in Gmytrasiewicz and Durfee (1993); a deep theory of belief may be found in Mack
(1994). Situation semantics, developed in the early 1980s and recently the subject of renewed
interest, represent a fundamentally new approach to modelling the world and cognitive systems
(Barwise & Perry, 1983; Devlin, 1991). However, situation semantics are not (yet) in the
mainstream of (D)AJ, and it is not obvious what impact the paradigm will ultimately have.
Logics which integrate time with mental states are discussed in Kraus and Lehmann (1988),
Halpern and Vardi (1989) and Wooldridge and Fisher (1994); the last of these presents a tableau
based proof method for a temporal belief logic. Two other important references for temporal
aspects are Shoham (1988. 1989). Thomas has developed some logics. for representing agent
theories as part of her framework for agent programming languages; see Thomas et al. (1991) and
Thomas (1993) and section 4. For an introduction to temporal logics and related topics, see
Goldblatt (1987) and Emerson (1990). A non-formal discussion of intention may be found in
Bratman (1987), or more briefly (Bratman, 1990). Further work on modelling intention may be
found in Grosz and Sidner (1990), Sadek (1992), Goldman and Lang (1991), Konolige and Pollack
(1993), Bell (1995) and Dongha (1995). Related work, focusing less on single-agent attitudes, and
more on social aspects, is Levesque et al. (1990), Jennings (1993a), Wooldridge (1994) and
Wooldridge and Jennings (1994).
Finally, although we have not discussed formalisms for reasoning about action here, we
suggested above that an agent logic would need to incorporate some mechanism for representing
agent's actions. Our reason for avoiding the topic is simply that the field is so big, it deserves a
whole review in its own right. Good starting points for AI treatments of action arc Allen (1984),
and Allen et al. (1990, 1991). Other treatments of action in agent logics arc based on formalisms
borrowed from mainstream computer science, notably dynamic logic (originally developed to
reason about computer programs) (Hare!, 1984). The logic of seeing to it that has been discussed in
the formal philosophy literature, but has yet to impact on (D)AI (Belnap & Perloff, 1988; Perloff,
1991; Belnap, 1991; Segerberg, 1989).

M. WOOLDRIDGE A:-!D NICHOi.AS JENNINGS 130
3 Agent architectures
Until now, this article has been concerned with agent theory-the construction of formalisms for
reasoning about agents, and the properties of agents expressed in such formalisms. Our aim in this
section is to shift the emphasis from theory to practice. We consider the issues surrounding the
construction of computer systems that satisfy the properties specified by agent theorists. This is the
area of agent architectures. Maes defines an agent architecture as:
"[A] particular methodology for building [agents]. It specifies how ... 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 ... and future internal state of the agent.
An architecture encompasses techniques and algorithms that support this methodology'· (Maes, 1991,
p.115)
Kaelbling considers an agent architecture to be:
"[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." (Kaelbling,
1991, p.86)
The classical approach to building agents is to view them as a particular type of knowledge-based
system. This paradigm is known as symbolic Al: we begin our review of architectures with a look at
this paradigm, and the assumptions that underpin it.
3.1 Classical approaches: deliberative architectures
The foundation upon which the symbolic AI paradigm rests is the physical-symbol system
hypothesis, formulated by Newell and Simon (1976). A physical symbol system is defined to be a
physically realisable set of physical entities (symbols) that can be combined to form structures, and
which is capable of running processes that operate on those symbols according to symbolically
coded sets of instructions. The physical-symbol system hypothesis then says that such a system is
capable of general intelligent action.
ft is a short step from the notion of a physical symbol system to Mc Carthy's dream of a sentential
processing automaton, or deliberative agent. (The term "deliberative agent" seems to have derived
from Genesercth"s use of the the term "deliberate agent'' to mean a specific type of symbolic
architecture (Genesereth and Nilsson, 1987, pp. 325-327).) We define a deliberative agent or agent
architecture to be one that contains an explicitly represented, symbolic model of the world, and in
which decisions (for example about what actions to perform) arc made via logical (or at least
pseudo-logical) reasoning, based on pattern matching and symbolic manipulation. The idea of
deliberative agents based on purely logical reasoning is highly seductive: to get an agent to realise
some theory of agency one might naively suppose that it is enough to simply give it logical
representation of this theory and "get it to do a bit of theorem proving" (Shardlow. 1990, section
3.2). If one aims to build an agent in this way, then there are at least two important problems to be
solved:
1. The transduction problem: that of translating the real world into an accurate, adequate
symbolic description, in time for that description to be useful.
2. The representation/reasoning problem: that of how to symbolically represent information
about complex real-world entities and processes, and how to get agents to reason with this
information in time for the results to be useful.
The former problem has led to work on vision, speech understanding, learning. etc. The latter has
led to work on knowledge representation, automated reasoning, automatic planning, etc. Despite
the immense volume of work that these problems have generated, most researchers would accept
that neither is anywhere near solved. Even seemingly trivial problems, such as commonsense

Intelligent agents: theory and practice 131
reasoning, have turned out to be extremely difficult (cf. the CYC project (Guba & Lenat, 1994)).
The underlying problem seems to be the difficulty of theorem proving in even very simple logics,
and the complexity of symbol manipulation algorithms in general: recall that first-order logic is not
even decidable, and modal extensions to it (including representations of belief, desire, time, and so
on) tend to be highly undecidable. Thus, the idea of building "agents as theorem provers"-what
might be called an extreme logicist view of agency-although it is very attractive in theory, seems,
for the time being at least, to be unworkable in practice. Perhaps more troubling for symbolic AI is
that many symbol manuipulation algorithms of interest are intractable. lt seems hard to build
useful symbol manipulation algorithms that will he guaranteed to terminate with useful results in an
acceptable fixed time bound. And yet such algorithms seem essential if agents are to operate in any
real-world. time-constrained domain. Good discussions of this point appear in Kaelbling (1986)
and Russell and Wefald (1991).
It is because of these problems that some researchers have looked to alternative techniques for
building agents; such alternatives are discussed in section 3.2. First, however, we consider efforts
made within the symbolic Al community to construct agents.
3.1.1 Planning agents
Since the early 1970s, the AI planning community has been closely concerned with the design of
artificial agents; in fact, it seems reasonable to claim that most innovations in agent design have
come from this community. Planning is essentially automaticprngamming: the design of aeourse of
action that, when executed, will result in the achievement of some desired goal. Within the
symbolic AI community, it has long been assumed that some form of Al planning system will be a
central component of any artificial agent. Perhaps the best-known early planning system was
STRIPS (Fikes & Nilsson, 1971). This system takes a symbolic description of both the world and a
desired goal state, and a set of action descriptions, which characterise the pre- and post-conditions
associated with various actions. It then attempts to find a sequence of actions that will achieve the
goal, by using a simple means-ends analysis. which essentially involves matching the post
conditions of actions against the desired goal. The STRIPS planning algorithm was very simple,
and proved to be ineffective on problems of even moderate complexity. Much effort was
subsequently devoted to developing more effective techniques. Two major innovations were
hierarchical and non-linear planning (Sacerdoti, 1974, 1975). However, in the mid 1980s, Chapman
established some theoretical results which indicate that even such refined techniques will ultimately
turn out to be unusable in any time-constrained system (Chapman, 1987). These results have had a
profound influence on subsequent AI planning research; perhaps more than any other, they have
caused some researchers to question the whole symbolic AI paradigm, and have thus led to the
work on alternative approaches that we discuss in section 3.2.
In spite of these difficulties, various attempts have been made to construct agents whose primary
component is a planner. For example: the Integrated Planning, Execution and Monitoring (IPEM)
system is based on a sophisticated non-linear planner (Ambros-Ingerson and Steel, 1988); Wood's
AUTODRIVE system has planning agents operating in a highly dynamic environment (a traffic
simulation) (Wood, 1993); Etzioni has built "soft bots" that can plan and act in a Unix environment
(Etzioni et al., 1994); and finally, Cohen's PHOENIX system includes planner-based agents that
operate in the domain of simulated forest fire management (Cohen et al., 1989).
3.1.2 Rratman, Israel and Pollack-IRMA
In section 2, we saw that some researchers have considered frameworks for agent theory based on
beliefs, desires, and intentions (Rao & Georgeff, 1991b). Some researchers have also developed
agent architectures based on these attitudes. One example is the Intelligent Resource-hounded
Machine Architecture (IRMA) (Bratman et a!., 1988). This architecture has four key symbolic data
structures: a plan library, and explicit representations of beliefs, desires, and intentions. Addition
ally, the architecture has: a reasoner, for reasoning about the world; a means-end analyser, for
determining which plans might be used to achieve the agent's intentions; an opportunity analyser,

M. WOOLDRIDGE AND NICHOLAS JENNINGS 132
which monitors the environment in order to determine further options for the agent; a filtering
process; and a deliberation process. The filtering process is responsible for determining the subset
of the agent's potential courses of action that have the property of being consistent with the agent's
current intentions. The choice between competing options is made by the deliberation process. The
IRMA architecture has been evaluated in an experimental scenario known as the Tileworld
(Pollack & Ringuette, 1990).
3.1.3 Vere and Bickmore-HOMER
An interesting experiment in the design of intelligent agents was conducted by Vere and Bickmore
(1990). They argued that the enabling technologies for intelligent agents are sufficiently developed
to be able to construct a prototype autonomous agent, with linguistic ability, planning and acting
capabilities, and so on. They developed such an agent, and christened it HOMER. This agent is a
simulated robot submarine, which exists in a two-dimensional "Seaworld'', about which it has only
partial knowledge. HOMER takes instructions from a user in a limited subset of English with about
an 800 word vocubulary; instructions can contain moderately sophisticated temporal references.
HOMER can plan how to achieve its instructions (which typically relate to collecting and moving
items around the Seaworld), and can then execute its plans, modifying them as required during
execution. The agent has a limited episodic memory, and using this, is able to answer questions
about its past experiences.
3.2.4 Jennings-GRATE*
GRATE* is a layered architecture in which the behaviour of an agent is guided by the mental
attitudes of beliefs, desires, intentions and joint intentions (Jennings, 1993b). Agents are divided
into two distinct parts: a domain level system and a cooperation and control layer. The former
solves problems for the organisation; be it in the domain of industrial control, finance or
transportation. The latter is a meta-level controller which operates on the domain level system with
the aim of ensuring that the agent's domain level activities are coordinated with those of others
within the community. The cooperation layer is composed of three generic modules: a control
module which interfaces to the domain level system, a situation assessment module and a
cooperation module. The assessment and cooperation modules provide an implementation of a
model of joint responsibility (Jennings, 1992), which specifics how agents should act both locally
and towards other agents whilst engaged in cooperative problem solving. The performance of a
GRATE* community has been evaluated against agents which only have individual intentions, and
agents which behave in a selfish manner, in the domain of electricity transportation management.
A significant improvement was noted when the situation became complex and dynamic (Jennings,
1995).
3.2 Alternative approaches: reactive architectures
As we observed above, there arc many unsolved (some would say insoluble) problems associated
with symbolic Al. These problems have led some researchers to question the viability of the whole
paradigm, and to the development of what are generally known as reactive architectures. For our
purposes, we shall define a reactive architecture to be one that does not include any kind of central
symbolic world model, and does not use complex symbolic reasoning.
3.2. l Brooks-behaviour languages
Possibly the most vocal critic of the symbolic AI notion of agency has been Rodney Brooks, a
researcher at MIT who apparently became frustrated by AI approaches to building control
mechanisms for autonomous mobile robots. In a 1985 paper, he outlined an alternative architec
ture for building agents, the so called subsumption architecture (Brooks, 1986). The review of
alternative approaches begins with Brooks' work.
In recent papers, Brooks (1990, 1991a,b) has propounded three key theses:

Intelligent agents: theory and practice 133
1. Intelligent behaviour can be generated without explicit representations of the kind that symbolic
AI proposes.
2. Intelligent behaviour can be generated without explicit abstract reasoning of the kind that
symbolic AI proposes.
3. Intelligence is an emergent property of certain complex systems.
Brooks identifies two key ideas that have informed his research:
1. Situatedness and embodiment: "Real" intelligence is situated in the world, not in disembodied
systems such as theorem provers or expert systems.
2. Intelligence and emergence: "Intelligent" behaviour arises as a result of an agent's interaction
with its environment. Also, intelligence is "in the eye of the beholder"; it is not an innate,
isolated property.
If Brooks was just a Dreyfus-style critic of AI, his ideas might not have gained much currency.
However, to demonstrate his claims, he has built a number of robots, based on the suhsumption
architecture. A subsumption architecture is a hierarchy of task-accomplishing behaviours. Each
behaviour "competes" with others to exercise control over the robot. Lower layers represent more
primitive kinds of behaviour (such as avoiding obstacles), and have precedence over layers further
up the hierarchy. It should be stressed that the resulting systems are, in terms of the amount of
computation they need to do, extremely simple, with no explicit reasoning of the kind found in
symbolic AI systems. But despite this simplicity, Brooks has demonstrated the robots doing tasks
that would be impressive if they were accomplished by symbolic AI systems. Similar work has been
reported by Steels, who described simulations of "Mars explorer" systems, containing a large
number of subsumption-architecture agents, that can achieve near-optimal performance in certain
tasks (Steels, 1990).
3.2.2 Agre and Chapman-PENG!
At about the same time as Brooks was describing his first results with the subsumption architecture,
Chapman was completing his Master's thesis, in which he reported the theoretical difficulties with
planning described above, and was coming to similar conclusions about the inadequacies of the
symbolic AI model himself. Together with his co-worker Agre, he began to explore alternatives to
the AI planning paradigm (Chapman & Agre, 1986).
Agre observed that most everyday activity is ··routine" in the sense that it requires little-if
any-new abstract reasoning. Most tasks, once learned, can be accomplished in a routine way, with
little variation. Agre proposed that an efficient agent architecture could be based on the idea of
''running arguments". Crudely, the idea is that as most decisions are routine, they can be encoded
into a low-level structure (such as a digital circuit), which only needs periodic updating, perhaps to
handle new kinds of problems. His approach was illustrated with the celebrated PENGI system
(Agre & Chapman, 1987). PENGI is a simulated computer game, with the central character
controlled using a scheme such as that outlined above.
3.2.3 Rosenschein and Kaelhling-situated automata
Another sophisticated approach is that of Rosenschein and Kaclbling (Rosenschein, 1985;
Rosenschein & Kaelbling, 1986; Kaelbling & Rosenschcin, 1990; Kaelbling, 1991). In their situated
automata paradigm, an agent is specified in declarative terms. This specification is then compiled
down to a digital machine, which satisfies the declarative specification. This digital machine can
operate in a provably time-bounded fashion; it does not do any symbol manipulation, and in fact no
symbolic expressions arc represented in the machine at all. The logic used to specify an agent is
essentially a modal logic of knowledge (see above). The technique depends upon the possibility of
giving the worlds in possible worlds semantics a concrete interpretation in terms of the states of an
automaton:

M. WOOLDRIDGE AND NICHOLAS JENNINGS 134
"[An agent} ... x iss aid to carry the information that p inw orld states, written s I= K(x,p), if for all world
states in which x has the same value as it does ins, the proposition pis true." (Kae!bling & Rosenschcin,
1990, p. 36)
An agent is specified in terms of two components: perception and action. Two programs are then
used to synthesise agents: RULER is used to specify the perception component of an agent;
GAPPS is used to specify the action component.
RULER takes as its input three components:
"[A j specification of the semantics of the [a gent's} inputs ("whenever bit 1 is on, it is raining"); a set of static
facts ("whenever it is raining, the ground is wet"); and a specification of the state transitions of the world ("if
the ground is wet, it stays wet until the sun comes out"). The programmer then specifies the desired
semantics for the output ("if this bit is on, the ground is wet"), and the compiler ... [synthesises] a circuit
whose output will have the correct semantics .... All that declarative '"knowledge" has been reduced to a
very simple circuit." (Kaelb!ing, 1991, p. 86)
The GAPPS program takes as its input a set of goal reduction rules (essentially rules that encode
information about how goals can be achieved), and a top level goal, and generates a program that
can be translated into a digital circuit to realise the goal. Once again, the generated circuit does not
represent or manipulate symbolic expressions; all symbolic manipulation is done at compile time.
The situated automata paradigm has attracted much interest, as it appears to combine the best
elements of both reactive and symbolic, declarative systems. However, at the time of writing, the
theoretical limitations of the approach are not well understood; there are similarities with the
automatic synthesis of programs from temporal logic specifications, a complex area of much
ongoing work in mainstream computer science (see the comments in Emerson, 1990).
3.2.4 Maes-Agent network architecture
Pattie Maes has developed an agent architecture in which an agent is defined as a set of competence
modules (Macs, 1989, 1990b, 1991 ). These modules loosely resemble the behaviours of Brooks'
subsumption architecture (above). Each module is specified by the designer in terms of pre- and
post-conditions (rather like STRIPS operators), and an activation level, which gives a real-valued
indication of the relevance of the module in a particular situation. The higher the activation level of
a module, the more likely it is that this module will influence the behaviour of the agent. Once
specified, a set of competence modules is compiled into a spreading activation network, in which the
modules are linked to one-another in ways defined by their pre-and post-conditions. For example,
if module a has post-condition rp, and module b has pre-condition cp, then a and bare connected by
a successor link. Other types of link include predecessor links and conflicter links. When an agent is
executing, various modules may become more active in given situations, and may be executed. The
result of execution may be a command to an effector unit, or perhaps the increase in activation !eve!
of a successor module.
There are obvious similarities between the agent network architecture and neural network
architectures. Perhaps the key difference is that it is difficult to say what the meaning of a node in a
neural net is; it only has a meaning in the context of the net itself. Since competence modules are
defined in declarative terms, however, it is very much easier to say what their meaning is.
3.3 Hybrid architectures
Many researchers have suggested that neither a completely deliberative nor completely reactive
approach is suitable for building agents. They have argued the case for hybrid systems, which
attempt to marry classical and alternative approaches.
An obvious approach is to build an agent out of two (or more) subsystems: a deliberative one,
containing a symbolic world model, which develops plans and makes decisions in the way proposed
by mainstream symbolic AI; and a reactive one, which is capable of reacting to events that occur in
the environment without engaging in complex reasoning. Often, the reactive component is given

Intelligent agents: theory and practice 135
some kind of precedence over the deliberative one, so that it can provide a rapid response to
important environmental events. This kind of structuring leads naturally to the idea of a layered
architecture, of which Touring Machincs (Ferguson, 1992) and Inte RRa P (Muller & Pischel, 1994)
are good examples. (These architectures are described below.) In such an architecture, an agent's
control subsystems are arranged into a hierarchy, with higher layers dealing with information at
increasing levels of abstraction. Thus, for example, the very lowest layer might map raw sensor
data directly onto effector outputs, while the uppermost layer deals with long-term goals. A key
problem in such architectures is what kind of control framework to embed the agent's subsystems
in, to manage the interactions between the various layers.
3.3.1 Georgeff and Lansky-PRS
One of the best-known agent architectures is the Procedural Reasoning System (PRS), developed
by Georgeff and Lansky (1987). Like IRMA (see above), the PRS is a belief-desire-intention
architecture, which includes a plan library, as well as explicit symbolic representations of beliefs,
desires, and intentions. Beliefs are facts, either about the external world or the system's internal
state. These facts are expressed in classical first-order logic. Desires are represented as system
behaviours (rather than as static representations of goal states). A PRSplan library contains a set of
partially-elaborated plans, called knowledge areas (KA), each of which is associated with an
ini,ocation condition. This condition determines when the KA is to be actil'ated. KAs may be
activated in a goal-driven or data-driven fashion; KAs may also be reactive, allowing the PRS to
respond rapidly to changes in its environment. The set of currently active KAs in a system represent
its intentions. These various data structures are manipulated by a system interpreter, which is
responsible for updating beliefs, invoking KAs, and executing actions. The PRS has been
evaluated in a simulation of maintenance procedures for the space shuttle, as well as other domains
(Georgcff & lngrand, 1989).
3.3.2 Ferguson-Touring Machines
For his 1992 Doctoral thesis, Ferguson developed the Touring Machines hybrid agent architecture
(Ferguson, 1992a,b).5 The architecture consists of perception and action subsystems, which
interface directly with the agent's environment, and three control layers, embedded in a control
framework, which mediates between the layers. Each layer is an independent, activity-producing,
concurrently executing process.
The reactive layer generates potential courses of action in response to events that happen too
quickly for other layers to deal with. It is implemented as a set of situation-action rules, in the style
of Brooks' subsumption architecture (see above).
The planning layer constructs plans and selects actions to execute in order to achieve the agent's
goals. This layer consists of two components: a planner, and a focus of attention mechanism. The
planner integrates plan generation and execution. and uses a library of partially elaborated plans,
together with a topological world map, in order to construct plans that will accomplish the agent's
main goal. The purpose of the focus of attention mechanism is to limit the amount of information
that the planner must deal with, and so improve its efficiency. It does this by filtering out irrelevant
information from the environment.
The modelling layer contains symbolic representations of the cognitive state of other entities in
the agent's environment. These models are manipulated in order to identify and resolve goal
conflicts-situations where an agent can no longer achieve its goals, as a result of unexpected
interference.
The three layers are able to communicate with each other (via message passing), and are
embedded in a control framework. The purpose of this framework is to mediate between the
5 1t is worth noting that Fergeson's thesis gives a good overview of the problems and issues associated with
building rational, resource-bounded agents. Moreover. the description given of the Touring Machines
architecture is itself extremely clear. We recommend it as a point of departure for further reading.

M. WOOLDRIDGE AND '.'JICHOLAS JENNINGS 136
layers, and in particular, to deal with conflicting action proposals from the different layers. The
control framework does this by using control rules.
3.3.3 Burmeister et al.-COSY
The COSY architecture is a hybrid BDI-architecture that includes elements of both the PRS and
IRMA, and was developed specifically for a multi-agent testbed called DASEDIS (Burmeister &
Sundermeyer; Haddadi, 1994). The architecture has five main components: (i) sensors; (ii)
actuators; (iii) communications (iv) cognition; and (v) intention. The first three components are
straightforward: the sensors receive non-communicative perceptual input, the actuators allow the
agent to perform non-communicative actions, and the communications component allows the
agent to send messages. Of the remaining two components, the intention component contains
"long-term goals, attitudes, responsibilities and the like ... the control elements taking part in the
reasoning and decision-making of the cognition component" (Haddadi, 1994, p. 15), and the
cognition component is responsible for mediating between the intentions of the agent and its beliefs
about the world, and choosing an appropriate action to perform. Within the cognition component
is the knowledge base containing the agent's beliefs, and three procedural components: a script
execution component, a protocol execution component, and a reasoning, deciding and reacting
component. A script is very much like a script in Schan k's original sense: it is a stereotypical recipe
or plan for achieving a goal. Protocols are stereotypical dialogues representing cooperation
frameworks such as the contract net (Smith, 1980). The reasoning, deciding and reacting
component is perhaps the key component in COSY. It is made up of a number of other subsystems,
and is structured rather like the PRS and IRMA (see above). An agenda is maintained, that
contains a number of active scripts. These scripts may be invoked in a goal-driven fashion (to satisfy
one of the agent's intentions), or a data-driven fashion (in response to the agent's current
situation). A filter component chooses between competing scripts for execution
3.3.4 Muller et al.~lnte RRa P
Intc RRa P, like Ferguson's Touring Machines, is a layered architecture, with each successive layer
representing a higher level of abstraction than the one below it (Millier & Pischel, 1994; Millier et
al., 1995; Millier, 1994). In Inte RRa P, these layers are further subdivided into two vertical layers:
one containing layers of knowledge bases, the other containing various control components, that
interact with the knowledge bases at their level. At the lowest level is the world interface control
component, and the corresponding world level knowledge base. The world interface component,
as its name suggests, manages the interface between the agent and its environment, and thus deals
with acting, communicating, and perception.
Above the world interface component is the behaviour-based component. The purpose of this
component is to implement and control the basic reactive capability of the agent. This component
manipulates a set of patterns of behaviour (Po B). A Po B is a structure containing a pre-condition
that defines when the Po B is to be activated, various conditions that define the circumstances under
which the Po B is considered to have succeeded or failed, a post-condition (a la STRIPS (Fikes &
Nilsson, 1971)), and an executable body, that defines what action should be performed if the Po B is
executed. (The action may be a primitive, resulting in a call on the agent's world interface. or may
involve calling on a higher-level layer to generate a plan.)
Above the behaviour-based component in Tnte RRa P is the plan-based component. This
component contains a planner that is able to generate single-agent plans in response to requests
from the behaviour-based component. The knowledge-base at this layer contains a set of plans,
including a plan library. The highest layer of Inte RRa P is the cooperation component. This
component is able to generate joint plans, that satisfy the goals of a number of agents, by
elaborating plans selected from a plan library. These plans arc generated in response to requests
from the plan-based component.
Control in lnte RRa P is both data- and goal-driven. Perceptual input is managed by the world
interface, and typically results in a change to the world model. As a result of changes to the world

Intelligent agents: theory and practice 137
model, various patterns of behaviour may be activated, dropped, or executed. As a result of Po B
execution, the plan-based <;omponent and cooperation component may be asked to generate plans
and joint plans respectively, in order to achieve the goals of the agent. This ultimately results in
primitive actions and messages being generated by the world interface.
3.4 Discussion
The deliberative, symbolic paradigm is, at the time of writing, the dominant approach in (D)AL
This state of affairs is likely to continue, at least for the near future. There seem to be several
reasons for this. Perhaps most importantly, many symbolic AI techniques (such as rule-based
systems) carry with them an associated technology and methodology that is becoming familiar to
mainstream computer scientists and software engineers. Despite the well-documented problems
with symbolic AI systems, this makes symbolic AI agents (such as GRATE*, Jennings, 1993b) an
attractive proposition when compared to reactive systems, which have as yet no associated
methodology. The need for a development methodology seems to be one of the most pressing
requirements for reactive systems. Anecdotal descriptions of current reactive systems implemen
tations indicate that each such system must be individually hand-crafted through a potentially
lengthy period of experimentation (Wavish and Graham, 1995). This kind of approach seems
unlikely to be usable for large systems. Some researchers have suggested that techniques from the
domain of genetic algorithms or machine learning might be used to get around these development
problems, though this work is at a very early stage.
There is a pressing need for research into the capabilities of reactive systems, and perhaps in
particular to the types of application for which these types of system are best suited; some
preliminary work has been done in this area, using a problem domain known as the Tile World
(Pollack & Ringuette, 1990) With respect to reactive systems, Ferguson suggests that:
"[T]he strength of purely non-deliberative architectures lies in their ability to exploit local patterns of
activity in their current surroundings in order to generate more or less hardwired action responses .. for a
given set of stimuli Successful operation using this method pre-supposes: i that the complete set of
environmental stimuli required for unambiguously determining action sequences is always present and
readily identifiable-in other words, that the agent's activity can be situationally determined; ii that the
agent has no global task constraints ... which need to be reasoned about at run time; and iii that the agent's
goal or desire system is capable of being represented implicitly in the agent's structure according to a fixed,
pre-compiled ranking scheme." (Ferguson. 1992a, pp. 29-30}
Hybrid architectures, such as the PRS, Touring Machines, Inte RRa P, and COSY, are currently a
very active area of work, and arguably have some advantages over both purely deliberative and
purely reactive architectures. However, an outstanding problem with such architectures is that of
combining multiple interacting subsystems (deliberative and reactive) cleanly, in a well-motivated
control framework. Humans seem to manage different levels of abstract behaviour with compari
tive ease; it is not clear that current hybrid architectures can do so.
Another area where as yet very little work has been done is the generation of goals and
intentions. Most work in AT assumes that an agent has a single, well-defined goal that it must
achieve. But if agents are ever to be really autonomous, and act pro-actively, then they must be
able to generate their own goals when either the situation demands, or the opportunity arises.
Some preliminary work in this area is Norman and Long (1995). Similarly, little work has yet been
done into the management and scheduling of multiple, possibly conflicting goals; some preliminary
work is reported in Dongha (1995).
Finally, we turn to the relationship between agent theories and agent architectures. To what
extent do the agent architectures reviewed above correspond to the theories discussed in section 2?
What, if any, is the theory that underpins an architecture? With respect to purely deliberative
architectures, there is a wealth of underlying theory. The close relationship between symbolic
processing systems and mathematical logic means that the semantics of such architectures can often
be represented as a logical system of some kind. There is a wealth of work establishing such

M. WOOLDRIDGE AND NICHOLAS JENNINGS 138
relationships in Al, of which a particularly relevant example is Rao and Georgeff (1992a). This
article discusses the relationship between the abstract BDI logics developed by Rao et al. for
reasoning about agents, and an abstract "agent interpreter", based on the PRS. However, the
relationship between the logic and the architecture is not formalised; the BDI logic is not used to
give a formal semantics to the architecture, and in fact it is difficult to see how such a logic could he
used for this purpose. A serious attempt to define the semantics of a (somewhat simple) agent
architecture is presented in Wooldridge (1995), where a formal model of the system My World, in
which agents are directly programmed in terms of beliefs and intentions, is used as the basis upon
which to develop a logic for reasoning about My World systems. Although the logic contains
modalities for representing beliefs and intentions, the semantics of these modalities are given in
terms of the agent architecture itself, and the problems associated with possible worlds do not,
therefore, arise; this work builds closely on Konolige's models of the beliefs of symbolic AI systems
(Konoligc, 1986a). However, more work needs to be done using this technique to model more
complex architectures, before the limitations and advantages of the approach are well-understood.
Like purely deliberative architectures, some reactive systems are also underpinned by a
relatively transparent theory. Perhaps the best example is the situated automata paradigm, where
an agent is specified in terms of a logic of knowledge, and this specification is compiled down to a
simple digital machine that can he realistically said to realise its corresponding specification.
However, for other purely reactive architectures, based on more ad hoc principles, it is not clear
that there is any transparent underlying theory. It could be argued that hybrid systems also tend to
bead hoc, in that while their structures are well-motivated from a design point of view, it is not clear
how one might reason about them, or what their underlying theory is. In particular, architectures
that contain a number of independent activity producing subsystems, which compete with each
other in real time to control the agent's activities, seem to defy attempts at formalisation. It is a
matter of debate whether this needs he considered a serious disadvantage, but one argument is that
unless we have a good theoretical model of a particular agent or agent architecture, then we shall
never really understand why it works. This is likely to make it difficult to generalise and reproduce
results in varying domains.
3.5 Further reading
Most introductory textbooks on Al discuss the physical symbol system hypothesis; a good recent
example of such a text is Ginsberg (1993). A detailed discussion of the way that this hypothesis has
affected thinking in symbolic AI is provided in Shardlow (1990). There are many objections to the
symbolic AI paradigm, in addition to those we have outlined above. Again, introductory textbooks
provide the stock criticisms and replies.
There is a wealth of material on planning and planning agents. See Georgeff (1987) for an
overview of the state of the art in planning (as it was in 1987), Allen et al. (1990) for a thorough
collection of papers on planning (many of the papers cited above are included), and Wilkins (1988)
for a detailed description of SIPE, a sophisticated planning system used in a real-world application
(the control of a brewery!) Another important collection of planning papers is Georgeff and
Lansky (1986). The book by Dean and Wellman and the book by Allen et al. contain much useful
related material (Dean and Wellman, 1991; Allen et al., 1991 ). There is now a regular international
conference on planning; the proceedings of the first were published as Hendler (1992).
The collection of papers edited by Maes (1990a) contains many interesting papers on alterna
tives to the symbolic AI paradigm. Kaelbling (1986) presents a clear discussion of the issues
associated with developing resource-bounded rational agents, and proposes an agent architecture
somewhat similar to that developed by Brooks. A proposal by Nilsson for teleo reactive programs
goal directed programs that nevertheless respond to their environment-is described in Nilsson
(1992). The proposal draws heavily on the situated automata paradigm; other work based on this
paradigm is described in Shoham (1990) and Kiss and Reichgelt (1992). Schoppcrs has proposed
compiling plans in advance, using traditional planning techniques, in order to develop universal

Intelligent agents: theory and practice 139
plans, which are essentially decision trees that can be used to efficiently determine an appropriate
action in any situation (Schoppers, 1987). Another proposal for building "reactive planners"
involves the use of reactive action packages (Firby, 1987).
Other hybrid architectures are described in Hayes-Roth (1990), Downs and Reichgelt (1991),
Aylett and Eustace (1994) and Bussmann and Demazeau (1994).
4 Agent languages
As agent technology becomes more established, we might expect to see a variety of software tools
become available for the design and construction of agent-based systems; the need for software
support tools in this area was identified as long ago as the mid-1980s (Gasser et al., 1987). The
emergence of a number of prototypical agent languages is one sign that agent technol0gy is
becoming more widely used, and that many more agent-based applications are likely to be
developed in the near future. By an agent language, we mean a system that allows one to program
hardware or software computer systems in terms of some of the concepts developed hy agent
theorists. At the very least, we expect such a language to include some structure corresponding to
an agent. However, we might also expect to sec some other attributes of agency (beliefs, goals, or
other mentalistic notions) used to program agents. Some of the languages we consider below
embody this strong notion of agency; others do not. However, all have properties that make them
interesting from the point of view of this review.
4.0. I Concurrent object languages
Concurrent object languages are in many respects the ancestors of agent languages. The notion of a
self-contained concurrently executing object, with some internal state that is not directly accessible
to the outside world, responding to messages from other such objects, is very close to the concept of
an agent as we have defined it. The earliest concurrent object framework was Hewitt's Actor model
(Hewitt, 1977; Agha, 1986); another well-known example is the ABCL system (Yonezawa, 1990).
For a discussion on the relationship between agents and concurrent object programming, see
Gasser and Briot (1992).
4.0.2 Shoham-agent-oriented programming
Yoav Shoham has proposed a "new programming paradigm, based on a societal view of
computation" (Shoham, 1990, p. 4; 1993). The key idea that informs this agent-oriented program
ming (AOP) paradigm is that of directly programming agents in terms of the mentalistic,
intentional notions that agent theorists have developed to represent the properties of agents. The
motivation behind such a proposal is that, as we observed in section 2, humans use the intentional
stance as an abstraction mechanism for representing the properties of complex systems. ln the same
way that we use the intentional stance to describe humans, it might be useful to use the intentional
stance to program machines.
Shoham proposes that a fully developed AOP system will have three components:
• a logical system for defining the mental state of agents;
• an interpreted programming language for programming agents;
• an "agentification" process, for compiling agent programs into low-level executable systems.
At the time of writing, Shoham has only published results on the first two components. (In Shoham
(1990, p. 12), he wrote that "the third is still somewhat mysterious to me", though later in the paper
he indicated that he was thinking along the lines of Rosenschcin and Kaelbling·s situated automata
paradigm (Rosenschcin & and Kaelbling, 1986).) Shoham·s first attempt at an AOP language was
the AGENT0 system. The logical component of this system is a quantified multi-modal logic,
allowing direct reference to time. No semantics are given, but the logic appears to be based on
Thomas et al. (1991). The logic contains three modalities: belief, commitment and ability. The
following is an acceptable formula of the logic, illustrating its key properties:

M. WOOLDRIDGE AND NICHOLAS JENNINGS 140
=
CA~ open(door)8 Bl CA~ open (door)8.
This formula is read: "if at time 5 agent a can ensure that the door is open at time 8, then at time 5
agent b believes that at time 5 agent a can ensure that the door is open at time 8".
Corresponding to the logic is the AGENT0 programming language. In this language, an agent is
specified in terms of a set of capabilities (things the agent can do), a set of initial beliefs and
commitments, and a set of commitment rules. The key component, which determines how the agent
acts, is the commitment rule set. Each commitment rule contains a message condition, a mental
condition, and an action. To determine whether such a rule fires, the message condition is matched
against the messages the agent has received; the mental condiiion is matched against the beliefs of
the agent. If the rule fires, then the agent becomes committed to the action. Actions may be private,
corresponding to an internally executed subroutine, or communicative, i.e., sending messages.
Messages are constrained to be one of three types: "requests" or "unrequests" to perform or refrain
from actions, and "inform" messages, which pass on information-Shoham indicates that he took
his inspiration for these message types from speech act theory (Searle, 1969; Cohen & Perrault,
1979). Request and unrequest messages typically result in the agent's commitments being
modified; inform messages result in a change to the agent's beliefs.
4.0.3 Tltomas-PLACA
AGENT0 was only ever intended as a prototype, to illustrate the principles of AOP. A more
refined implementation was developed by Thomas, for her 1993 doctoral thesis (Thomas, 1993).
Her Planning Communicating Agents (PLACA) language was intended to address one severe
drawback to AGENT0: the inability of agents to plan, and communicate requests for action via
high-level goals. Agents in PLACA are programmed in much the same way as in AGENT0, in
terms of mental change rules. The logical component of PLACA is similar to AGENTO's, but
includes operators for planning to do actions and achieve goals. The semantics of the logic and its
properties are examined in detail. However, PLACA is not at the "production" stage; it is an
experimental language.
4.0.4 Fisher-Concurrent Metate M
One drawback with both AGENT0 and PLACA is that the relationship between the logic and
interpreted programming language is only loosely defined: in neither case can the programming
language be said to truly execute the associated logic. The Concurrent Metate M language
developed by Fisher can make a stronger claim in this respect (Fisher, 1994). A Concurrent
Metate M system contains a number of concurrently executing agents. each of which is able to
communicate with its peers via asynchronous broadcast message passing. Each agent is pro
grammed by giving it a temporal logic specification of the behaviour that it is intended the agent
should exhibit. An agent's specification is executed directly to generate its behaviour. Execution of
the agent program corresponds to iteratively building a logical model for the temporal agent
specification. It is possible to prove that the procedure used to execute an agent specification is
correct, in that if it is possible to satisfy the specification, then the agent will do so (Barringer et al.,
1989).
The logical semantics of Concurrent Metate M are closely related to the semantics of temporal
logic itself. This means that, amongst other things, the specification and verification of Concurrent
Metate M systems is a realistic proposition (Fisher & Wooldridge, 1993). At the time of writing,
only prototype implementations of the language are available; full implementations are expected
soon.
4.0.5 The IMAGINE Project-APRIL and MAIL
,APRIL (Mc Cabe & Clark, 1995) and MAIL (Haugeneder ct al., 1994) are two languages for
developing multi-agent applications that were developed as part of the ESPRIT project IMAGINE
(Haugenedcr, 1994). The two languages arc intended to fulfil quite different roles. APRIL was

Intelligent agents: theory and practice 141
designed to provide the core features required to realise most agent architectures and systems.
Thus APRIL provides facilities for multi-tasking (via processes, which are treated as first-class
objects, and a Unix-like fork facility), communication (with powerful message-passing facilities
supporting network-transparent agent-to-agent links); and pattern matching and symbolic process
ing capabilities. The generality of APRIL comes at the expense of powerful abstractions-an
APRIL system builder must implement an agent or system architecture from scratch using
APRIL's primitives. In contrast, the MAIL language provides a rich collection of pre-defined
abstractions, including plans and multi-agent plans. APRIL was originally envisaged as the
implementation language for MAIL. The MAIL system has been used to implement several
prototype multi-agent systems, including an urban traffic management scenario (Haugeneder and
Steiner, 1994).
4.0.6 General Magic, lnc.-TELESCRIPT
TELESCRIPT is a language-based environment for constructing agent societies that has been
developed by General Magic, Inc.: it is perhaps the first commercial agent language.
TELESCRIPT technology is the name given by General Magic to a family of concepts and
techniques they have developed to underpin their products. There are two key concepts in
TELESCRIPT technology: places and agents. Places are virtual locations that are occupied by
agents. Agents are the providers and consumers of goods in the electronic marketplace applications
that TELESCRIPTwas developed to support. Agents are software processes, and are mobile: they
are able to move from one place to another, in which case their program and state are encoded and
transmitted across a network to another place, where execution recommences. Agents are able to
communicate with one-another: if they occupy different places, then they can connect across a
network, in much the standard way; if they occupy the same location, then they can meet one
another.
Four components have been developed by General Magic to support TELESCRIPT tech
nology. The first is the TELESCRIPT language. This language "is designed for carrying out
complex communication tasks: navigation, transportation, authentication, access control, ·and so
on" (White, 1994, p.17). The second component is the TELESCRIPT engine. An engine acts as an
interpreter for the TELESCRIPT language, maintains places, schedules agents for execution,
manages communication and agent transport, and finally, provides an interface with other
applications. The third component is the TELESCRIPT protocol set. These protocols deal
primarily with the encoding and decoding of agents, to support transport between places. The final
component is a set of software tools to support the development of TELESCRIPT applications.
4.0.7 Connah and Wavish-ABLE
A group at Philips research labs in the UK have developed an Agent Behaviour Language (ABLE),
in which agents are programmed in terms of simple, rule-like licences (Connah & Wavish, 1990;
Wavish, 1992). Licences may include some representation of time (though the language is not
based on any kind of temporal logic): they loosely resemble behaviours in the subsumption
architecture (see above). ABLE can be compiled down to a simple digital machine, realised in the
"C" programming language. The idea is similar to situated automata, though there appears to
be no equivalent theoretical foundation. The result of the compilation process is a very fast
implementation, which has been used to control a Compact Disk-Interactive (CD-I) application.
ABLE has recently been extended to a version called Real-Time ABLE (RTA) (Wavish &
Graham, 1995).
4.1 Discussion
The emergence of various language-based software tools for building agent applications is clearly
an important development for the wider acceptance and use of agent technology. The release of
TELESCRIPT, a commercial agent language (albeit one that does not embody the strong notion of

M. WOOLDRIDGE A:-1D NICHOLAS JENNINGS 142
agency discussed in this paper) is particularly important, as it potentially makes agent technology
available to a user base that is industrially (rather than academically) oriented.
While the development of various languages for agent-based applications is of undoubted
importance, it is worth noting that all of the academically produced languages mentioned above are
in some sense prototypes. Each was designed either to illustrate or examine some set of principles,
and these languages were not, therefore, intended as production tools. Work is thus needed, both
to make the languages more robust and usable, and to investigate the usefulness of the concepts
that underpin them. As with architectures, work is needed to investigate the kinds of domain for
which the different languages are appropriate.
Finally, we turn to the relationship between an agent language and the corresponding theories
that we discussed in section 2. As with architectures, it is possible to divide agent languages into
various different categories. Thus AGENT0, PLACA, Concurrent Metate M, APRIL, and MAIL
arc deliberative languages, as they arc all based on traditional symbolic AI techniques. ABLE, on
the other hand, is a purely reactive language. With AGENT0 and PLACA, there is a clear (if
informal) relationship between the programming language and the logical theory the language is
intended to realise. In both cases, the programming language represents a subset of the
corresponding logic, which can be interpreted directly. However, the relationship between logic
and language is not formally defined. Like these two languages, Concurrent Metate M is intended
to correspond to a logical theory. But the relationship hetween Concurrent Metate M and the
corresponding logic is much more closely defined, as this language is intended to be a directly
executable version of the logic. Agents in Concurrent Metate M, however, are not defined in terms
of mentalistic constructs. For a discussion on the relationship between Concurrent Metate M and
AGENT0-like languages, see Fisher (1995).
4.2 Further reading
A recent collection of papers on concurrent object systems is Agha ct al. (1993). Various languages
have been proposed that marry aspects of object-based systems with aspects of Shoham·s agent
oriented proposal. Two examples are AGENTSPEAK and DAISY. AGENTSPEAK is loosely
based on the PRS agent architecture, and incorporates aspects of concurrent-object technology
(Weerasooriya et al., 1995). In contrast, DAISY is based on the concurrent-object language CUBL
(Adorni & Poggi, 1993), and incorporates aspects of the agent-oriented proposal (Poggi, 1995).
Other languages of interest include OZ (Henz et al., 1993) and IC PRO LOG Il (Chu, 1993).
The latter, as its name suggests, is an extension of PROLOG, which includes multiple-threads,
high-level communication primitives, and some object-oriented features.
5 Applications
Although this article is not intended primarily as an applications review. it is nevertheless worth
pausing to examine some of the current and potential applications of agent technology.
5.1 Cooperative problem solving and distributed Al
As we observed in section 1, there has been a marked flowering of interest in agent technology
since the mid-1980s. This interest is in part due to the upsurge of interest in Distributed Al
Although DAI encompasses most ofthc issues we have discussed in this paper, it should be stressed
that the dassical emphasis in DAI has been on macro phenomena (the social level), rather than the
micro phenomena (the agent level) that we have been concerned with in this paper. DAI thus looks
at such issues as how a group of agents can be made to cooperate in order to efficiently solve
problems, and how the activities of such a group can be efficiently coordinated. DAI researchers
have applied agent technology in a variety of areas. Example applications include power systems
management (Wittig, 1992; Varga et al., 1994), air-traffic control (Steeb et al., 1988), particle

Intelligent agents: theory and practice 143
accelerator control (Jennings et al., 1993), intelligent document retrieval (Mukhopadhyay et al.,
1986), patient care (Huang et al., 1995), telecommunications network management (Weihmayer &
Velthuijsen, 1994), spacecraft control (Schwuttke & Quan, 1993), computer integrated manufac
turing (Parunak, 1995), concurrent engineering (Cutkosky et al., 1993), transportation manage
ment (Fischer et al., 1993), job shop scheduling (Morley & Schelberg, 1993), and steel coil
processing control (Mori et al., 1988). The classic reference to DAI is Bond and Gasser (1988),
which includes both a comprehensive review article and a collection of significant papers from the
field; a more recent review article is Chaib-draa et al. (1992).
5.2 Interface agents
Macs defines interface agents as:
"[C]omputer programs that employ artificial Intelligence techniques in order to provide assistance to a user
dealing with a particular application .... The metaphor is that of a personal assistant who is collaborating
with the user in the same work environment." (Macs, 1994h, p. 71)
There are many interface agent prototype applications: for example, the New T system is a
USENET news tilter (along the lines mentioned in the second scenario that introduced this article)
(Maes, 1994a, pp. 38-39). A Ncw T agent is trained by giving it series of examples, illustrating
articles that the user would and would not choose to read. The agent then begins to make
suggestions to the user, and is given feedback on its suggestions. New T agents are not intended to
remove human choice, but to represent an extension of the human's wishes: the aim is for the agent
to be able to bring to the attention of the user articles of the type that the user has shown a
consistent interest in. Similar ideas have been proposed by Mc Gregor, who imagines prescient
agents-intelligent administrative assistants that predict our actions, and carry out routine or
repetitive administrative procedures on our behalf (Mc Gregor, 1992).
There is much related work being done by the computer supported cooperative work (CSCW)
community. CSCW is informally defined by Haecker to be "computer assisted coordinated activity
such as problem solving and communication carried out by a group of collaborating individuals"
(Haecker, 1993, p. l). The primary emphasis of CSCW is on the development of (hardware and)
software tools to support collaborative human work-the term gruupware has been coined to
describe such tools. Various authors have proposed the use of agent technology in groupware. For
example, in his participant systems proposal, Chang suggests systems in which humans collaborate
with not only other humans, but also with artificial agents (Chang, 1987). We refer the interested
reader to the collection of papers edited by Haecker (1993) and the article by Greif (1994) for more
details on CSCW,
5.3 Information agents and cooperative information systems
An information agent is an agent that has access to at least one, and potentially many information
sources, and is able to collate and manipulate information obtained from these sources to answer
queries posed by users and other information agents (the network of interoperating information
sources are often referred to as intelligent and cooperative information systems (Papazoglou et al.,
1992)). The information sources may be of many types, including, for example, traditional
databases as well as other information agents. Finding a solution to a query might involve an agent
accessing information sources over a network. A typical scenario is that of a user who has heard
about somebody at Stanford who has proposed something called agent-oriented programming,
The agent is asked to investigate, and, after a careful search of various FTP sites, returns with an
appropriate technical report, as well as the name and contact details of the researcher involved. A
number of studies have been made of information agents, including a theoretical study of how
agents are able to incorporate information from different sources (Levy et al., 1994; Gruber, 1991 ),
as well a prototype system called IRA (information retrieval agent) that is able to search for loosely

M. WOOLDRIDGE AND NICHOLAS JENNINGS 144
specified articles from a range of document repositories (Voorhees, 1994). Another important
system in this area is called Carnot (Huhns et al., 1992), which allows pre-existing and hetero
geneous database systems to work together to answer queries that are outside the scope of any of
the individual databases.
5.4 Believable agents
There is ohvious potential for marrying agent technology with that of the cinema, computer games,
and virtual reality. The Oz project6 was initiated to develop:
.. artistically interesting, highly interactive, simulated worlds . . to give users the experience of living in
(not merely watching) dramatically rich worlds that include moderately competent, emotional agents"
(Batesetal., 1992b,p. 1)
To construct such simulated worlds, one must first develop believable agents: agents that "provide
the illusion oflife, thus permitting the audience's suspension of disbelief" (Bates, 1994, p. 122). A
key component of such agents is emotion: agents should not be represented in a computer game or
animated film as the flat, featureless characters that appear in current computer games. They need
to show emotions; to act and react in a way that resonates in tune with our empathy and
understanding of human behaviour. The Oz group have investigated various architectures for
emotion (Bates et al., 1992a), and have developed at least one prototype implementation of their
ideas (Bates, 1994).
6 Concluding remarks
This paper has reviewed the main concepts and issues associated with the theory and practice of
intelligent agents. It has drawn together a very wide range of material, and has hopefully provided
an insight into what an agent is, how the notion of an agent can be formalised, how appropriate
agent architectures can be designed and implemented, how agents can be programmed, and the
types of applications for which agent-based solutions have been proposed. The subject matter of
this review is important because it is increasingly felt, both within academia and industry, that
intel!igent agents will be a key technology as computing systems become ever more distributed,
interconnected, and open. In such environments, the ability of agents to autonomously plan and
pursue their actions and goals, to cooperate, coordinate, and negotiate with others, and to respond
flexibly and intelligently to dynamic and unpredictable situations will lead to significant improve
ments in the quality and sophistication of the software systems that can be conceived and
implemented, and the application areas and problems which can be addressed.
Acknowledgements
Much of this paper was adapted from the first author's 1992 Ph D thesis (Wooldridge, 1992), and as
such this work was supported by the UK Science and Engineering Research Council (now the
EPSRC). We arc grateful to those people who read and commented on earlier drafts of this article,
and in particular to the participants of the 1994 workshop on agent theories, architectures, and
languages for their encouragement, enthusiasm, and helpful feedback. Finally, we would like to
thank the referees of this paper for their perceptive and helpful comments.
References
Adorni, G and Poggi, A, 1993. "An object-oriented language for distributed artificial intelligence" Inter
national Journal of Man-Machine Studies 38 435-453.
Agha, G, 1986. ACTORS: A Model of Concurrent Computation in Distributed Systems. MIT Press.
Agha, G, Wegner, P and Yonezawa, A (eds.), 1993. Research Directions in Concurrent Objec1-0riented
Programming. MIT Pre:.s.
6 Not to be confused with the Oz programming language (Henz ct al., 1993).

Intelligent agents: theory and practice 145
Agre, P and Chapman, D, 1987. "PENGI: An implementation of a theory of activity" In: Proceedings of the
Sixth National Conference on Artificial Intelligence (AAAI-87), pp 268-272, Seattle, WA.
Allen, JF, 1984. "Towards a general theory of action and time" Artificial Intelligence 23 (2) 123--154.
Allen, JF, Hendler, J and Tate, A (eds.), 1990. Readings in Planning. Morgan Kaufmann.
Allen, JF, Kautz, H, Pelavin, Rand Tenenberg, J, 1991. Reasoning About Plans. Morgan Kaufmann.
Ambros-Ingerson, J and Steel, S, 1988. "Integrating planning, execution and monitoring" In: Proceedings of
the Seventh National Conference on Artificial Intelligence (AAAI-88), pp 83-88, St. Paul, MN.
Austin, JL, 1962. How to Do Things With Words. Oxford University Press.
Ay!ett, Rand Eustace, D, 1994. "Multiple cooperating robots-combining planning and behaviours'' In: SM
Deen (ed) Proceedings of the 1993 Workshop on Cooperating Knowledge Based Systemv (CKBS-93), pp 3--
11. DAKE Centre, University of Keele, UK.
Haecker, RM (ed.) 1993. Readings in Groupware and Computer-Supported Cooperative Work. Morgan
Kaufmann.
Barringer, H, Fisher, M, Gabbay, D, Gough, G and Owens, R, 1989. "Metate M: A framework for
programming in temporal logic" In: REX Workshop on Stepwise Refinement of Distributed Systems:
Models, Formalisms, Correctness (LNCS Volume 430) pp 94-129. Springer-Verlag.
Barwise, J and Perry, J, 1983. Situations and Attitudes, MIT Press.
Bates, J, 1994. "The role of emotion in believable agents" Communications of the ACM 37 (7) 122-125.
Bates, J, Bryan Loyall, A and Scott Reilly, W, 1992a. "An architecture for action, emotion, and social
behaviour". Technical Report CMU-CS-92-144, School of Computer Science, Carnegie-Mellon Univer
sity, Pittsburgh, PA.
Bates, J, Bryan Loyall, A and Scott Reilly, W, 19926. "Integrating reactivity, goals, and emotion in a broad
agent". Technical Report CMU-CS-92-142, School of Computer Science, Carnegie-Mellon University,
Pittsburgh, PA.
Bell, J, 1995. "Changing attitudes". In: M Wooldridge and NR Jennings (eds.) Intelligent Agents: Theories,
Architectures, and Languages (LNAI Volume 890), pp 40--55, Springer-Verlag.
Belnap, N, 1991. ''Backwards and forwards in the modal logic of agency'' Philosophy and Phenomenological
Research LI (4) 777-807.
Belnap, N and Perloff, M, 1988. "Seeing to it that: a canonical form for agentives" Theoria 54175-199.
Bond, AH and Gasser, L (eds.) 1988. Readings in Distribwed Artificial Intelligence, Morgan Kaufmann.
Bratman, ME, 1987. Intentions, Plans, and Practical Reason, Harvard University Press.
Bratman, ME, 1990. "What is intention?" In: PR Cohen, JL Morgan and ME Pollack (eds.) Intentions in
Communication, pp 15-32, MIT Press.
Bratman, ME, Israel, DJ and Pollack, ME, 1988. "Plans and resource-bounded practical reasoning''
Computational Intelligence 4 349-355.
Brooks, RA, 1986. "A robust layered control system for a mobile robot" IEEE Journal of Robotics and
Aulomation 2 (1) 14-23.
Brooks, RA, 1990. "Elephants don't play chess" In: P Maes (ed.) Designing Aulonomous Agents, pp 3-15,
MIT Press.
Brooks, RA, 1991a. "Intelligence without reason" In: Proceedings of the Twelflh International Joint
Conference on Artificial Intelligence (JJCAl-91), pp 569-595, Sydney, Australia.
Brooks, RA, 19916. "Intelligence without representation" Artificial Intelligence 47 139-159.
Burmeister, Band Sundermeyer, K. 1992. ''Cooperative problem solving guided by intentions and percep
tion" In: E Werner and Y Demazeau (eds) Decentralized Al 3-Proceedings of the Third European
Workshop on Modelling Autonomous Agents and Multi-Agent Worlds (MAAMAW-91), pp 77-92,
Elsevier.
Bussman, Sand Demazeau, Y, 1994. "An agent model combining reactive and cognitive capabilities" In:
Proceedings of the IEEE International Conference on Intelligent Robots and Systems (IROS-94), Munich,
Germany.
Caste!franchi, C, 1990. "Social power" In: Y Demazeau and J-P MU!ler(eds.) Decentralized Al-Proceedings
of the First European Workshop on Modelling Autonomous Agents in Multi-Agent Worlds (MAAMA W8Y), pp 49--62, Elsevier.
Castelfranchi, C, 1995. "Guarantees for autonomy in cognitive agent architecture" In: M Wooldridge and NR
Jennings (eds.) Intelligent Agents. Theories, Architectures, and Languages (LNAI Volume 890), pp 56--70,
Springer-Verlag.
Castelfranchi, C, Miceli, Mand Cesta, A, 1992. '·Dcpcndcncc relations among autonomous agents" In: E
Werner and 'Y Dema:ceau (eds.) Decentralized Al 3-Proceedings of the Third European Workshop on
Mode Uing Awonomous Agents and Multi-Agent Worlds (MAAMA W-Yl), pp 215-231, Elsevier.
Catach, L, 1988. "Norma! multimodal logics"' In: Proceedings of the Seventh National Conference on Anificial
Intelligence (AAAJ-88), pp 491-495. St. Paul, MN.

M. WOOLDRIDGE AND NICHOLAS JENNINGS 146
Chaib-draa, B, Moulin, B, Mandiau, Rand Millot, P, 1992. "Trends in distributed artificial intelligence··
Artificial Intelligence Review 6 35-66.
Chang, E, 1987. "Participant systems" In: M Huhns (ed.) Distributed Artificial Intelligence, pp 311-340,
Pitman.
Chapman, D, 1987. "Planning for conjunctive goals" Artificial Intelligence 32 333--378.
Chapman, D and Agre, P, 1986. "Abstract reasoning as emergent from concrete activity" In: MP Georgeff
and AL Lansky (eds.) Reasoning About Actions & Plans~Proceedings of the 1986 Workshop pp 411-424,
Morgan Kaufmann.
Chel\as, B, 1980. Modal Logic: An Introduction, Cambridge University Press.
Chu, D, 1993. "l.C. PROLOG JI: A language for implementing multi-agent systems" In: SM Deen (ed.)
Proceedings of the 1992 Workshop on Cooperating Knowledge Based Systems (CKBS-92), pp 61-74,
DAKE Centre, University of Keele, UK.
Cohen, PR, Greenberg ML, Hart OM and Howe AE. 1989. "Trial by fire: Understanding the design
requirements for agents in complex environments" Al Magazine 10 (3) 32-48.
Cohen, PR and Levesque, HJ, 1990a. "'Intention is choice with commitment" Artificial Intelligence 42 213261.
Cohen, PR and Levesque, HJ, 1990h. ·'Rational interaction as the basis for communication" In: PR Cohen, J
Morgan and ME Pollack (eds.) Intentions in Communication, pp 221-256. MIT Press.
Cohen, PR and Perrault, CR, 1979. ·'Elements of a plan based theory of speech acts" Cognitive Science 3177212.
Connah, D and Wavish. P, 1990. '·An experiment in cooperation" In: Y Demazeau and J-P Miiller (eds.)
Decentralized AI-Proceedings of the First European Workshop on Modelling Autonomous Agents in
Multi-Agent Worlds (MAAMA W-89), pp 197-214, Elsevier.
Cutkosky, MR, Engelmorc, RS, Fikes. RE, Gcnesereth. MR, Gruber, T, Mark, WS, Tenenbaum, JM and
Weber, JC, 1993. "PACT: An experiment in integrating concurrent engineering systems'' IEEE Computer
26 (l) 28-37.
Davies, NJ, 1993. Truth, Modality, and Action, Ph D thesis, Department of Computer Science, University of
Essex, Colchester, UK.
Dean, TL and Wellman, MP. 1991. Planning and Control, Morgan Kaufmann.
Dennett, DC, 1978. Brainstorm~, MIT Press.
Dennett, DC. 1987. The Intentional Stance, MIT Press.
des Rivieres. J and Levesque. HJ, 1986. ''The consistency of ~yntactical treatments of knowledge" In: JY
Halpern ( ed,) Proceedings of the 1986 Conference on Theoretical Aspects of Reasoning About Knowledge,
pp 115-130, Morgan Kaufmann.
Devlin, K, 1991. Logic and Information, Cambridge University Press.
Dongha, P, 1995 "Toward a formal model of commitment for rc~ource-bounded agent~" In: M Wooldridge
and NR Jennings (eds.) Intelligent Agents: Theories, Architectures, and Languages ( LNAJ Volume 890), pp
86--101. Springer-Verlag.
Down~, J and Rcichgelt, II, 1991. "Integrating classical and rcm.:tive planning within an architecture for
autonomous agents·· In: J Hertzberg (ed,) European Worhhop on Planning (LNAI Volume 522), pp 13--
26.
Doyle. J, Shoham, Y and Wellman, MP, 1991. ''A logic of relative desire" In: ZW Ras and M Zemankova
(cd~.) Methodologies for Intelligent Systems-Sixth International Symposium, ISMIS-91 (LNAI Volume
542). Springer-Verlag.
Emerson EA, 1990. "Temporal and modal logic" In: J van Leeuwen (ed.) Handhookof Throretical Computer
Science. pp 996--1072, Elsevier.
Emerson, EA and Halpern, JY, 1986. "'Sometimes' and 'not never' revisited: on branching time versus linear
time temporal logic" Journal of the ACM 33 (1) 151-178.
Etzioni, 0. Lesh, N and Segal, R, 1994. ·'Building soft bots for UNIX'' In: 0 Etzioni (ed.) Software Agents
Paper~ from the 1994 Spring Symposium (Technical Report SS-94-03), pp 9-16, AAAI Press.
Fagin, Rand Halpern, JY, 1985. "Belief, awarenes~. and limited reasoning" In: Proceedings of the Ninth
International Joint Conference on Artificial Intelligence (IJCAl-85), pp 480--490. Los Angeles, CA.
Fagin. R, Halpern, JY and Yard!, MY, 1992. "What can machines know? on the properties of knowledge in
distributed system~·· Journal of rhe A CM 39 (2) 328--376.
Ferguson IA. 1992a. Touring Machines: An Architecture for Dynamic, Rational, Mobile Agents, Ph D thesis,
Clare Hall, University of Cambridge, UK. (Also available as Technical Report No. 273, University of
Cambridge Computer Laboratory.)
Ferguson. IA, 1992h. ·Towar<ls an architecture for adaptive, rational, mobile agents·• In: E Werner and Y
Demazeau (eds.) Decentralized Al 3-Proceedings of the Third European Workshop on Modelling
Autonomous Agents and Multi-Agent Worlds (MAAMA W-91), pp 249-262, Elsevier.

Intelligent agents: theory and practice 147
Fikes, RE and Nilsson. N, 1971. "STRIPS: A new approach to the application of theorem proving to problem
solving'· Artificial Intelligence 5 (2) 189-208.
Firby, JA. 1987. "An investigation into reactive planning in complex domains"' In: Proceedings of the Tenth
International Joint Conference on Artificial Intelligence (IJCA/-87), pp 202-206, Milan, Italy.
Fischer, K, Kuhn. N, MUiler, HJ, MUiler, JP and Pischel, M, 1993. ·'Sophisticated and distributed: The
transportation domain" In: Proceedings of the Fifth European Workshop on Modelling Autonomous
Agents and Multi-Agent Worlds (MAAMA W-93), Neuchatel, Switzerland.
Fisher, M, 1994. "A survey of Concurrcnt Metate M-the language and its applications"" In: DM Gabbay and
HJ Ohlbach (eds.) Temporal Logic-Proceedings of the First International Conference (LNAJ Volume
827). pp 480-505, Springer-Verlag.
Fisher, M. 1995. "Representing and executing agent-based systems·· In: M Wooldridge and NR Jennings
(eds.) Intelligent Agents: Theories, Architectures, and Languages (LNAJ Volume 890), pp 307-323,
Springer-Verlag.
Fisher, M and Wooldridge, M, 1993. "Specifying and verifying distributed intelligent systems" In: M
Filgueiras and L Damas (eds.) Progress in Artificial Intelligence-Sixth Portuguese Conference on Anificial
Intelligence (LNAI Volume 727), pp 13-28. Springer-Verlag.
Galliers. JR, 1988a. "A strategic framework for multi-agent cooperative dialogue" In: Proceedings of the
Eighth European Conference on Artificial Intelligence (ECAl-88), pp 415-420, Munich, Germany.
Galliers, JR, 1988b. A Theoretical Framework for Computer Models of Cooperative Dialogue, Acknowledg
ing Multi-Agent Conflict. Ph D thesis, Open University, UK.
Gasser, L, 1991. "Social conceptions of knowledge and action: DAI foundations and open systems semantics"
Artificial Intelligence 47 107-138.
Gasser, L, Braganza, C and Hermann, N, 1987. "MACE: A flexible testbed for distributed AI research" In:
M Huhns (ed.) Distributed Artificial Intelligence, pp 119-152, Pitman.
Gasser, Land Briot, JP, 1992. "Object-based concurrent programming and DAI" In: Distributed Artificial
Intelligence: Theory and Praxis. pp 81-108, Kluwer Academic.
Geissler, C and Konolige. K, 1986. "A resolution method for quantified modal logics of knowledge and
helief'· In: JY Halpern (ed.) Proceedings of the 1986 Conference on Theoretical Aspects of Reasoning About
Knowledge, pp 309-324, Morgan Kaufmann.
Genescreth, MR and Ketchpe L SP, 1994. "Software agents" Communications of the ACM 37 (7) 48-53.
Genesereth. MR and Nils~on, N, 1987. Logical Foundations of Artificial Intelligence, Morgan Kaufmann.
Gcorgeff, MP, 1987. "Planning" Annual Review of Computer Science 2 359-400.
Georgeff, MP and lngrand, FF, 1989. "Decision-making in an embedded reasoning system" In: Proceedings
of the Eleventh International Joint Conference on Artificial Intelligence ( IJCAl-89), pp 972-978, Detroit,
ML
Gcorgcff. MP and Lansky, AL (eds.) 1986. Reasoning About Actions & Plans-Proceedings of the 1986
Workshop, Morgan Kaufmann.
Georgeff, MP and Lansky, AL, 1987. "Reactive reasoning and planning In: Proceedings of the Sixth National
Conference on Artificial Intelligence (AAAJ-87), pp 677-682, Seattle, WA.
Ginsberg, M. 1993. Essentials of Artificial Intelligence, Morgan Kaufmann.
Gmytrasiewicz, P and Durfee. EH, 1993. "Elements of a utilitarian theory of knowledge and action" In:
Proceedings of the Thirteenth International Joint Conference on Artificial Intelligence (IJCA/-93), pp 396--
402, Chamb Cry. France.
Goldblatt, R. 1987. Logics of Time and Computation, Centre for the Study of Language and lnformation
Lccturc Notes Series. (Distributed by Chicago University Press.)
Goldman, RP and Lang. RR, 1991 "Intentions in time"', Technical Report TUTR 93-101, Tulane University.
Goodwin, R. 1993. "Formalizing properties of agents", Technical Report CMU-CS-93-159, School of
Computer Science. Carnegie-Mellon University, Pittsburgh, PA.
Greif. I, 1994. "Desktop agent~ in group-enabled products" Communications of the ACM 37 (7) 100-105.
Grosz, BJ and Sidner, CL, 1990. ''Plans for discourse" In: PR Cohen, J Morgan and ME Pollack (eds.)
Intentions in Communication. pp 417-444, MIT Press.
Gruber. TR, 1991. "The role of common ontology in achieving sharable, reusable knowledge bases'· In: R
Fikes and E Sandcwall (eds.) Proceedings of Knowledge Representation and Reasoning (KR&R-91),
Morgan Kaufmann.
Guha, RV and Lcnat, DB, 1994. "Enabling agents to work together" Communications of the ACM 37 (7) 127142.
Haas, A, 1986. "A syntactic theory of belief and knowledge" Artificial Intelligence 28 (3) 245-292.
Haddadi, A. 1994. "A hybrid architecture for multi-agent systems" In: SM Deen (ed.) Proceedings of the 1993
Workshop 011 Cooperating Knowledge Based Systems (CKBS-93), pp 13-26. DAKE Centre, University of
Keele, UK.

M. WOOLDRIDGE AND NICHOLAS JENNINGS 148
Halpern, JY, 1986. "Reasoning about knowledge: An overview" In: JY Halpern (ed.) Proceedings oft he 1986
Conference on Theoretical Aspects of Reasoning About Knowledge, pp 1-18, Morgan Kaufmann.
Halpern, JY, 1987. "Using reasoning about knowledge to analyze distributed systems" Annual Review of
Computer Science 2 37--68.
Halpern, JY and Moses, Y, 1992. "A guide to completeness and complexity for modal logics of knowledge and
belief' Artificial Intelligence 54 319-379.
Halpern, JY and Vardi, MY, 1989. "The complexity of reasoning about knowledge and time. I. Lower
bounds" Journal of Computer and System Sciences 38 195-237.
Hare!, D, 1984. "Dynamic logic" In: D Gabbay and F Guenther (eds.) Handbook of Philosophical Logic
Volume ll-Extensions of Classical Logic, pp 497-604, Reidel.
Haugeneder, H, 1994. IMAGINE final project report.
Haugeneder, Hand Steiner, D, 1994. "A multi-agent approach to cooperation in urban traffic" In: SM Deen
(ed.) Proceedings of the 1993 Workshop on Cooperating Knowledge Based Systems (CKBS-93), pp 83-98,
DAKE Centre, University of Keele, UK.
Haugeneder, H, Steiner, D and Mc Cabe, FG, 1994. "IMAGINE: A framework for building multi-agent
systems" In: SM Deen (ed.) Proceedings of the 1994 Jnternational Working Conference on Cooperating
Knowledge Based Systems (CKBS-94), pp 31--64, DAKE Centre, University of Keele, UK.
Hayes-Roth, B, 1990. "Architectural foundations for real-time performance in intelligent agents" The
Journal of Real-Time Systems 2 99-125.
Hendler, J (ed.) 1992. Artificial intelligence Planning: Proceedings of the First International Conference,
Morgan Kaufmann.
Henz, M, Smolka, G and Wuertz, J, 1993. "Oz-a programming language for multi-agent systems" ln:
Proceedings of the Thirteenth international Joint Conference on Artificial Intelligence ( IJCAI-93), pp 404409, Chamb Cry, France.
Hewitt, C, 1977. "Viewing control structures as patterns of passing messages" Artificial intelligence 8 (3) 323364.
Hintikka, J, 1962. Knowledge and Belief, Cornell University Press.
Houlder, V, 1994. "Special agents" In: Financial Times, 15 August, p 12.
Huang, J, Jennings, NR and Fox. J, 1995. "An ag-ent architecture for distributed medical care" In: M
Wooldridge and NR Jennings (eds.) Intelligent Agents: Theories, Architectures, and Languages (LNA!
Volume 890), pp 219-232, Springer-Verlag.
Hughes, GE and Cresswell, MJ, 1968. lntroduction to Modal Logic, Methuen.
Huhns, MN, Jacobs, N, Ksiezyk, T, Shen, WM, Singh, MP and Cannata, PE, 1992. "Integrating enterprise
information models in Carnot" In: Proceedings of the International Conference on Intelligent and
Cooperative information Systems, pp 32-42, Rotterdam, The Netherlands.
Israel, DJ, 1993. "The role(s) of logic in artificial intelligence" In: DM Gabbay, CJ Hogger and JA Robinson
(eds.) Handbook of Logic in Artificial Intelligence and Logic Programming, pp 1-29, Oxford University
Press.
Jennings, NR, 1992. "On being responsible" In: E Werner and Y Demazeau (eds.) Decentralized Al]
Proceedings of the Third European Workshop on Modelling Autonomous Agents and Multi-Agent Worlds
(MAAMA W-91), pp 93-102, Elsevier.
Jennings. NR, 1993a. "Commitments and conventions: The foundation of coordination in multi-agent
systems" Knowledge Engineering Review 8 (3) 223-250.
Jennings, NR, 1993b. "Specification and implementation of a belief desire joint-intention architecture for
collaborative problem solving" Journal of intelligent and Cooperative Information Systems 2 (3) 289-318.
Jennings, NR, 1995. "Controlling cooperative problem solving in industrial multi-agent systems using joint
intentions" Artificial Intelligence 14 (2) (to appear).
Jennings, NR. Varga, LZ, Aarnts, RP, Fuchs, J and Skarek, P, 1993. "Transforming standalone expert
systems into a community of cooperating agents'' International Journal of Engineering Applications of
Artificial Intelligence 6 (4 ) 317-331.
Kaelbling, LP, 1986. "An architecture for intelligent reactive systems" In: MP George ff and AL Lansky (eds.)
Reasoning About Actions and Plans-Proceedingofthe 1986 Workshop, pp 395-410, Morgan Kaufmann.
Kaelbling, LP, 1991. ·'A situated automata approach to the design of embedded agents" SIG ART Bulletin 2
(4) 85-88.
Kaelbling, LP and Roscnschein, SJ, 1990. "Action and planning in embedded agents" In: P Maes (ed.)
Designing Autonomous Agents, pp 35-48, MIT Press.
Kinny, D, Ljungberg, M, Rao, AS, Sonenberg, E, Tidhar, G and Werner, E, 1992. "Planned team activity"
In: C Castelfranchi and E Werner (eds.) Artificial Social Systems-Selected Papers from the Fourth
European Workshop on Modelling Autonomous Agents and Multi-Agent Worlds, MAAAMA W-92 (LNA!
Volume 830), pp 226-256, Springer-Verlag.

Intelligent agents: theory and practice 149
Kiss, G and Reichgelt, H, 1992. "Towards a semantics of desires" In: E Werner and Y Demazeau (eds.)
Decentralized Al 3-Proceedings of the Third European Workshop on Modelling Autonomous Agents and
Multi-Agent Worlds (MAAMAW-91), pp 115-128, Elsevier.
Konolige, K, 1982. "A first-order formalization of knowledge and action for a multi-agent planning system"
In: JE Hayes, D Michie and Y Pao (eds.) Machine Intelligence JO, pp 41-72, Ellis Horwood.
Konolige, K, 1986a. A Deduction Model of Belief, Pitman.
Konolige, K, 1986b. "What awareness isn't: A sentential view of implicit and explicit belief (position paper)"
In: JY Halpern (ed.) Proceedings of the 1986 Conference on Theoretical Aspects of Reasoning About
Knowledge, pp 241-250, Morgan Kaufmann.
Konolige, Kand Pollack, ME, 1993. "A representationalist theory of intention" In: Proceedings of the
Thirteenth International Joint Conference on Artificial Intelligence (IJCAJ-93), pp 390--395, Chambefy,
France.
Kraus, Sand Lehmann, D (1988) "Knowledge, belief and time" Theoretical Computer Science 58 155-174.
Kripke, S, 1963. "Semantical analysis of modal logic" Zeitschrift f Ur Mathematische Logik und Grundlagen
der Mathematik 9 67-96.
Lakemeyer, G, 1991. "A computationally attractive first-order logic of belief" In: JELIA-90: Proceedings of
the European Workshop on Logics in AI (LNAI Volume 478), pp 333-347, Springer-Verlag.
Lesperance, Y, 1989. "A formal account of self knowledge and action" In: Proceedings of the Eleventh
Imernational Joint Conference on Artificial Intelligence (IJCAI-89), pp 868-874, Detroit, Ml.
Levesque, HJ, 1984. "A logic of implicit and explicit belief" In: Proceedings oft he Fourth National Conference
on Artificial Intelligence (AAAI-84), pp 198-202, Austin, TX.
Levesque, HJ, Cohen, PR and Nunes, JHT, 1990. "On acting together" In: Proceedings of the Eighth National
Conference on Artificial Intelligence (AAAJ-90), pp 94-99, Boston, MA.
Levy, A Y, Sagiv, Y and Srivastava, D, 1994. "Towards efficient information gathering agents" In: 0 Etzioni
(ed.) Software Agents-Papers from the 1994 Spring Symposium (Technical Report SS-94-03), pp 64-70,
AAAI Press.
Mack, D, 1994. "A new formal model of belief" In: Proceedings of the Eleventh European Conference on
Artificial Intelligence (ECAI-94), pp 573-577, Amsterdam, The Netherlands.
Maes, P, 1989. "The dynamics of action selection" In: Proceedings of the Eleventh International Joint
Conference on Artificial Intelligence (IJCAI-89), pp 991-997, Detroit, MI.
Maes, P (ed.) 1990a. Designing Autonomous Agents, MIT Press.
Maes, P, 1990b. "Situated agents can have goals" In: P Maes (ed.) Designing Autonomous Agents, pp49-70,
MIT Press.
Maes, P, 1991. "The agent network architecture (ANA)" SIGA RT Bulletin 2 (4) 115-120.
Maes, P, 1994a. "Agents that reduce work and information overload" Communications of the ACM 37 (7) 3140.
Macs, P, 1994b. "Social interface agents: Acquiring competence by learning from users and other agents" In:
0 Etzioni (ed.) Software Agents-Papers from the J 994 Spring Symposium (Technical Report SS-94-03), pp
71-78, AAA! Press.
Mc Cabe, FG and Clark, KL, 1995. "April-agent process interaction language" In: M Wooldridge and NR
Jennings (eds.) Intelligent Agents: Theories, Architectures, and Languages (LNAI Volume 890), pp 324340, Springer-Verlag.
Mc Carthy, J, 1978. "Ascribing mental qualities to machines." Technical report, Stanford University Al Lab.,
Stanford, CA 94305.
Mc Gregor, SL, 1992. "Prescient agents" In: D Coleman (ed.) Proceedings of Groupware-92, pp 228-230.
Montague, R, 1963. "Syntactical treatments of modality, with corollaries on reflexion principles and finite
axiomatizations" Acta Philosophica Fennica 16153-167.
Moore, RC, 1990. '"A formal theory of knowledge and action" In: JF Allen, J Hendler and A Tate (eds.)
Readings in Planning, pp 480-519, Morgan Kaufmann.
Morgenstern, L, 1987. "Knowledge preconditions for actions and plans" In: Proceedings of the Tenth
International Joint Conference on Artificial Intelligence (JJCAI-87), pp 867-874, Milan, Italy.
Mori, K, Torikoshi, H, Nakai, Kand Masuda, T, 1988. "Computer control system for iron and steel plants"
Hitachi Review 37 (4) 251-258.
Morley, RE and Schelberg, C, 1993. "An analysis of a plant-specific dynamic scheduler'· In: Proceedings of the
NSF Workshop on Dynamic Scheduling, Cocoa Beach, Florida.
Mukhopadhyay, U, Stephens, Land Huhns, M, 1986. "An intelligent system for document retrieval in
distributed office environments'' Journal of the American Society for Information Science 37 123-135.
Millier, JP, 1994. "A conceptual model for agent interaction" In: SM Deen (ed.) Proceedings of the Second
International Working Conference on Cooperating Knowledge Based Systems (CKBS-94), pp 213-234,
DAKE Centre, University of Keele, UK.

M. WOOLDRIDGE AND N[CHOLAS JENNINGS 150
MUiler, JP and Pischel, M, 1994. "Modelling interacting agents in dynamic environments .. In: Proceedings of
the Eleventh European Conference on Artificial Intelligence (ECAI-94), pp 709-713, Amsterdam, The
Netherlands.
Mi.i\ler, JP, Pischcl, M and Thiel, M, 1995. "Modelling reactive behaviour in vertically layered agent
architectures" In: M Wooldridge and NR Jennings (eds.) Intelligent Agents: Theories, Architectures, and
Languages (LNAI Volume 890), pp 261-276, Springer-Verlag.
Newell, A and Simon, HA, 1976. "Computer science as empirical enquiry" Communications of the ACM 19
113-126.
Nilsson, NJ, 1992. "Towards agent programs with circuit semantics", Technical Report STAN-CS-92-1412,
Computer Science Department, Stanford University, Stanford, CA 94305.
Norman, TJ and Long, D, 1995. "Goal creation in motivated agents" In: M Wooldridge and NR Jennings
(eds.) Intelligent Agents: Theories, Architectures, and Languages (LNAI Volume 890), pp 277-290,
Springer-Verlag.
Papazoglou, MP, Laufman, SC and Sel\is, TK, 1992. "An organizational framework for cooperating
intelligent information systems" Journal of Intelligent and Cooperative Information Systems 1 (1) 169-202.
Parunak, HVD, 1995. "Applications of distributed artificial intelligence in industry'" In: GMP O'Hare and
NR Jennings (eds.) Foundations of Distributed Al, John Wiley.
Patil, RS, Fikes, RE, Patel-Schneider, PF, Mc Kay, D, Finin, T, Gruber, T and Neches, R, 1992. "The
DARPA knowledge sharing effort: Progress report" In: C Rich, W Swartout and B Nebel (eds.)
Proceedings of Knowledge Representation and Reasoning (KR&R-92), pp 777-788.
Perlis, D, 1985. "Languages with self reference I: Foundations" Artificial Intelligence 25 301-322.
Perlis, D, 1988. "Languages with self reference II: Knowledge, belief, and modality" Artificial Intelligence 34
179-212.
Perloff, M, 1991. "STIT and the language of agency" Synthese 86 379-408.
Poggi, A, 1995. "DAISY: An object-oriented system for distributed artificial intelligence" In· M Wooldridge
and NRJennings (eds.) Intelligent Agents: Theories, Architectures, and Languages (LNA!Volume890). pp
341-354, Springer-Verlag.
Pollack, ME and Ringuette, M, 1990. "Introducing the Tileworid: Experimentally evaluating agent architec
tures" In: Proceedings of the Eighth National Conference on Artificial Intelligence (AAAl-90), pp 183-189,
Boston, MA.
Rao, AS and Georgeff, MP, 1991a. '"Asymmetry thesis and side-effect problems in linear time and branching
time intention logics'" In: Proceedings oft he Twelfth International Joint Conference on Artificial Intelligence
(IJCAl-91), pp 498-504, Sydney, Australia.
Rao, AS and Georgeff, MP, 1991b. "Modeling rational agents within a BDI-architccture·• In: R Fikes and E
Sandewal! (eds.) Proceedings of Knowledge Representation and Reasoning (KR&R-9/), pp 473-484,
Morgan Kaufmann.
Rao, AS and Georgeff, MP, 1992a. "An abstract architecture for rational agents"' In: C Rich, W Swartout and
B Nebel (eds.) Proceedings of Knowledge Representation and Reasoning (KR&R-92), pp 439-449.
Rao, AS and Georgeff, MP, 1992b. "Social plans: Preliminary report" In: E Werner and Y Demazeau (eds.)
Decentralized Al 3-Proceedings of the Third European Workshop on Modelling Autonomous Agents and
Multi-Agent Worlds (MAAMA W-9I), pp 57-76, Elsevier.
Rao, AS and Georgeff, MP, 1993. "A model-theoretic approach to the verification of situated reasoning
systems" In: Proceedings of the Thirteenth International Joint Conference on Artificial Intelligence (IJCAI93), pp 318-324, Chambery, France.
Reichgclt, H, 1989a "A comparison of first-order and modal logics of time" In: P Jackson, H Reichge!t and F
van Harmclen (eds.) Logic Based Knowledge Representation, pp 143-176, MIT Press.
Reichgelt. H, 198%. "Logics for reasoning about knowledge and belief" Knowledge Engineering Review4 (2)
119-139.
Rosenschein, JS and Genesereth, MR, 1985 "Deals among rational agents·• In: Proceedings of the Ninth
International Joint Conference on Anificial Intelligence (Il CAl-85), pp 91-99, Los Angeles, CA
Rosenschein, S, 1985. "Formal theories of knowledge in AI and robotics'' New Generation Computing, pp
345-357.
Rosenschein, S and Kae\bling, LP, 1986. '·Toe synthesis of digital machines with provable epistemic
properties" In: JY Halpern (ed.) Proceedings of the 1986 Conference on Theoretical Aspects of Reasoning
About Knowledge, pp 83-98, Morgan Kaufmann.
Russell, SJ and Wefald, E, 1991 Do the Right Thing-Studies in Limited Rationality. MIT Press.
Sacerdoti. E, 1974. "Planning in a hierarchy of abstraction spaces" Artificial Intelligence 5 115-135.
Sacerdoti, E, 1975. '"The non-linear nature of plans'' In: Proceedings of the Fourth International Joint
Conference on Artificial Intelligence ( Il CAl-75), pp 206--214. Stanford, CA.
Sadek. MD, 1992. '·A study in the logic of intention" In: C Rich, W Swartout and B Nebel (eds.) Proceedings
of Knowledge Representation and Reasoning (KR&R-92). pp 462-473.

Intelligent agents: theory and practice 151
Sargent, P, 1992. "Back to school for a brand new ABC" In: The Guardian, 12 March, p 28.
Schoppe N;, MJ, 1987. "Universal plans for reactive robots in unpredictable environments" In: Proceedings of
the Tenth International Joint Conference on Artificial Intelligence ( /JCAl-87), pp 1039-1046, Milan, Italy.
Schwuttke, UM and Quan, AG, 1993. '·Enhancing performance of cooperating agents in real-time diagnostic
systems" In: Proceedings of the Thirteenth International Joint Conference on Artificial Intelligence (JJCAl93), pp 332-337, Chambfay, France.
Searle, JR, 1969. Speech Acts: An Essay in the Philosophy of Language, Cambridge University Press.
Seel, N, 1989. Agent Theories and Architectures, Ph D thesis, Surrey University, Guildford, UK.
Segerbcrg, K, 1989. "Bringing it about" Journal of Philosophical Logic 18 327-347.
Shardlow, N, 1990. "Action and agency in cognitive science", Master's thesis, Department of Psychology,
University of Manchester, Oxford Road, Manchester M13 9PL, UK.
Shoham, Y, 1988. Reasoning About Change: Time and Causation from the Standpoint of A rtificial Intelligence,
MIT Press.
Shoham, Y, 1989. "Time for action: on the relation between time, knowledge and action" In: Proceedings of
the Eleventh /nternational Joint Conferenceon Artificial Intelligence (l JCAl-89), pp 954-959, Detroit, Ml.
Shoham, Y, 1990. "Agent-oriented programming", Technical Report STAN-CS-1335-90, Computer Science
Department, Stanford University, Stanford, CA 94305.
Shoham, Y, 1993. "Agent-oriented programming" Artificial Intelligence 60 (1) 51-92.
Singh, MP, l990a. '·Group intentions" In: Proceedings of the Tenth International Workshop on Distributed
Artificial Intelligence (IWDAI-90).
Singh, MP. 1990b. "Towards a theory of situated know-how" In: Proceedings of the Ninth European
Conference on Artificial Intelligence (ECAJ-90), pp 604-609, Stockholm, Sweden.
Singh, MP, 1991a. "Group ability and structure" In: Y Dcmazeau and JP MUiler (eds.) Decentralized Al 2Proceedings of the Second European Workshop on Modelling Autonomous Agents and Multi-Agent Worlds
(MAAMAW-90), pp 127-146, Elsevier.
Singh, MP, 1991b '·Towards a formal theory of communication for multi-agent systems" In: Proceedings of rhe
Twelfth International Joint Conference on Artificial Intelligence (l JCAI-91), pp 69-74, Sydney, Australia.
Singh, MP, 1992. "A critical examination of the Cohen-Levesque theory of intention" In: Proceedings of the
Tenth European Conference on Artificial Intelligence (ECAl-92), pp 364-368. Vienna, Austria.
Singh, MP, 1994. Multiagent Systems: A Theoretical Framework for Intentions, Know-How, and Communi
cations (LNAI Volume 799), Springer-Verlag.
Singh, MP and Asher, NM, 1991. "Towards a formal theory of intentions" In: Logics in Al-Proceedings of
the European Workshop JELIA-90 (LNA! Volume 478), pp 472-486, Springer-Verlag.
Smith, RG, 1980. A Framework for Distributed Problem Solving, UMI Research Press.
Steeb, R, Cammarata S, Hayes-Roth FA, Thorndyke PW and Wesson RB, 1988. "Distributed intelligence for
air fleet control" In: AH Bond and L Gasser (eds.) Readings in Distributed Artificial Intelligence, pp 90101, Morgan Kaufmann.
Steels, L, 1990. "Cooperation between distributed agents through self organization" In: Y Demazeau and JP
MUiler (eds.) Decentralized Al-Proceedings of the First European Workshop on Modelling Autonomous
Agents in Multi-Agent Worlds (MAAMA W-89), pp 175-196, Elsevier.
Thomas, SR, 1993. PLACA, an Agent Oriented Programming Language, Ph D thesis, Computer Science
Department, Stanford University, Stanford, CA 94305. (Available as technical report STAN-CS-931487).
Thoma~, SR, Shoham Y, Schwartz A and Kraus S, 1991. "Preliminary thoughts on an agent description
language" International Journal of Intelligent Systems 6 497-508.
Thomason, R, 1980. "A note on syntactical treatments of modality" Synthese 44 391-395.
Turner, R, 1990. Truth and Modality for Knowledge Representation, Pitman.
Varga, LZ, Jennings, NR and Cockburn, D, 1994. "Integrating intelligent systems into a cooperating
community for electricity distribution management" International Journal of Expert Systems with Appli
cations 1 (4) 563-579.
Vere, Sand Bickmore, T, 1990. "A basic agent" Computational Intelligence 6 41-60.
Voorhees, EM, 1994. •'Software agents for information retrieval" In: 0 Etzioni (ed.) Software Agents
Papers from the 1994 Spring Symposium (Technical Report SS-94-03), pp 126--129, AAAI Press.
Wainer, J, 1994. "Yet another semantics of goals and goal priorities" In: Proceedings of the Eleventh
European Conference on Artificial Intelligence (ECIA-94), pp 269-273, Amsterdam, The Netherlands.
Wavish, P, 1992. "Exploiting emergent behaviour in multi-agent systems" In: E Werner and Y Demazeau
(eds.) Decentralized Al 3-Proceedings of the Third European Workshop on Modelling Autonomous
Agents and Multi-Agent Worlds (MAAMA W-91), pp 297-310, Elsevier.
Wavish, P and Graham, M, 1995. "Role, skills, and behaviour: a situated action approach to organising
systems of interacting agents" In: M Wooldridge and NR Jennings (eds.) Intelligent Agents: Theories,
Archi1ectures, and Languages (LNAI Volume 890), pp 371-385, Springer-Verlag.

M. WOOLDRIDGE AND NICHOLAS JENNINGS 152
Weerasooriya, D, Rao, A and Ramamohanarao, K, 1995. "Design of a concurrent agent-oriented language"
In: M Wooldridge and NR Jennings (eds.) Intelligent Agents: Theories, Architectures, and Languages
(LNAJ Volume 890), pp 386----402, Springer-Verlag.
Weihmayer, Rand Velthuijsen, H, 1994. "Application of distributed AI and cooperative problem solving to
telecommunications" In: J Liebowitz and D Prcreau (eds.) Al Approaches to Telecommunications and
Network Management, IOS Press.
Werner, E, 1988. "Toward a theory of communication and cooperation for multiagent planning" In: MY
Var di (ed.) Proceedings of the Second Conference on Theoretical Aspects of Reasoning About Knowledge,
pp 129-144, Morgan Kaufmann.
Werner, E, 1989. "Cooperating agents: A unified theory of communication and social structure" lq: L Gasser
and M Huhns (eds.) Distributed Artificial Intelligence Volume 11, pp 3---36, Pitman.
Werner, E, 1990. "Wh.at·can agents do together: A semantics of cooperative ability" In: Proceedings of the
Ninth European Conference on Artificial Intelligence (ECAJ-90), pp 694-701, Stockholm, Sweden.
Werner, E, 1991. "A unified view of information, intention and ability" In: Y Demazeau and JP Miiller (eds.)
Decentralized Al 2-Proceedings of the Second European Workshop on Modelling Autonomous Agents
and Multi-Agent Worlds (MAAMA W-90), pp 109-126, Elsevier.
White, JE, 1994. "Telescript technology: The foundation for the electronic marketplace'', White paper,
General Magic, Inc., 2465 Latham Street, Mountain View, CA 94040.
Wilkins, D, 1988. Practical Planning: Extending the Classical Al Planning Paradigm, Morgan Kaufmann.
Wittig, T (ed.) 1992. ARCHO N: An Architecture for Multi-Agent Systems, Ellis Horwood.
Wood, S, 1993. Planning and Decision Making in Dynamic Domains, Ellis Horwood.
Wooldridge, M, 1992. The Logical Modelling of Computational Multi-Agent Systems, Ph D thesis, Depart
ment of Computation, UMIST, Manchester, UK. (Also available as Technical Report MMU-DOC-94-01,
Department of Computing, Manchester Metropolitan University, Chester Street, Manchester, UK.)
Wooldridge, M, 1994. ·"Coherent social action" In: Proceedings of the Ele J.'enth European Conference on
Artificial Intelligence (ECAI-94), pp 279-283, Amsterdam, The Netherlands.
Wooldridge, M, 1995. "This is MYWORLD: The logic of an agent-oriented testbed for DAI" In: M
Wooldridge and NR Jennings (eds.) Intelligent Agents: Theories, Architectures, and Languages (LNAI
Volume 890), pp 160-178, Springer-Verlag.
Wooldridge, M and Fisher M, 1992. "A first-order branching time logic of multi-agent systems" In:
Proceedings of the Tenth European Conference on Artificial Intelligence (ECAl-92), pp 234-238, Vienna,
Austria.
Wooldridge, Mand Fisher M, 1994. "A decision procedure for a temporal belief logic" In: OM Gabbay and
HJ Ohlbach. (eds.) Temporal Logic-Proceedings of the First International Conference (LNAI Volume
827), pp 317-331, Springer-Verlag.
Wooldridge, M and Jennings NR, 1994. "Formalizing the cooperative problem solving process" In:
Proceedings of the Thirteenth International Workshop on Distributed Artificial Intelligence (IWDAJ-94), pp
403---417, Lake Quinalt, WA.
Yonczawa, A (ed.) 1990. ABCL-An Object-Oriented Concurrent System, MIT Press.