Agent’s Programming from a Mental States Framework Milton Corrêa1 and Helder Coelho2 1 Serviço
2
Federal de Processamento de Dados, Rio de Janeiro, Brazil
[email protected]
Departamento de Informática, Faculdade de Ciências, Universidade de Lisboa, Portugal
[email protected]
Abstract. In the paper we present a methodology to agents’ programming based on a Mental States Framework (MSF) in which mental states types are defined in terms of basic components as an External Content; criterions to determine unsatisfaction, uncertainty, urgency, insistence, intensity and importance associated to a mental state; laws of causality through which a mental state can produce another; and control mechanisms for provoking, selecting, suspending and canceling the processing of a mental state. Besides the mental states types named Belief, Desire and Intention as usually defined, this framework also includes Expectation and any possible other mental states’ type that may be important when modeling interactions among agents in a more complex society. According to this methodology, the agents’ architectures are defined by selecting a set of mental states as established in this framework. And, in the paper it is shown that the Object Oriented Programming is well suitable and interesting to the implementation of those architectures. Therefore an Agent Oriented Programming is seen as an interaction among mental states and architectures spaces by MSF and the Object Oriented Programming.
1
Introduction
The agents’ programming paradigm being the support for the fast technology development based on agents in the 90’s comes to supply the growing processing complexity in what concerns information in organization, in process control in distributed system in which the techniques upon centralized control become impossible to be applied as for instance, international air traffic control, interoperability among knowledge systems or information exchange among heterogeneous information systems as in internet either in a company environment or in groups of companies where a lot of information are found in legacy systems [25]. Great efforts have been devoted in agents’ conceptualization and theories concerning agents’ interactions in a cooperative society, i.e. in building multi-agents systems (MAS). Regarding the engineering out look, there are proposals for agents’ architectures and programming techniques that are still under development. Nevertheless, there is a great gap not fulfilled between theory and agents’ building engineering.
1
Most of MAS architectures may be described as a common set of components which suit the basic capacities for the agents: reactivity, that is, react to stimulation or environment changes; deliberation, that means, the agent decides according to reasoning; interactions with other agents; a set of actions; goal oriented; autonomy, that is, being able to take decision based on their internal states without external intervention [16]. A sort of architectures that has been largely explored in recent years is based in the agents’ modeling from Beliefs, Desires and Intentions (the so-called BDI architectures) [2, 19]. In Corrêa and Coelho [10] is proposed a Mental States Framework (MSF) to standardize the commonly notion of usual mental states as Beliefs, Desires and Intentions and other not so much used in DAI as Expectations. It is shown that from this framework the research and application to build agents with new other possible types of mental states (as Hopes) can be possible. A space of mental states is built up on types of mental states and a space for possible agents’ architectures is built up on that mental state space. In this paper is shown that this framework can be applied to agent programming based on the Object Oriented Programming paradigm. Our aim is to establish a methodology: 1) The mental states are characterized in terms of a small set of basic attributes; 2) we argue that these attributes are, at least, sufficient to define mental states; 3) other possible mental states could also be characterized by assigning them a set of those attributes; 4) This mental states model can explain agents' interactions; 5) An Agent Programming as an extension of an Object Oriented Programming results from this mental states framework. Our starting point is the observation that mental states (MS) as Beliefs, Desires and Intentions, are commonly defined as being structured on basic components. Cohen and Levesque [5] defined Intention as being structured on “Choice” plus “Commitment”, in other works "Commitment” is also thought as more basic than Intention [4, 19]. Werner [24] treated Intentions as Strategies, and Bratman [2] defined them as Plans. The components of Belief usually are a Proposition and a "true" or "false" value associated to it or a degree of Certainty and can also have a degree of Importance [18]. On the other hand, Sloman[20, 21] pointed out that urgency, importance, intensity and insistence are behind agent's actions. In Corrêa and Coelho[8] and Corrêa[9] is shown that a complete functioning agent needs a specification based on the notion of architecture, but highlighted by the dynamics of mental reactions. The way all mental sates change and interact is carefully controlled [11]. We argue that other mental states than Belief, Desires and Intentions can be necessary to understand and to explain the complexities of agents behavior in a society and a complete and consistent theory about agents' mental states has not only to explain and build the artificial agents but also to understand the natural ones [12]. For instance, Expectation is a mental state that enables more flexibility and more complex behaviors [8, 17, and 23]. Expectations and Intentions can complement one another. Intentions can embody an agent's expectations that the agent act in ways to satisfy those intentions, while expectations can lead an agent to form intentions to check that those expectations are satisfied. Our approach considers mental states as
2
organizations of agent's internal processing of information related to their actions, and a fundamental feature of these organizations is that they are related to situations in the world (intentionality). Mental states are guides for agent's actions such that the agent behavior can be explained on them and, on the other hand, mental states can interact to produce the agent behavior. We assume that there is a limited set of basic attributes such that a mental state is defined in terms of some combination of them.
2 A Framework for mental states In order to have a theoretical structure to define mental states we need to find an agreement about their basic components or attributes. This can be obtained by observing the conceptions and applications of the usual mental states and filtering a common set of them. On the other hand, these attributes must be put together and analyzed if they are capable to offer a base to define the usual mental states of DAI (Belief, Desires and Intentions) and other not well applied yet, although known from Psychology, Philosophy and Economics as relevant to explain human behavior as Expectations, Hopes and Necessities. When we diversify applications and work within Social and Human Sciences or more complexity interactions of economic agents we are forced to adopt such other mental states [9, 20, 21]. To make references to these attributes we organize them in three groups: the first called "nucleus" contains the mental state's external content and a set of criterions for unsatisfaction, uncertainty, urgency, intensity, insistence and intensity; the second called "laws" contains a set of possible causal relationships among mental states and the last called "controls" contains a set of controls to trigger, choose, suspend and cancel mental states. 2.1 Nucleus These attributes define proper characteristics of MS that is, the MS with the same set of attributes are classified as the same type as shown in the paragraphs below, but every particular MS is distinguished, at least, by its external content. Normally an MS is distinguished from other MS of same type by the attributes of the nucleus. - External Content (Ex. Content). The mental states have external significance; that is to say, they are linked to the world in terms of what they are about. This external content is represented by a logical proposition (e.g. in a multimodal logic). - Unsatisfaction is related to the stimulation of actions or changes of mental states. We consider that this component is the central motivator of actions and the producer of other mental states. A function sx ( sx: Á Á ... Á§Á) is defined as a “degree of unsatisfaction”. For instance, a MS X is satisfied if sx(r) = 0 (r°Á Á ...Á) So, we need some criteria to decide if a MS is satisfied or not. - Uncertainty is a measure of agent confidence regarding the situation that corresponds to the mental state. A function cx (cx: Á Á ... Á§Á) is defined as a “degree of uncertainty”.
3
- Urgency is a measure of how much time remains to the point the corresponding MS X must be satisfied. A function ux (ux: Á Á ... Á§Á) is defined as the “degree of urgency”. - Intensity is related to the agent's pledge and energy dedicated to an MS. For example, if an agent is trying to satisfy an MS, the intensity of this MS is connected to how actively or vigorously this satisfaction is pressed and adopted. A function vx (vx: Á Á ... Á§Á) is defined as a “degree of intensity”. - Importance is related to a valuation in terms of benefits and costs the agent has of a corresponding mental state situation. A function mx (mx: Á Á ... Á§Á) is defined as a “degree of importance". - Insistence is related to how much dificult it is for an agent to abandon a MS. For instance, if the agent strongly insists on some goal, this goal will not be abandoned easily . A function nx (nx: Á Á ... Á§Á) is defined as a “degree of insistence”. 2.2 Laws The laws define how the mental states are combined to cause other mental states or agent's actions. For example, under certain conditions, a Belief and a Desire cause another Desire: an agent A's Desire to learn mathematics and the Belief that agent B knows mathematics and that B can teache A, cause A's Desire to learn mathematics from B. Another law is: a Desire and a Belief can cause an Intention: A's Desire to learn mathematics from B and A's Belief that in order to learn from agent B there is a strategy. That is, A must know how to learn from another agent. Thus, there is an A's Intention to learn mathematics from B. A collection of laws relating Belief, Desire, Intention and Expectation is presented in figure 1 according to Corrêa [9]. Another demonstration of such mental states dynamics applied in a Tutor/Learner session is shown in Moussale et al. [15]. 2.3 Controls These attributes define how and when an MS causes another, can be suspended, canceled or stayed active. An MS is active when it produces or influences another MS, causes an agent action or contributes indirectly to an agent action. An MS is suspended when it is temporarily inactive. A MS is cancelled when it is permanently inactive. Let X be a mental state. The possible laws will be triggered if sx(r) > 0 (r°Á_Á_...Á) and at least one of the conditions (C1 to C7) below occurs: C1) ux (r) > 0 and vx (r) > 0 and mx (r) > 0; C2) ux (r) > 0 and mx (r) > 0; C3) ux (r) > 0 and vx (r) > 0; C4) vx (r) > 0 and mx (r) > 0; C5) mx (r) > 0; C6) vx (r) > 0;
C7) ux(r) > 0
C8) Canceling or suspending a MS: If X is an active MS and gx(r) < 0 then cancel X If X and Y are active and conflicting MS then suspend X if gx(r) < gy(r) or suspend Y if gy(r) < gx(r);
4
else if gx(r) = gy(r) choose X or Y randomly. Where gx(r) and gy(r) are the interrupting functions, defined in terms of urgency, intensity and importance of the corresponding MS. We will not present here a specific definition of conflicting mental states. We consider that two MS of the same type are conflicting when they cannot coexist as active mental states at the same time in the agent mind. C9) Activation of a suspended mental state. If X is a suspended MS and there is no other active MS conflicting with X then X is activated, unless if there is an active MS Y conflicting with X then X is activated if nx(r) > 0, X is maintained suspended if gx(r) >0 or X is canceled. C10) Finding a strategy to satisfy an MS. If there is no strategy or means to satisfy a MS X then if “it is possible to find a strategy or means to satisfy X" then “find and adopt this strategy or means” else if gx(r) > 0 X is suspended otherwise X is canceled. C11) Find alternatives when an adopted strategy doesn't work anymore. If K is an adopted strategy to satisfy an MS X and, and at some moment, it is not possible to satisfy X through K then find another strategy if possible and nx(r) >0; suspend X if gx(r) >0 or cancels X. The relationships among these attributes and the mental states, are shown in the table of figure 1.
5
Ex. Content Unsatisfaction Uncertainty Urgency Importance Intensity Insistence
B x
D
I
x x
x x x x x x x
x x x
x x x x
L1) B
