Catálogo de publicaciones - libros

In this tutorial paper we summarise the key features of the multi-threaded Qu-Prolog language for implementing multi-threaded communicating agent applications. Internal threads of an agent communicate using the shared dynamic database used as a generalisation of Linda tuple store. Threads in different agents, perhaps on different hosts, communicate using either a thread-to-thread store and forward communication system, or by a publish and subscribe mechanism in which messages are routed to their destinations based on content test subscriptions. We illustrate the features using an auction house application. This is fully distributed with multiple auctioneers and bidders which participate in simultaneous auctions. The application makes essential use of the three forms of inter-thread communication of Qu-Prolog. The agent bidding behaviour is specified graphically as a finite state automaton and its implementation is essentially the execution of its state transition function. The paper assumes familiarity with Prolog and the basic concepts of multi-agent systems.

In an earlier paper we [6] presented a declarative approach for agent control. In that work we described how control can be specified in terms of cycle theories , which define declaratively the possible alternative behaviours of agents, depending on their internal state and (their perception of) the external environment in which they are situated. This form of control has been adopted for logic-based KGP agents [8, 2]. In this paper we show how using this form of control specification we can specify different profiles of agents, how they would vary the behaviour of agents and what advantages they have with respect to factors in the application and in the environment, such as time-criticality.

We propose a simple event calculus representation of contracts and a reactive belief-desire-intention agent architecture to enable the monitoring and execution of contract terms and conditions. We use the event calculus to deduce current and past obligations, obligation fulfilment and violation. By associating meta-information with the contracts, the agent is able to select which of its contracts with other agents are relevant to solving its goals by outsourcing. The agent is able to handle an extendable set of contract types such as standing contracts, purchase contracts and service contracts without the need for a first-principles planner.

Amongst several fundamental aspects in multi-agent systems design, the definition of the agent interaction space is of the utmost importance. The specification of the agent interaction has several facets: syntax, semantics, and compliance verification. In an open society, heterogenous agents can participate without showing any credentials. Accessing their internals or their knowledge bases is typically impossible, thus it is impossible to prove a priori that agents will indeed behave according to the society rules. Within the SOCS (Societies Of ComputeeS) project, a language based on abductive semantics has been proposed as a mean to define interactions in open societies. The proposed language allows the designer to define open, extensible and not over-constrained protocols. Beside the definition language, a software tool has been developed with the purpose of verifying at execution time if the agents behave correctly with respect to the defined protocols. This paper provides a tutorial overview of the theory and of the tools the SOCS project provided to design, define and test agent interaction protocols.

In open multi-agent systems agent interaction is usually ruled by public protocols defining the rules the agents should respect in message exchanging. The respect of such rules guarantees interoperability. Given two agents that agree on using a certain protocol for their interaction, a crucial issue (known as “a priori conformance test”) is verifying if their interaction policies, i.e. the programs that encode their communicative behavior, will actually produce interactions which are conformant to the agreed protocol. An issue that is not always made clear in the existing proposals for conformance tests is whether the test preserves agents’ capability of interacting, besides certifying the legality of their possible conversations. This work proposes an approach to the verification of a priori conformance, of an agent’s conversation policy to a protocol, which is based on the theory of formal languages. The conformance test is based on the acceptance of both the policy and the protocol by a special finite state automaton and it guarantees the interoperability of agents that are individually proved conformant. Many protocols used in multi-agent systems can be expressed as finite state automata, so this approach can be applied to a wide variety of cases with the proviso that both the protocol specification and the protocol implementation can be translated into finite state automata. In this sense the approach is general. Easy applicability to the case when a logic-based language is used to implement the policies is shown by means of a concrete example, in which the language DyLOG , based on computational logic, is used.

The paper addresses the issue of contextual representations of ontologies, as it arises in the area of normative system specifications for modeling multiagent systems. To this aim, the paper proposes a formalization of a notion of contextual terminology, that is to say, a terminology holding only with respect to a specific context. The formalization is obtained by means of a formal semantics framework which enables the expressivity of common description logics to reason within contexts (intra-contextual reasoning), allowing at the same time the possibility to reason also about contexts and their interplay (inter-contextual reasoning). Using this framework, two complex scenarios are discussed in detail and formalized.

In this paper, we consider the design of normative multiagent systems composed of both constitutive and regulative norms. We analyze the properties of constitutive norms, in particular their lack of reflexivity, and the trade-off between constitutive and regulative norms in the design of normative systems. As methodology we use the metaphor of describing social entities as agents and of attributing them mental attitudes. In this agent metaphor, regulative norms expressing obligations and permissions are modelled as goals of social entities, and constitutive norms expressing “counts-as” relations are their beliefs.

This paper studies compositional semantics of nonmonotonic logic programs. We suppose the answer set semantics of extended disjunctive programs and consider the following problem. Given two programs P _1 and P _2, which have the sets of answer sets AS ( P _1) and AS ( P _2), respectively; find a program Q which has answer sets as minimal sets S ∪ T for S from AS ( P _1) and T from AS ( P _2). The program Q combines answer sets of P _1 and P _2, and provides a compositional semantics of two programs. Such program composition has application to coordinating knowledge bases in multi-agent environments. We provide methods for computing program composition and discuss their properties.

In multi-agents systems, incompleteness, due to either communication failure or response delay, is a major problem to handle. To face incompleteness, frameworks for speculative computation were proposed (see references [5, 6, 7]). The idea developed in such frameworks is to allow the asking agent, while waiting for the slave agents to reply, to reason using default beliefs until replies are sent. In particular, K. Satoh and K. Yamamoto [7] proposed a framework that allows an agent not only to perform speculative computation, but also to accept iterative answer revision for yes/no questions. In this paper, we present an extension of the framework for more general types of questions using constraint logic programming (CLP).

A method to recognize agent’s intentions is presented in a framework that combines the logic of Situation Calculus and Probability Theory. The method is restricted to contexts where the agent only performs procedures in a given library of procedures, and where the system that intends to recognize the agent’s intentions has a complete knowledge of the actions performed by the agent. An original aspect is that the procedures are defined for human agents and not for artificial agents. The consequence is that the procedures may offer the possibility to do any kind of actions between two given actions, and they also may forbid to perform some specific actions. Then, the problem is different and more complex than the standard problem of plan recognition. To select the procedures that partially match the observations we consider the procedures that have the greatest estimated probability. This estimation is based on the application of Bayes’ theorem and on specific heuristics. These heuristics depend on the history and not just on the last observation. A PROLOG prototype of the presented method has been implemented.