Comprehensive-Information-Based Formal Pragmatic Approach For Agent Communication

Multi-Agent System (MAS) is essential to distributed Artificial Intelligence in the network environment. However, there are many challenges in current MAS researches, ranging from infrastructure, methodology, reusability, to interaction and collaboration. Agent communication, which is essential to the question about Agent's interaction and collaboration, has a very important value for researches.Agent has two key characteristics, which are autonomy and heterogeneity. The heterogeneity means that Agents are usually developed by different vendors, through different developing paradigms, for different service goals, accordingly with diverse internal structures. The autonomy means that every Agent, running in an open environment, holds its own internal autonomous problem-solving mechanism, which makes it be able to control its own behaviors and states in order to achieve expected goals based on internal policies and external conditions, without external guidance and orders. An autonomous and heterogeneous Agent shows itself the intelligent capability, which is a key feature of Agent's internal processing mechanism. Accordingly, Agent is obliged to hold some intelligent capability in its external communication issues, for example, negotiation, conflict resolution, cooperation, proof finding, and etc. Thus, simple message exchanges are inadequate to Agent communication. Instead, Agent communication should has some kind of coherent characteristic, just the same as human communication, which means an utterance should have relevant syntax, semantic, and pragmatic links with other utterances in the same context. This kind of relativity in syntax, semantic, and pragmatic triunity provides the foundation for Agent's intelligent communication.The motivation of this dissertation is to discover the mechanism of syntax, semantic, and pragmatic triunity in the process of Agent communication, and exploit a formal approach based on this mechanism to promote the flexibility of Agent's interaction. For Agent communication, communicative act is the key point to explore the triunity and Interaction Protocol (IP) is a kind of high-level abstraction representing Agent's interaction process by a series of communicative acts, which is also regarded as a pragmatic mechanism that should be researched more deeply. For this reason, our fundamental research methodology is to define Agent's communicative acts formally first. Based on that, we then provide a flexible IP specification method. The whole system is called CI-based Formal Pragmatic Approach (CIFPA, CI refers to Comprehensive Information).Our main contributions include:1. We established CIFPA logical model for Agent's conversation, which provides a foundation for the flexible IP specification method. This job is guided by the Comprehensive Information Theory and Speech Act Theory, based on Social Commitment semantics. Uniquely, Agent's reasoning and argumentation are introduced into this model, so that semantic elements and pragmatic elements, public pragmatic elements and private pragmatic elements, are integrated into one whole logical framework. Compared with other ACL semantics based on social approach, CIFPA has some unique features: 1) The orientation of this logical framework and model is to start from discovery of syntax, semantic, pragmatic triunity and then to serve the triunity. 2) We defined the semantic of commitment through modal reachability semantics, and established complete commitment category. 3) The logical model is based on Computational Tree Logic and Dynamic Logic, so that it can describe the Agent's interaction process as static-dynamic-combined State Transition System in a natural, intuitive, unambiguous, concise way. 4) We provided a complete category for commitment's states, in which commitment's existence state and content state are distinguished, which accords with intuition. 5) We defined plenty of communicative acts, with consideration of Agent's reasoning and argumentation capability.2. We developed a flexible IP specification method based on CIFPA semantics, which has the considerations about Agent's reasoning and argumentation capability so that opportunities and exceptions in the process of interaction can be handled properly, and hence IP has better flexibility. This method has some advantages: 1) The specification of IP has a high-level abstraction, and the key element of specification, that is commitment, is simple and has meanings. This makes the designer of IP be able to define IP specification in a high-effective and direct way. 2) IP specification naturally embeds Agent's reasoning ability, so that it can represent and handle various situations flexibly. 3) IP specification is declarative (not imperative), which provides a foundation for the combine of reasoning and interpreting execution. This embodies Agent's feature reasoning based on knowledge, which enhances the flexibility of IP more.3. We developed a set of methods to simulate IP specified in CIFPA method and verify some key properties of these IPs in formal verification technology. We must do that because CIFPA method provides flexible IP specifications, which surely demands higher requirements in the simulation and verification of IP. For example, validity should be verified to judge whether Agent complies with IP specification and fundamental semantics or not. Another example is correctness, to ensure an IP specification designed by some person is meaningful, reasonable, safe, terminable, and meets other key properties. Excellent IP simulation and verification mechanism can help designers detect errors and potential non-consistency of IP specification as early as possible, so that flexible, valid, correct, safe IPs can be specified.4. We developed an Agent implementation framework in FIPA Agent platform JADE, which makes an Agent be able to have conversation in a flexible way. In the conversation, Agent will continuously consult a knowledge base, which combines IP specification and Agent's internal policy, to ensure the interaction complies with IP at the same time that Agent can make flexible decisions and choices based on its policy during the interaction process.In a summary, the main contribution of this dissertation is establishing a formal approach for Agent communication, centered on a pragmatic element, which is commitment, under the guidance of Comprehensive Information Theory. This approach includes a fundamental semantics to define communicative acts, an IP specification method based on this semantics, and a set of methods for the simulation, verification, implementation and execution of IP specified in above IP specification method.