Towards Robust Open Multi-Agent Systems: Domain-Independent Exception Handling Services

这个建议解决了一个简单的问题：我们如何开发有效的多代理系统，以避免在开放系统环境中可能遇到的容易出现异常（错误、恶意或“愚蠢”）的代理和基础设施？这个问题将通过开发技术来解决，这些技术可以在运行时确定可能发生的异常，以及适合动态变化的代理种群和系统环境的异常检测和解决策略。这项工作将建立在两个关键思想的基础上：(1)有可能识别高度可重用的独立于领域的异常处理专业知识，(2)这种专业知识可以以不同的异常处理服务的形式实例化，这些服务将异常处理从代理中外包出去，只要求它们遵守一些相对简单的准则。这项工作的潜在影响将包括：(1)允许创建能够在开放环境中有效运行的多代理系统，同时(2)通过避免创建具有复杂异常处理能力的代理来减轻代理开发人员的负担。

This proposal addresses one simple question: how can we develop effective multi-agent systems out of the exception-prone (buggy, malicious, or simply "dumb") agents and infrastructures we can expect to encounter in open system contexts? This problem will be addressed by developing technologies that can determine, at run time, what exceptions can occur, as well as what exception detection and resolution strategies are appropriate, for a dynamically changing agent population and system environment. This workwill build on two key ideas: (1) it is possible to identify highly reusable domain-independent exception handling expertise, and (2) this expertise can be instantiated in the form of distinct exception handling services that outsource exception handling from the agents, requiring only that they adhere to some relatively simple guidelines. The potential impact of this work will include (1) enabling the creation of multi-agent systems that can function effectively in open contexts, while (2) reducing the burden on agent developers by obviating the need to create agents with sophisticated exception handling capabilities.