Decision Making in Heterogenous Multiagent Systems

最近发表的“人工智能百年研究”强调了开发人类感知系统的重要性，这种系统的设计明确考虑到人类是一个不可分割的组成部分。例如，人们可以设想与人类无缝交互的自主系统，根据需要利用计算代理和人类的优势，增强两者的能力。然而，为了实现这样的系统，有许多研究挑战需要解决，从支持人机协作的接口设计到有效交互的正式模型。******一个核心挑战是确保人类和代表他们的代理的偏好和激励适当地对齐，因为误解和不对齐的激励可能导致意想不到的结果和不受欢迎的行为。此外，系统必须明确设计，以确保稳健和透明的结果，尊重代理人（包括人类和计算）的偏好和价值观。在本提案中，我描述了一个研究议程，通过发展和利用算法博弈论、计算社会选择以及更广泛地说，多智能体系统和人工智能方面的研究，将支持强大的人类意识系统的发展。******这项研究的影响将在多个维度上感受到。首先，它将解决多智能体系统和人工智能中的基本问题。这一领域的进展将支持健壮的多代理系统的开发，这些系统可以在广泛的应用领域中使用。最后，该提议的核心是培养人工智能领域的高素质人才，这一领域在全国所有行业和部门都有强劲的需求。

The recently released ``One Hundred Year Study on Artificial Intelligence" emphasized the importance of developing human-aware systems, that are explicitly designed taking into consideration the fact that humans are an integral part. For example, one can envision autonomous systems that seamlessly interact with humans, leveraging the strengths of both computational agents and humans as required, augmenting the abilities of both. However, in order for such systems to be realized, there are numerous research challenges which need to be addressed, ranging from the design of interfaces to support human-machine collaboration to formal models of effective interaction.******One central challenge is ensuring that preferences and incentives of both humans and the agents representing them are appropriately aligned, as misunderstandings and misalignment of incentives can lead to unexpected outcomes and undesirable behaviour. Furthermore, systems must be explicitly designed so as to ensure robust and transparent outcomes, respecting the agents' (both human and computational) preferences and values. In this proposal I describe a research agenda that, by developing and levering research in algorithmic game theory, computational social choice and, more broadly, multiagent systems and artificial intelligence, will support the development of robust human-aware systems.******The impact of this research will be felt across multiple dimensions. First, it will address fundamental questions in multiagent systems and artificial intelligence. Progress in this area will support the development of robust multiagent systems that can used across a wide range of application domains. Finally, central to the proposal is the training of highly qualified personnel in artificial intelligence, an area where there is strong national demand across all industries and sectors.