Self-organised Distributed Information Processing in cyber-physical and socio-technical systems

在数字化转型中，由一组异构自治和独立单元组成的网络物理和社会技术系统的许多实例可能会遇到做出集体决策的要求。为了达到这个目的，系统的各个单元向彼此提出问题，交换信息，并将一组单独的、独立的、主观的测量或意见组合成一个相互同意的决定。使用社会启发计算的方法论，我们将设计一种被称为社会影响调节理论（RTSI）的心理学理论的算法形式化。RTSI有许多独特的特点，提供了另一种方法，表明社会影响包括影响目标的来源以及目标寻找受其影响的来源，并区分了“形成性”和信息性交流。此外，我们将开发一个创新的自组织多代理模拟器，我们将在其中嵌入该算法，并运行一系列实验来测试分布式信息处理和社会影响日益丰富的方面。首先，我们将研究专业知识的出现，知识的多样性，以及在这种社会技术分布式系统中的工作分配。在处理稳定性和灵活性问题时，我们将强调保持两者之间的平衡，避免停滞不前，避免在变化中花费太多精力。此外，我们打算研究神谕和机器学习在知识的编纂、集体记忆的保存以及减轻寡头政治和多数专制等危害方面的作用，我们将特别研究福柯提出的知识和权力的双重性质。我们将在动态系统中扩展分析，我们将探索“有或没有神谕”的教育是否以及如何作为一种把关知识和验证创新的机制。考虑到已经在多智能体系统中进行了许多研究并且具有许多细微差别和观点的信任在社会技术分布式系统中尤为重要，本研究还将研究欺骗和代价高昂的信号在系统个体之间信任发展中的作用，从而在多智能体系统中做出明智的集体决策的能力。在观察到问题的类型在理解答案中的重要性之后，我们将开发单元提出不同类型的问题并遵循不同反思过程的模型。最后，本研究还可能调查基于区块链的昂贵信号方法的潜在影响，以及这如何有助于问责制和责任概念关系的社会构建，并研究它们在系统性自我完善反思和稳定中的作用。这个EPSRC“人工智能技术”和“信息系统”领域的项目可以在物理或网络分布式信息处理系统中有各种应用。例如，它可以为可持续负责任的信息系统的设计奠定基础，也可以用于许多工业5.0应用，包括智能城市或智能工厂、自动驾驶汽车和机器人系统。

In digital transformation, many instances of cyber-physical, and socio-technical systems, consisting of a group of heterogeneous autonomous and independent units, might encounter the requirement of making collective decisions. To that end, the units of the system address questions to one another, exchange information, and combine a set of individual, independent, subjective measurements or opinions into a single mutually-agreed decision. Using the methodology of socially inspired computing, we are going to design an algorithmic formalisation of a theory of psychology called Regulatory Theory of Social Influence (RTSI). RTSI has a number of distinctive features that offer an alternative approach which suggests that social influence consists of sources influencing targets, as well as, of targets seeking sources by whom to be influenced, and distinguishes between "formative" and informative communication. In addition, we are going to develop an innovative self-organising multi-agent simulator, in which we will embed this algorithm, and run a series of experiments to test the increasingly richer aspects of Distributed Information Processing and Social Influence. Initially, we are going to study the emergence of expertise, the diversity of knowledge, and the distribution of work in such socio-technical distributed systems. Addressing problems of stability and flexibility, we will highlight the significance of maintaining a balance between them, as a way to avoid stagnancy or spending too much effort in changing. Furthermore, we intend to investigate the role of oracles and machine learning in the codification of knowledge, the preservation of collective memory, and the mitigation of hazards such as oligarchy and majoritarian tyranny, and we will, particularly, look into the dual nature of knowledge and power as proposed by Foucault. We will extend the analysis in dynamic systems, and we will explore whether and how education ''with or without oracles" can act as a mechanism of gatekeeping knowledge and validating innovation. Considering that trust, which has been many studied in multi-agent systems and has many nuances and perspectives, is particularly important in socio-technical distributed systems, this research will, also, examine the role of deception and costly signalling in the development of trust between the individuals of the system, and consequently, in the competence of the multi-agent system to make educated collective decisions. After observing the importance of the typology of the question in the understanding of the answer, we will develop models in which the units ask different kinds of questions and follow different processes of reflection. Finally, this research might also investigate the potential impact of a blockchain-based approach to costly signalling and how can this contribute to the social construction of conceptual relations of accountability and responsibility, and examine their role in systemic self-improvement reflection and stability. This project in the EPSRC areas of "Artificial Intelligence technologies" and "Information Systems" could have various applications into physical or cyber distributed information processing systems. For instance, it could establish the basis for the design of sustainable responsible information systems and can, also, be used in many Industry 5.0 applications, including smart cities or smart factories, autonomous cars, and robotic systems.