The Effects of Information Dynamics on Trust and Reputation

现代系统越来越复杂，具有多个独立的决策代理。这些代理包括物联网、自动驾驶汽车和人类。单个系统可以包含多种类型的代理。代理人的目标是在他们的最佳利益范围内行动，这就需要一些机制来选择在不确定的情况下与谁互动。类似的有机系统以信任的形式提供了这一点。信任对计算机科学来说并不新鲜，密码学就是一个范例。然而，传统的密码方法往往不可行或不适合复杂的社会。因此，计算信任的现代概念旨在捕捉广义现象来描述现有系统并构建改进的系统。在人口密集的环境中，代理需要某种方式“在地面上”建立信任。一个agent不可能安全地与每一个潜在的合作伙伴进行交互和判断。声誉是用来克服这一点的。代理将在整个网络中共享关于其他人的信息，以建立一个比他们单独更丰富的社区概念。因此，所有这类系统的信任基础是消息传递。在试图理解信息的影响时，很容易把注意力集中在信息的内容上。然而，了解信息传播的时间、地点和方式对于理解其影响至关重要。恶意实体利用网络中的信息流有许多尚未探索的方法。无论是利用缓慢、错误的信息传播来“隐藏”其行为的后果，还是利用快速但未经验证的信息传播来传播消息，信息动态都是任何声誉系统的弱点。本项目旨在研究信息动力学研究的一些基本原则及其对多主体决策的影响，重点是利用和保护个体主体和整个社会。研究方法一般计算信任的研究本身就很新颖。捕捉这一现象的正式尝试甚至更新。因此，要研究信任的新方面，如信息动力学，至少需要对现有方法进行新颖的应用。用于分析网络（网络理论）和相互作用的正式工具，战略代理（博弈论）是存在的，但它们在计算信任的现代概念中的应用是稀疏的。因此，本项目努力借鉴来自各个领域的既定正式方法，以解决我们对复杂系统中的信任和社会决策的理解中目前尚未解决但基本的问题。更确切地说，该项目的目标如下：-了解网络的结构如何影响信息在网络上的流动，以及这对网络成员的影响。+这可以用来构建不太容易受到此类问题影响的网络。-分析恶意实体如何利用这些动态（例如利用信息传播的延迟来“隐藏”后果）。+这可以用来告知机制的发展，以减轻此类网络上的恶意行为。-分析无辜各方的策略，这些策略或多或少会使他们容易受到恶意实体的攻击。+在工程系统的情况下，这可以用来构建更健壮的代理。+在社交网络的情况下，这可以为围绕人类行为者的教育和政策决策提供信息（例如，当信任其他个人、网站或在线新闻来源时）。EPSRC策略数学科学；全球的不确定性;数字经济合作者虽然目前还没有正式的合作者，但一些公司已经表达了对这项工作的影响和成果的兴趣。谈判正在进行中。

ContextModern systems are increasingly complex, feature multiple independent, decision-making agents. These agents including IoT, autonomous vehicles and human beings. A single system can include multiple types of agent. Agents aim to act within their best interests, requiring some mechanism for selecting who to interact with under uncertainty. Similar organic systems provide this in the form of trust. Trust is not novel to computer science with cryptography being the paradigm. However, conventional cryptographic methods are often infeasible or unsuitable for complex societies. As such, the modern conception of computational trust aims to capture the generalized phenomenon to describe existing systems and construct improved ones.In highly populated environments, agents needs some way of establishing trust "off the ground". It is infeasible for an agent to safely interact with and judge every potential partner. Reputation is used to overcome this. Agents will share information about others throughout the network to build a richer idea of the community than they could alone. As such, the basis for trust on all such systems is message passing. It is easy to focus on the content of a message when trying to understand its impact. However, knowing when, where and how a message spreads is crucial to understanding its influence. There are many unexplored ways in which a malicious entity can exploit information flow in a network. Whether exploiting slow, faulty information propagation to "hide" from the consequences of their actions or exploiting fast but unverified information propagation to spread a message, information dynamics are a vulnerability of any reputation system. This project aims to look at some fundamental tenets of the study of information dynamics and its influence on multi-agent decision-making, with a focus on the exploitation and protection of individual agents and the community as a whole.Research MethodologyThe study of general computational trust is itself quite novel. Formal attempts at capturing this phenomenon are even newer. As such, to study a novel aspect of trust such as information dynamics would require at least the novel application of existing methods. Formal tools for analysing networks (network theory) and interacting, strategic agents (game theory) exist but their application to the modern conception of computational trust is sparse if at all present. As such, this project endeavours to draw from established formal methods from a variety of fields to tackle a currently unresolved but fundamental issue in our understanding of trust and social decision-making in complex systems. More precisely, the project aims for the following:- Understanding how the structure of a network affects the movement of information on it and the impact of this on the members of the network. + This can be used to then construct networks which are less vulnerable to such issues.- Analysing how malicious entities may exploit these dynamics (e.g. exploiting delays in information propagation to "hide" from consequences). + This can be used to inform the development of mechanisms to mitigate malicious behaviour on such networks.- Analysing strategies of innocent parties that are more or less likely to make them vulnerable to the endeavours of a malicious entity. + In the case of engineered systems, this can be used to construct more robust agents. + In the case of social networks, this could inform education and policy decisions around human actors (e.g. when trusting other individuals, websites or news sources online).EPSRC StrategiesMathematical Sciences; Global Uncertainties; Digital EconomyCollaboratorsWhile there are no formal collaborators as of yet, some companies have expressed their interest in the ramifications and outcomes of this work. Talks are ongoing.