Trust Objects: Enabling advanced reputation services on the Web of Trust
Presented by Moses Ma and Dr. Rutu Mulkar-Mehta, FutureLab
Submitted to the 5th Rebooting the Web of Trust Technical Workshop as a discussion paper
Boston, October 3-5, 2017
Keywords: reputation, trust, verified claims, collaboration, innovation, framework, blockchain, decentralized, self-sovereign
We propose to facilitate the collaborative drafting of a technical paper that describes the principles and key design considerations for an online framework to manage fully functional reputation services within a web of trust. The approach uses both transactional and non-transactional reputation data. Non-transactional reputation data includes both trust primitives such as verified claims as well as indeterminate trust assertions. We will also describe the potential use of incentive tokens for incremental optimization of the eco-system, in a manner that is especially suited for decentralized, self-sovereign eco-systems. Finally, we propose to discuss the complexities of online disagreements and how to resolve and adjudicate them in a pareto-optimal manner.
We base much of our work on key design considerations for decentralized reputation, developed by C. Allen and by A.C. de Crespigny et al (see references), at the Spring 2017 RWOT design conference in Paris.
Reputation is social concept that is an essential component of social and business networks, because it serves as an optimizing influence on such systems. However, while there have been numerous analyses of how reputation may be computed and managed, there has to date been no systematic approach for implementing reputation systems, nor strategies for self-optimizing reputation management, proposed for decentralized networks. Our goal is to develop an inter-disciplinary, game-theoretic model for computational trust and reputation based on psychology and micro-economics.
The proposed approach utilizes both transactional and non-transactional trust data. Transactional data includes a record of failed vs successful transactions, such as the history of successful vs unsuccessful transactions at eBay. Non-transactional data includes trust primitives such as verified claims, as well as indeterminate trust assertions. This paper also shows that it is possible use incentive tokens to drive incremental optimization of the eco-system, in a manner that is especially suited for decentralized, self-sovereign eco-systems. Finally, we propose to discuss the complexities of online disagreements and propose key design considerations for systems that could more effectively resolve and adjudicate disputes, in a pareto-optimal manner.
We believe there are several important principles that apply to non-transactional reputation systems, which could be used to help govern their design and operation within enterprises and organizations. These are:
Reputation is complex.
Trust and reputation are transitive.
Reputation is a convolution of trust primitives.
Reputation is a narrative, a dynamic social process, not a static credit score.
Reputation is a currency.
Reputation is all about people.
Using these principles, we offer a model for computational reputation that is functional and useful, adaptive and self-optimizing. In our model, reputation is defined to be a convolution of transactional and non-transactional data, with associated weighting based on the trustability of the rater. We will also discuss the requirements for continuous reputation tracking, adaptive weighting with artificial intelligence based adaptive learning, neural networks for behavioral pattern detector, and automatic normalization of voting weights.
Finally, the most important goal for this working draft is to map the emergent model to the proposed DID and Verified Claims standards. Our proposal is simply to add a field to the basic verified claim system, in the form of a “protocol cookie”. Cookies were designed to be a reliable mechanism for websites to remember stateful information or to record the user's browsing activity. Therefore, this field would enable the claim to remember stateful information – such as a URI for the claim offerer’s reputation rating, or to record the history of entities that have accessed the claim, or to manage visibility settings for the claim, so that disclosure of the claim could be selectively permissioned. The most valuable function of the field would be to provide a URI to an ontology or classification system for the claim to provide metadata about the size or scope of the claim.
The paper will include a design philosophy for the implementation of reputation engines that are “self-optimizing” using “optimization tokens” that promote more effective and truthful rating and reporting by people. These tokens can also be earned by artificial intelligence systems that act in a manner similar to miners, but providing optimization services. For example, neural network based fraud detection systems could be developed to look for “ratings extortionists”, who threaten negative reviews if not provided a discount.
We would like to collaborate with the participants of the Rebooting Web of Trust Workshop to refine the concepts and use cases. Our goal is to produce both an improved, more easily extensible standard, as well as a demo that demonstrates a compelling use case directly after the workshop.