SingularityNET Moves Toward Social Computing With Proof-of-Reputation

Our latest design overcomes one of blockchain’s major barriers: designing a decentralized reputation management system.

Our internal development team continues to experiment with new features for the SingularityNET platform. This post discusses our latest design for a Proof-of-Reputation mechanism, an important advancement that aims to replace legacy Proof-of-Work and Proof-of-Stake consensuses. While this mechanism is still in the design phase, it will serve as a core feature of SingularityNET to allow for better AI discovery through acquired reputation.

In Part 1 of this technical series, we covered how SingularityNET will make it possible for any non-professional user to create, educate, train, and launch their own personal AI agent.

In Part 2, we covered some proposed integrations with the OpenCog platform.

Now, let’s get to the social computing aspect of SingularityNET, with reputation management being a critical component.

Reputation Management System Overview

The reliability of a decentralized financial system such as blockchain relies on distributed consensus. But designing a functional reputation system remains one of the key challenges for any social computing platform.

To address both needs, the SingularityNET team has designed a Proof-of-Reputation consensus. Our latest design features two breakthroughs:

1. It can accurately aggregate ratings while preventing users from gaming the system. A unique property of our design is its iterative calculation of hierarchical reputation trees in sliding temporal windows. Primarily, it makes reputations of nodes in a network bound to reputations of other nodes providing the ratings, combined with actual values of economical interactions between nodes. Proper tuning of system parameters should make it possible for two things to happen. First, newcomers can get on social elevators and attract attention to high quality services. Second, the reputation system should prevent intruders from gaming it and taking over the consensus.
2. It’s adaptable, with the long-term potential of replacing Proof-of-Work and Proof-of-Stake consensuses. Evaluated reputations within our design can be used for different purposes. On one hand, the best providers of AI services and algorithms can be selected based on accumulated ratings. At the same time, reputations could be used to supply reliable consensus for underlying blockchain or hash-graph systems, potentially replacing legacy Proof-of-Work and Proof-of-Stake consensuses.

Designing Our Reputation Management System

The SingularityNET reputation system is designed in two isolated layers.

1. Capturing reputation data — The first layer is responsible for measuring network dynamics for evaluation of reputations in an open, distributed ledger. It will also store snapshots of reputation evaluations in the same ledger data.
2. Making sense of the reputation data — The next layer is in charge of collecting explicit ratings and performing calculations for reputations. The calculation is done based on the dynamics of performed services, financial transactions, and the explicit ratings exchanged by Nodes.

To prevent gaming the system while iteratively improving the reputation system’s design, we may introduce multiple competing Reputation Agencies that leverage different techniques for reputation evaluation. This work would be open source, allowing anyone to contribute and experiment in improving the network.

Technical Explanation of How Our Aigents Integration Can Aid in Reputation Management

In February, we announced our partnership with Aigents. Our ongoing collaboration has already proved fruitful.

For example, the integration of the Aigents platform into SingularityNET can provide the data necessary for many of our proposed reputation mechanisms. Aigents can mine social graphs and relationships and patterns in public networks such as Ethereum, Steemit, and Golos. We just need to register the SinglularityNET Agent Service identifier in service ontology, then provide adapter classes for each of the service methods, and then register each adapter class with a service id. This is can be done with a few lines of code in Python, as Python is used for reference implementation in an early version of SinglularityNET.

At the moment, there are four services of the Aigents platform already integrated into SingularityNET.

1. Social Grapher — Used to study social structures in existing blockchains.
2. News Feeder — Capable of producing self-generated RSS news, which is extracted from online sources by proprietary agents owned and trained by users.
3. Text Clusterer — Intended for aggregative clustering of texts and web pages.
4. Text Extractor — Can extract meaningful relationships from texts in terms specified by users.

Plugging Aigents into SinglularityNET can also improve the richness of the environment. Besides mining social graphs in public blockchains, Aigents can learn and extract patterns and categories from online sources. They can learn
from these categories and patterns from the experiences of users and their social interactions, implementing so-called experiential learning.

In this scenario, users can create his or her own personal agent and feed it with explicit knowledge known to the user. Then, the agent can collect implicit knowledge possessed by user in their social interactions on social networks. Further, the agent can start feeding the user with information such as detected news items and discovered categories and relationships. In turn, users can give agent explicit feedback on its discoveries so the agent can improve over time.

For example, Aigents can study social interaction patterns in existing social networks, strictly within the permissible bounds of APIs supplied by these networks. It makes it possible to detect multiple behavioral patterns and social types based on online activity, such as opinion leaders, opinion translators, listeners, collaborators, flamers, etc. By connecting such individual patterns into graphs, it is possible to figure out paths of opinion and reputation acquisition.

Our current work is focused on augmenting these graphs with emotional context learned from textual communications.

We’re Just Getting Started

2018 will be a pivotal year for SingularityNET. There will be much more news to come, including platform updates, new team members, and network upgrades.

Subscribe to our newsletter, join our Telegram Channel, and check out the community-governed subreddit for more information.