OriginTrail 백서

OriginTrail

원문

bitpro가 원문에서 글자만 모은 내용이에요. 표·그림·수식은 빠지거나 깨질 수 있어요. 전체 2쪽 중 2쪽

s knowledge mining and paranet-specific AI services. ● A supported blockchain on which the paranet is running and assembling Knowledge Assets, The characteristics of a paranet, including its knowledge asset parameters and how services are provisioned, are all defined by the paranet operator, which can be an individual, an organization, or a Decentralized Autonomous Organization (DAO). Paranets together form the DKG, leveraging the common underlying network infrastructure. Given the DKG is a permissionless system, anyone can initiate a paranet. 7 Figure 5: Paranets as a collection of Knowledge Assets Paranets provide a powerful substrate for AI systems, leveraging network effects of verifiable inputs from multiple sources to receive accurate answers through dRAG, allowing it to gather information from the (1) graph of public knowledge and (2) privately held knowledge in relevant Knowledge Assets that it has access to. Paranets can form around topics of different nature and size, for example: ● Industry 4.0 ● Decentralized Science ● Sustainability ● Public company reports ● LLM training data ● Metaverse ● Video content networks ● Social media ● Art discovery networks ● Prediction markets ● Sports and betting ● Entertainment While some of the examples are inspired by the ongoing adoption seen on the DKG, the list is by no means exhaustive and serves as a showcase of the variety and inclusiveness of different autonomous paranets that DKG can support and make available to AI systems. All paranets are together part of the DKG and can be queried and accessed in combination. 5 Incentivising Verifiable Internet for AI on NeuroWeb The dRAG framework offers a strong approach to addressing many of the key challenges of AI, however, the value of using dRAG grows exponentially with the amount of knowledge AI tools can access in the DKG. To bend the adoption curve steeper, incentivization mechanisms are a key element . The incentivization of Verifiable Internet for AI is implemented by allowing paranets operators to leverage NeuroWeb incentives to fuel the paranet 8 growth . The size of emissions and proposed rules to drive knowledge mining on a particular paranet are decided on with decentralized governance. The core activity of creating knowledge on paranets to earn incentives is Knowledge Mining. 5.1 Knowledge Mining When launching an autonomous paranet, the paranet operators define the key characteristics. One of the most important ones is the incentivization mechanism for knowledge mining on their paranet. Such mechanisms may include conditions like topic relevance, ontology compliance, size limitations, and many others. In addition to knowledge miners, incentivization mechanisms may also reward service providers. Services can range from simple ontology compliance checks to AI-powered chat applications. A paranet can be proposed and operated by anyone and they can be in any size that the operators feel is relevant - they can be functional both as narrow and wide. Paranet operators compete to obtain the rewards for their paranets with other paranets through NeuroWeb governance voting . Each operator creates a reward proposal specifying the reward emissions they require, the knowledge mining mechanism, incentivized services, and other key elements that operators or the voting NeuroWeb community may find to be important. Figure 6: Incentives flow in autonomous paranet In this way, there is an ongoing motivation to publish knowledge on the DKG which is tailored to the relevant domains, increasing the value of DKG for those building dRAG solutions. 6 AI - powered Autonomous DKG The combination of the DKG, dRAG, and Verifiable Internet for AI incentives creates a structure that supports the constant growth of global knowledge. Knowledge miners will start to mine knowledge in a particular paranet by gathering data from relevant sources using manual and automated techniques. However, once more knowledge becomes available in a paranet, the possibility of autonomous knowledge mining is unlocked. As high-quality, annotated, ontology-fit data is present in a paranet, we can run deductive reasoning which is a logical process where new knowledge is deterministically made from existing knowledge based on ontology rules and connections/relationships between knowledge. The autonomous knowledge graph equally offers opportunities for inductive reasoning where new knowledge can be created in a probabilistic manner, by using AI to identify patterns and regularities in existing knowledge and create predictions on the best new knowledge. 9 An important body of work advancing this field is being created with foundation models for knowledge graph reasoning [6] which merge the foundation models used in AI systems onto a knowledge graph structure. This introduces the possibility of universal and transferable graph representations which would allow an AI system [7] to infer and populate any knowledge graph using its foundation model and vice-versa, get fine-tuned by consuming graph representations from existing knowledge graphs. Particularly powerful at such predictive tasks are the previously mentioned Graph Neural Networks (GNNs), since this is one of their key use cases. The autonomous Decentralized Knowledge Graph will effectively use both types of reasoning (logical and probabilistic) to enrich and expand the DKG, driving a better understanding of relationships and entities, and making more informed predictions and recommendations to autonomously create new knowledge. 7 Economics of Verifiable Internet for AI The Verifiable Internet fosters a cohesive approach towards the vision of more inclusive and democratic AI with two basic concepts driving its unit economics: 1. AI-ready Knowledge Assets 2. Knowledge paranets Each of these units is supported with a tokenomics design ensuring the OriginTrail DKG network is enhanced in security, resilience, and usability. The incentive structure enacted on the NeuroWeb blockchain ensures that activities contributing to the DKG network growth receive attractive rewards and create further opportunities for building knowledge as a foundation for Verifiable Internet for AI. 7.1. TRAC utility token driving Verifiable Internet for AI infrastructure Knowledge Assets are the most basic unit economics concept discussed as part of the Verifiable Internet for AI. Each Knowledge Asset generated consumes a certain unit of service within the OriginTrail DKG which is provisioned in a permissionless way by node runners and delegated stakers. Every time knowledge publishers add a new Knowledge Asset to the DKG, they use TRAC to compensate for the use of the DKG network services and use a gas token for transactions. The gas token depends on the blockchain the DKG is being used for publishing (e.g. NEURO on NeuroWeb, DAI on Gnosis, ETH on Ethereum, etc). The nodes in the DKG network compete to collect publishing fees provided by knowledge publishers. They compete on three core elements of (1) providing adequate service (storage and availability of published knowledge) for paranets, (2) the amount of TRAC stake on their node, and (3) node paranet address required for efficient distribution of knowledge and rewards across the network. The nodes with the highest scores according to the criteria are eligible to collect their portion of rewards throughout the duration of the period a Knowledge Asset has been published for. Since TRAC staked is a critical component to the node’s success, node operators can also allow other TRAC holders to delegate their TRAC to their node stake. In exchange for their delegated stake, node operators share a portion of the node rewards with delegators, according to the relative size of their delegations. TRAC stake also has a vital security role as it ensures that the DKG nodes are performing their services adequately. If they act maliciously, their stake can get slashed. 7.2. NEURO native token of NeuroWeb - incentivizing Verifiable Internet for AI Knowledge paranets , as explained in previous chapters, are a collection of Knowledge Assets on a particular topic and associated services. Paranets are proposed by paranet operators , an entity that also defines the core rules of the paranet. In addition to the technical rules (e.g. ontologies, data processing rules, service requirements, etc), paranet operators may also define the paranet incentive system. The operators may choose to incentivize knowledge mining (publishing of new Knowledge Assets to their paranet), knowledge validation services (ensuring a required level of data quality), knowledge access services (e.g. running an AI discoverability service, public interfaces, or AI agent service) or other paranet specific services. 10 Once defined, the paranet proposal for incentives can be submitted to a governance vote towards the DKG incentives pool on the NeuroWeb blockchain . The NEURO token community then acts as a (de)central(ized) bank and decides which areas of the knowledge economy will create the most positive effects for the ecosystem and vote on them receiving the proposed amount of NEURO emission for running their incentive structure. Important to highlight is that NEURO incentives can also be issued for paranets running on other blockchains by providing verifiable proof of activity. The NEURO rewards, however, have to always be collected on NeuroWeb as that is where they are issued. In addition to being used for voting on paranet incentives, NEURO has utility as the native asset of the NeuroWeb blockchain on Polkadot . By being a native asset, NEURO is used as gas for transactions, for running and staking on parachain collators, and for governance voting on NeuroWeb. 7.3 Token economics overview Growing and using the Verifiable Internet for AI via the OriginTrail DKG means always interacting with at least two decentralized networks, the DKG, and a corresponding blockchain. For this reason, using the infrastructure always requires at least one more utility token besides TRAC. Below is an overview of key actions you can perform on the two layers and the corresponding asset you would have to use to execute them. Table 1: Overview of key actions and corresponding assets for two layers TRAC NEURO Other assets Knowledge Assets Publishing Knowledge Assets Compensating for use of DKG. Gas token for txs if using NeuroWeb. Gas tokens for txs (e.g. DAI, ETH) if using other chains Running the DKG infrastructure (Delegated) staking on DKG nodes. Gas token for txs if using NeuroWeb. Gas tokens for txs (e.g. DAI, ETH) if using other chains Knowledge Paranets Incentivizing Knowledge mining Used when publishing Knowledge Assets Providing incentives for publishers. Not applicable Incentivizing Knowledge services (e.g. knowledge validation, discoverability...) Not applicable Providing incentives for service providers. Not applicable NeuroWebAI Blockchain use Not applicable Gas token for txs on NeuroWeb. Not applicable Collators system Not applicable Running and staking on collators. Not applicable Governance Not applicable Voting on updates and treasury and paranets growth incentives for knowledge mining. Not applicable 11 8 Conclusion The exploration of the Verifiable Internet for Artificial Intelligence requires us to look into the details of three converging technologies of AI, Internet, and blockchain. The fusion of these cutting-edge domains promises to reshape our digital landscape, creating unprecedented opportunities while addressing critical challenges. This paper introduces an AI ecosystem that is not only powerful and dynamic but also trustworthy and transparent. The OriginTrail Decentralized Knowledge Graph (DKG), with Knowledge Assets as its primary resource, is a pivotal innovation in this context, offering a robust framework for ensuring the ownership, discoverability, and verifiability of information utilized by AI systems through the dRAG framework. Autonomous paranets, characterized by their self-operating and community-owned nature, offer a modular and scalable approach to organizing Knowledge Assets and growing the DKG. The incentives for continued growth of Verifiable Internet for AI span across multiple networks. Different assets are used according to their respective designs - TRAC for operating, securing, and using the DKG infrastructure and NEURO for incentivizing growth in addition to the utility as the native NeuroWeb blockchain token. The convergence of AI, the Internet, and crypto as embodied in the Verifiable Internet for AI represents a transformative step towards a more transparent, equitable, and intelligent digital world. 9 References [1] Chui M, Roberts R, Yee L., Hazan Z., Singla A., Smaje K., Sukharevsky A, Zemmel R, (2023) The economic potential of generative AI: The next productivity frontier. URL: https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/the-economic-potential-of-generative-AI-the-n ext-productivity-frontier [2] Shumailov, I., Shumaylov, Z., Zhao, Y., Gal, Y., Papernot, N., Anderson, R. (2023) The curse of Recursion: Training on generated data makes models forget. New York, May 31., URL: https://arxiv.org/abs/2305.17493 [3] Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N., Küttler, H., Lewis, M., Yih, W., Rocktäschel, T., Riedel, S. and Kiela, D. (2021) Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. New York, April 12., URL: https://arxiv.org/abs/2305.17493v2 [4] Pan, S., Luo, L., Wang, Y., Chen, C., Wang, J., Wu, X., Fellow. (2024) Unifying Large Language Models and Knowledge Graphs: A Roadmap. Jan 25., URL: https://arxiv.org/pdf/2306.08302.pdf [5] Sporny M., Longley D., Chadwick D., Steele O., Verifiable Credentials Data Model v2.0 URL: https://www.w3.org/TR/vc-data-model-2.0/ [6] Galkin M., Yuan X., Mostsafa H., Tang J., Zhu Z., (2023) Towards Foundation Models for Knowledge Graph Reasoning., URL: https://arxiv.org/pdf/2310.04562.pdf [7] LeCun, Y. (2022) A Path Towards Autonomous Machine Intelligence. New York, June 27., URL: https://openreview.net/pdf?id=BZ5a1r-kVsf 12
프로젝트로 돌아가기