What is PlatON 2.0

Vision ①[7yA]

Combining blockchain and privacy-preserving computation technologies, PlatON is building a decentralized and collaborative AI network and global brain to drive the democratization of AI for safe artificial general intelligence.

• To build the infrastructure needed for autonomous AI agents and their collaboration, to facilitate the emergence and evolution of advanced AI, and to explore the path to general AI.
• Make the power of AI available to anyone through our decentralized network and open-source software tools to make AI technology better for the masses.

The overall goal is achieved in three phases.

• Decentralized privacy-preserving computation network, establishing a decentralized data sharing and privacy-preserving computation infrastructure network that connects data owners, data users, algorithm developers and arithmetic providers.
• A decentralized AI marketplace that enables the common sharing of AI assets, agile smart application development, and provides the whole process of products and services from AI computing power and algorithms to AI capabilities and their production, deployment, and integration.
• A decentralized AI collaboration network that allows AI to collaborate at scale, bringing together collective intelligence to accomplish complex goals.

Privacy-preserving Artificial Intelligence Network

Three−Tier AI network model

​The entire privacy-preserving AI network is divided into three layers.

Layer1: Consensus Network

Consensus network is a decentralized blockchain network composed of blockchain nodes, which are connected to each other through P2P protocol and can be consensual through consensus protocol in an environment where no one needs to be trusted. On the blockchain network, smart contracts can be executed, but due to performance and transaction cost limitations, smart contracts do not support computational logic that cannot be overly complex, and can only access data on the chain and store limited data.

On the blockchain, each participant can have a complete copy of the data and all transaction is open and transparent, so the native blockchain technology does not have the ability to protect privacy. By overlaying privacy-preserving computation protocols based on homomorphic encryption, zero-knowledge proof, TEE and other technologies on the consensus network, the privacy of data and computations on the chain can be protected.

Layer2: Privacy-preserving computation network

The basic elements of computing are data, algorithms, and arithmetic power. Data nodes and computing nodes can be connected to the privacy-preserving computation network through P2P protocols to publish data and arithmetic power, and algorithms can be computed using data and computing power. Through smart contracts on the blockchain, a decentralized sharing and trading market for data, algorithms and computing power can be built. Based on the cryptographic economics on the blockchain, data, computing power and algorithms can be monetized, forming an effective incentive mechanism to motivate more data, algorithms and computing power to join the network.

Privacy-preserving computation networks can privately execute smart contracts of consensus networks and also run popular deep learning frameworks.

The data in privacy-preserving computation networks are generally kept locally and are available invisible through secure multi-party computing, federated learning, and other techniques for collaborative computation. Not only the privacy of the data is protected, but also the privacy of the computation results such as the completed AI models trained.

Layer3: Collaborative AI network

Using the datasets and computing power of privacy-preserving computation networks, AI models can be trained, deployed, and served externally, forming a marketplace for AI services. Through technologies such as Multi Agent System, AI agents can operate independently and communicate and collaborate with each other to create more and more innovative AI services, enabling AI DAO and forming autonomous AI networks.

Technology Stack​

​The technology stack of the privacy-preserving AI network is shown in Figure 6, which is generalized based on the resources and technologies that privacy-preserving AI relies on. Some existing blockchain projects can be mapped to this stack, although certain projects are not a good match.

There are projects that try to combine blockchain, privacy-preserving computation and artificial intelligence, some combine privacy-preserving computation and blockchain to enhance blockchain privacy protection and computing capabilities, some combine blockchain and AI to provide a marketplace for AI services, and some use the decentralization of blockchain to build decentralized cloud computing platforms or data trading marketplaces. Although there seem to be more projects, all of them can only meet part of the needs of privacy protection AI in a fragmented manner and cannot be combined organically, and have not yet formed a mature privacy AI ecology.

Layer1: Consensus Network

Layer1 is the basic protocol of blockchain, the core is consensus and smart contract, Layer1 is the basis of decentralized computing, smart contract is a simple computing model, in a sense it is a kind of Serverless. There are a lot of blockchain projects to implement Layer1 of Ethereum model, such as Eos, Cosmos, Polkadot, Algorand, Dfinity, Solana, Near, etc. Algorand, Dfinity, Solana, Near, etc.

There are three main ways to implement privacy-preserving computation protocols on Layer1: a confidential computing scheme using TEE, typically Oasis, and Enigma has moved from the initial MPC to the TEE camp; a scheme that uses cryptographic techniques such as zero-knowledge proofs and homomorphic encryption for cryptographic computing of blockchain data, such as Monero, Zcash, Manta, and Aztec and Raze Framework based on other Layer1 networks.

Layer2: Privacy-preserving computation network

The amount of data that can be stored on Layer1 is limited, the logic of the smart contract cannot be too complex and does not have access to off-chain data either, so the training of AI models cannot be done in the smart contract.

Privacy-preserving computation network closely combines data, algorithms and computing power to build a complete computing ecology where all subjects, including individuals and institutions, would be financially incentivized to provide personal and professional data. Data security and privacy are guaranteed through through decentralization and secure computing, and subjects are feel more comfortable sharing sensitive data (spending, health information). Over time, the market will accumulate more and higher quality data. Artificial intelligence experts will be motivated to create and share higher performance AI models.

Here we analyze the three key building blocks of decentralized AI: data, models, and computing power.

• Data

Ocean and Computable Labs are working to build data marketplace protocols. snips is using crypto economics to incentivize a network of workers involved in synthetic data generation. gems and Effect are also building decentralized interactive marketplace for data labeling that require human intelligence.

• Computing power

lot of the recent progress in AI has been facilitated by a massive ramp in computing power, that resulted both from better leveraging existing hardware, and also building new high performance hardware specifically for AI (Google TPUs, etc).

DeepBrain aims to share idle computing resources from around the world to enable decentralized arithmetic networks. Its general philosophy is comparable to other projects such as Akash, Golem, but DeepBrain Chain is more specifically focused on the type of computing power needed for AI.

Starkware, zkSync are all zkRollup scaling solutions for scaling payment transactions and smart contract transactions on Ether. LoopRing, Hermez are also zkRollup scaling solutions focused on scaling payment transactions and token transfer transactions.

• Algorithm

For a decentralized computing networks to work, it is important to guarantee that whatever data is provided by individuals and companies is processed in a completely private manner.

Enigma, Phala and OpenMined all provide secure computing solutions, Enigma and Phala target general computing, OpenMinded focuses on privacy-preserving machine learning. Enigma and Phala use TEE techniques, and OpenMinded primarily uses Federated Learning, championed by Google, and Differential Privacy, championed by Apple. The Algorithmia project enables an interactive machine learning model marketplace with the help of blockchain, which is actually a model transaction enabled by smart contracts.

Layer3: Collaborative AI network

The privacy-preserving computation network provides the three key elements needed for AI: data, models and computing power. A decentralized AI marketplace will help create better AI. People provide their data, developers compete to provide the best machine learning models, and the entire system acts as a self-reinforcing network that attracts more and more participants and creates better and better AI.

AI continues to thrive and accelerate through decentralized AI marketplaces. We will have the ability to create many types of AI for almost every task. These AI robots need an effective organizational model to help them cooperate in a transparent manner. fetch works to build and enable Autonomous Economic Agents (AEAs) to cooperate in an organized manner. An AEA is a software entity that can perform actions without external stimuli, and an AEA can intelligently search for and interact with other AEAs. SingularityNET is another very ambitious and complex project that aims to be the leading protocol for networking artificial intelligence and machine learning tools to form efficient applications across vertical markets, ultimately resulting in coordinated artificial general intelligence. The SingularityNET platform currently focuses on providing a commercial launchpad for developers to launch their AI services on the web where they can interoperate with other AI services and paying subscribers. the Botchain project is a system that gives autonomous AI agents to provide identity authentication.

A step further than autonomous AI agent cooperation is that the entire network operates completely autonomously, supported by AI. This is the AI DAO, a decentralized autonomous organization supported by AI, which can be a decentralized organization run entirely by AI, with no or limited human intervention. Many companies in this field have ambitious plans, but are actually in the conceptual stage.

Competitive Landscape

​First of all, PlatON is an underlying public chain, which is not inferior to any mainstream public chains such as Ethereum, Eos, Cosmos, Polkadot, Algorand, Dfinity, Solana, Near, etc. in terms of decentralization, security, performance, and smart contract development.These public chains mainly aim to build WEB3 network infrastructure and decentralized application platform, while PlatON is to build privacy-protected computing network as well as artificial intelligence collaboration network, and the main applications are training and service of artificial intelligence, and autonomous agents.

Compared to other projects with privacy-preserving computation such as Enigma, Oasis and Phala, PlatON focuses on the combination of privacy-preserving computation and AI.

• PlatON supports more complex privacy-preserving computations and even deep learning, and PlatON will also provide specific privacy-preserving computation acceleration hardware for AI.
• PlatON is more focused on privacy-preserving training of AI models and construction of AI agents, as well as interoperability of AI agents, rather than just layer2 computational enhancement of blockchain networks.

Competitive Advantages

PlatON uses a combination of blockchain, privacy computing and artificial intelligence technologies with the following advantages.

Decentralization

Any user and node can connect to the network permissionless. Any data, algorithms and computing power can be securely shared, connected and traded. Anyone can develop and use artificial intelligence applications.

Privacy-preserving

Modern cryptography-based privacy-preserving computation techniques provide a new computing paradigm that makes data and models available but not visible, allowing privacy to be fully protected and data rights to be safeguarded.

High-performance

High-performance asynchronous BFT consensus is achieved through optimization methods such as pipeline verification, parallel verification, and aggregated signatures, and its safety, liveness, and responsiveness are proven using formal verification methods.

Low training costs

With blockchain and privacy-preserving computation technologies, anyone can share data and algorithms in a secure and frictionless marketplace, truly reducing marginal costs and drastically reducing training costs.

Low development threshold

Visualize AI model development and debugging, automated machine learning (AutoML), MLOps simplifies the whole process of managing AI models from model development, training to deployment, reducing the development threshold of AI models and improving development efficiency.

Regulatable and auditable

All data, variables, and processes used in the AI training decision-making process have tamper-evident records that can be tracked and audited. The use of privacy-preserving technologies allows the use of data to satisfy regulatory regulations such as the right to be forgotten, the right to portability, conditional authorization, and minimal collection.