Microsoft Yahei, STHeitiSC-Light, simsun, Times New Roman,Microsoft Yahei, STHeitiSC-Light, simsun, Times New Roman,

The relationship between artificial intelligence and blockchain is like the relationship between computers and the Internet. Computers provide production tools for the Internet, and the Internet realizes information interconnection for computers; artificial intelligence will solve the problems of autonomy, efficiency, and energy saving in blockchain. However, the blockchain will realize general intelligence by sharing isolated and fragmented artificial intelligence. The former is a tool, and the latter is a purpose.

We believe that the respective characteristics and pain points of AI and blockchain determine that the combination of the two is inevitable. A distributed and decentralized blockchain will bring AI a broad and free-flowing data market, AI Module resources and algorithm resources. At the same time, adding AI to the blockchain can make the blockchain more energy-efficient, secure, and efficient, and its smart contracts and autonomous organizations will also become smarter.

Therefore, there are two different ways to combine the two, and the focus of these two ways is also different: one is based on the blockchain, using AI functions to optimize the blockchain (including private chains, alliance chains, etc.) , public chain); the second is based on AI, using the decentralization of the blockchain and the natural attributes of the value network to distribute the deployment of the overall AI system and realize the free flow of data, algorithms, and model resources.

Blockchain and AI are 2 extreme sides of the technology spectrum: on the one hand fostering centralized intelligence in closed data platforms, and on the other hand promoting decentralized applications in open data environments. If we find an intelligent way to make them work together, the total positive externality could be magnified for a split second.

First of all, aiming at the pain points of AI, we will explore the milestone changes made by the blockchain to AI.

When it comes to artificial intelligence, big data cannot be avoided. Artificial intelligence includes three core parts: algorithm, computing power and data. An excellent AI algorithm model requires big data training and sufficient computing power support for continuous optimization and upgrading. The relationship between artificial intelligence and data is like the shuttle and the Jenny machine. In the past, humans have always conceived machine learning methods, but suffered from insufficient data for verification and training. The explosion of the Internet finally ushered in the era of big data, but many of the current data They are all in the hands of central institutions, such as Google, Facebook, BAT, etc., but what AI needs for development, such as: personal consumption records, medical data, educational data, behavioral data, etc., cannot be controlled by individuals at will, and the data market has not yet formed , The result of centralized big data is information islands.

Several main features of the blockchain, such as: consensus system of distributed nodes, immutable information, anonymization, and decentralization. The blockchain also has a very magical calculation method called zero-knowledge proof, which can prove to you that my data is very valuable, but it does not tell you where the real private data is. With the blockchain, the data market can make society more fair, and the incentive mechanism makes data sharing possible, so we use a benign data market, in this market, blockchain and artificial intelligence will achieve The new ideas that coexist with each other finally realize their different values.

Change 1: The distributed AI brought by the blockchain can realize mutual calls between different AI functions and accelerate the development of AI.

Nowadays, every enterprise has a demand for AI to varying degrees, and the current AI products rarely meet the needs of enterprises, and the development of personalized AI products has high technical and financial barriers. Even giants that can hire developers to build custom AI products will struggle to hire enough AI experts to meet all demand. For example, as we are familiar with now, iFlytek focuses on voice intelligence, Megvii is good at image and face recognition, DJI focuses on drones, and Baidu concentrates on developing unmanned driving. Even these industry giants can only focus on artificial intelligence. One field and one direction cannot be comprehensive. But artificial intelligence is a system, and the visual and auditory systems are just a subsystem of this system. If we want to truly embrace the intelligent era, we must break the boundaries between various systems and realize mutual calls between various systems.

image description

image description

(Source: Benyi Capital)

Change 2: Blockchain can break the closed AI development model, share AI resources, and encourage data sharing between traditional silos.

Many advanced AI tools exist only in GitHub (programming community) repositories created by graduate students or independent researchers. This makes them out of reach for anyone to install, configure and run them. Most AI developers are academics, not businessmen, and their algorithms and models are inaccessible to the outside world. The sharing mechanism and incentive mechanism of the blockchain can encourage developers of AI models to share their development results. Both machine learning and deep learning require a sufficiently large data set, and AI personnel cannot create and manage such a large data set. At the same time, the current closed development model also makes it difficult for developers to share data sets.

The decentralized nature of blockchain facilitates data sharing: if no single entity controls the infrastructure that stores data, there is less friction for sharing data. Data sharing can occur within the enterprise (data merging between different branches can reduce the cost of internal audits), alliance databases (data from integrated banks can effectively reduce fraud) or on public blockchains (energy usage + auto parts supply chain data). The debate on data sharing VS data moat: Whether to share data or use data to build barriers, the essence depends on the game between the benefits brought by these two behaviors. To share, there must be enough driving force to exceed the "moat" benefits.

The way of token incentives on the blockchain provides a model of an incentive mechanism for sharing data. If there are enough early benefits, data sharing will become inevitable. When the data from the islands are merged, a new data set can be obtained. When we train on the new data set, it will bring a new data set that can be used for the new A new model for business.

Change 3: Blockchain can also be used to audit and track data and models to obtain more reliable predictions: AI likes data, the more data, the better the model, but the proportional relationship between the amount of data and the AI ​​model is based on a good based on data quality. If you train on garbage data, you will also get a garbage model, and the test data is the same, so the data also needs credibility, and the model trained on valid data is also effective, so that the model also gains reputation and credibility. can also be used more widely.

Change 4: The decentralized data market can reduce the friction caused by data sharing: In the process of decentralized management, data and models are exchanged as intellectual property assets, and no entity can control the data storage infrastructure, which makes It is easier for organizations to work together or share data. With this kind of decentralized exchange, we will see a truly open market for data.

image description

image description

(Source: Benyi Capital)

Compared with the current data islands, the distributed AI platform that will add blockchain technology in the future hopes to achieve the goal of free scheduling of data, algorithms, and AI resources (including development tools, data packages, etc.), and establish a truly free-flowing market.

The value of this platform lies in the construction of the underlying protocol, the docking of data, and AI resources, not just the introduction of resources into the platform. Introducing the platform is only the first step, and then it is more important to realize: these data and resources can be called through the data/AI interface. Further extension is the convenience and speed of calling.

Blockchain is essentially a new filing system for digital information that stores data in an encrypted distributed ledger format. Because data is encrypted and distributed across many different computers, it is possible to create tamper-proof, highly reliable databases that can only be read and updated by authorized users. However, the rapid development of blockchain technology in recent years has exposed many problems, hindering its commercialization process. In summary, there are four main issues: efficiency, security, expansion, and supervision.

Based on the blockchain, we discuss the subversive innovation brought by AI to the blockchain.

Innovation 1: With the optimization of AI algorithm, the consensus mechanism combined with POW and POS can save the power and energy consumption of the blockchain.

Above we mentioned the three core components of artificial intelligence: data, algorithm, and computing power. The optimization of algorithms can save computing power. According to this logic, using artificial intelligence for POW consensus mechanism and hash operation can greatly improve computing efficiency , thereby saving power and energy. For example: the new startup Matrix uses AI to combine POW and POS, adopts a layered consensus mechanism, first uses random clustering algorithms to generate multiple small clusters in the entire node network and selects representative nodes mainly based on the POS mechanism, and then The elected representative nodes conduct POW competition for bookkeeping rights, which can greatly reduce energy waste compared with the full-node competition bookkeeping method.

Innovation 2: AI can introduce a new decentralized learning system to solve the data redundancy problem on the blockchain and expand the system. Decentralized learning systems, such as federated learning and new data sharding techniques, can make the system more efficient. In addition, practice has proved that through the optimization of AI models and algorithms, the natural evolution and dynamic adjustment of the blockchain can also be realized, and the occurrence of forks can also be effectively prevented.

Innovation 3: AI can more effectively manage the autonomous organization of the blockchain: Traditionally, computers cannot complete tasks without clear instructions on how to perform them. Due to the encrypted nature of the blockchain, operating on a computer with blockchain data requires a lot of computer processing power (such as Bitcoin mining). And AI can manage tasks in a smarter, more thoughtful way. Just like an expert who is good at breaking codes can be trained to break codes faster and faster, a machine learning-driven mining algorithm can improve its professional skills almost immediately if it is given the correct training data. If skills are used in community management, then the efficiency of community management will be greatly improved.

image description

image description

(Source: Benyi Capital)

Project Description

Project Description

image description

image description

(Source: Benyi Capital)

SingularityNET's team includes experts in artificial intelligence, big data, and software engineering, as well as a professional marketing team, with a relatively complete personnel structure.

image description

image description

(Source: Internet, Benyi Capital)

image description

image description

(Source: Project White Paper, Benyi Capital)

The SingularityNET project was launched in June 2017. From 2017 to 2018, it will focus on the development of infrastructure and AI tools, especially in terms of language and image processing, biomedical analysis, and finance. At the same time, in December 2017 and July 2018, SingularityNET will release the main network of Alpha and Beta test versions. It is expected that by 2019, the entire system structure, including infrastructure and AI tools, will be perfected, and the SingularityNET 1.0 version will officially start running.

image description

image description

text

AI agents issue AI tasks, exchange data, and provide AI feedback, etc., mainly through a series of smart contracts. In addition, non-AI agents want to obtain AI services from AI agents in the network, which is also realized through smart contracts. At present, in order to cultivate market vitality, SingularityNET does not charge transaction fees.

Strictly speaking, SingularityNET cannot be called an autonomous organization, because the SingularityNET Foundation is responsible for the supervision of the network, but with the improvement of the network, SingularityNET will eventually have the ability of autonomy. The SingularityNET Foundation is made up of 5 organizations: OpenCog Foundation, Hanson Robotics, Vulpem, Novamente LLC, and Bureau of Economic Aviation.

Services provided: At present, there are four main service forms of the SingularityNET network: exchanging hardware and software resources, exchanging AGI tokens for hardware and software resources, matching and alliance between AI agents, and voting on autonomous organizations. The implementation of the first two transactions is to write a certain standard software and hardware API into the smart contract template, and provide it to others by embedding AI tools into the contract template. The smart contracts provided by SingularityNET include: API interfaces for information exchange and terms negotiation between AI agents; exchange services with AGI tokens; democratic autonomy for specific issues.

Figure 8: Four services provided by SingularityNET

(Source: Benyi Capital)

Consensus mechanism: Currently SingularityNET is based on Ethereum, and its consensus mechanism is similar to that of Ethereum. It adopts POW+POS. After leaving Ethereum in the future, SingularityNET will create a POR (Proof of Reputation) reputation consensus mechanism. The mechanism incorporates a number of factors: stake, overall activity in the network, specific rating aspects, length of activity, and ratings above a certain threshold, among others. In addition, this mechanism is similar to the POI (Proof of Importance) consensus mechanism adopted by the Nem blockchain, so the algorithm of the Nem blockchain will be widely used in the future.

Underlying design: The initial common underlying design of SingularityNET was designed by the Foundation member-Economic Space Agency (ECSA). Currently, the company is cooperating with ECSA to develop a tool called Gravity, which provides a highly flexible blockchain economy. The operating system can support various efficient interactions based on smart contracts on various platforms. In addition, Gravity comes with a set of smart contract templates for the design of various types of decentralized blockchain organizations.

The infrastructure of SingularityNET is divided into four layers: the first layer is an open JSON/RPC protocol and its reference implementation for service interoperability; the second layer is that the service description is a shared and extensible ontology, so transaction details can be Negotiate between parties; three layers are: the reputation system tracks service quality and incentivizes better quality; four layers are: micropayment transactions based on distributed ledger technology will have a financial layer, so all parties can pass frictionless Monetize high-quality artificial intelligence services in a new way. This enables different AI methods and developments, such as machine learning and deep learning, ontologies and semantic graphs with probabilistic reasoning, natural language processing, evolutionary and genetic programming, and robotics.

Figure 9: Conception of SingularityNET infrastructure

(Source: Benyi Capital)

Currently SingularityNET can provide: an API and a set of smart contracts for interacting with AI agents; a token-based economy for processing financial transactions with AI agents; a democratic governance mechanism to adjust economic parameters. The current AI modules are provided by Hanson Robotics and OpenCog Foundation, as well as other open source AI modules.

The name of the token issued by SingularityNET is AGI, the total amount is 1 billion, the current circulating supply is about 540 million, the market price is 0.2 US dollars per token, and the market value is about 110 million US dollars. The first ICO was launched in December 2017, based on the Ethereum platform, with a price of 1AGI=0.1USD, and a total financing of 36 million US dollars.

Project Description

Project Description

The Matrix AI Network project was initiated by Professor Deng Yangdong, associate professor of the School of Software, Tsinghua University, China, and has assembled a group of experts and scholars in the fields of artificial intelligence, blockchain, chips and underlying hardware. The team has a strong professional background.

image description

image description

image description

Figure 11: MAN distribution mechanism (source: public information, Benyi Capital)

The Matrix network was officially launched in September 2016, and the preparatory work is still in progress, such as blockchain design, artificial intelligence-based mining machine design, etc. The Matrix mainnet is expected to be launched before March 2019, and the MAN coins that have been sold through ICO will be migrated to the Matrix mainnet at a ratio of 1:1. By the beginning of 2020, the Matrix main network ecology, including mining, AI application development, and related infrastructure construction, will be gradually improved, and the ecological chain will basically be formed.

image description

image description

(Source: Matrix white paper, Benyi Capital)

As mentioned earlier, Matrix has four major advantages over existing blockchains: easier to use, safer, faster, and more flexible. These advantages are basically due to the architectural innovation of Matrix blockchain. In general, the Matrix network has four major innovation points: hierarchical consensus mechanism, multi-functional heterogeneous nodes, multi-chain blockchain structure, and dynamic optimization of protocol parameters.

Layered Consensus Mechanism

Matrix does not directly adopt consensus algorithms such as POW, POS, or the more popular DPOS and PBFT, nor does it adopt the mixed consensus algorithm of early POW to late POS that is often used in some ICO projects. Instead, it combines POW and POS in an original way. And the traditional Hash operation is transformed into the cornerstone of its artificial intelligence system - Monte Carlo Markov calculation (MCMC) required by Bayesian decision theory, so that miners can generate actual value in the process of mining.

image description

image description

(Source: Benyi Capital)

Multi-chain blockchain structure

Compared with the leading basic public chain, the Matrix network adopts the design idea of ​​separation of control and data, and has a multi-chain parallel innovative structure. The basic structure consists of a distributed control chain and a data chain. The distributed control chain is mainly composed of block headers, control chain parameters, and AI model lists. The block mainly contains the parameter configuration of each data chain. The AI ​​version number is used to obtain AI parameters and models. Multiple data chains can be defined in the Matrix network and their parameter models can be mapped to the control chain block.

In addition, except for the control chain and the first data chain that need to follow Matrix’s POS/POW hybrid consensus mechanism, the consensus mechanism of subsequent new data chains can be set by yourself. Currently, Matrix supports almost all mainstream consensus algorithms. With the further increase of AI services and attached data chains provided by its blockchain, blocks in each data chain can realize data interaction and cross-chain transactions through control chains and smart contracts, and eventually Matrix will evolve into a blockchain network.

Dynamic optimization of protocol parameters

Based on the design concept of the separation of Matrix control chain and data chain, Matrix blockchain has the ability of natural evolution and dynamic adjustment, which is its biggest innovation compared with the current blockchain. Matrix establishes a Bayesian network model for real-time statistics and analysis of various AI parameters, and then optimizes parameters through standard optimization algorithms. The current optimization model adopted by the Matrix blockchain is a Bayesian enhanced learning algorithm.

multifunctional heterogeneous node

The Matrix system model includes six types of system equipment: standard Matrix nodes, cloud access nodes, cloud storage nodes, Matrix trusted gateways, external data storage pools, and AI service equipment.

A standard Matrix node runs a Matrix distributed control chain and at least one data chain, and has the ability to run an AI virtual machine, execute the AI ​​model defined by the control chain and data chain, and run smart contracts.

The cloud access node is a service node deployed by Matrix in the cloud, which mainly provides convenient access methods for users without Matrix nodes, especially mobile devices. Users only need to provide access authentication and connect to the cloud access node after relevant security encryption operations to obtain the AI ​​services provided by Matrix.

image description

image description

(Source: Benyi Capital)

Matrix trusted gateway is an important device for Matrix network to interact with external data. Based on AI intelligent audit and node voting, Matrix can give credit ratings for various external data sources. The gateway helps improve the scalability of the Matrix network, making it safer and more convenient for smart contracts running to obtain external data.

The external data storage pool mainly stores the external data acquired through the trusted gateway and processes it with security encryption, which can only be accessed by specific users. Matrix will deploy multiple distributed storage pools, and only when the data content of the storage pools reaches a consensus, can it be written into a block.

The AI ​​service device mainly includes two functions. The first function is to help support the parameter optimization of the entire Matrix system and various AI models in it; the second function is to provide services for external users using the Matrix network as a whole, or to call Matrix nodes serve internal users. AI service equipment needs hardware support to provide the required computing power.

Potential risks of Matrix

Time risk: The main network of Matrix is ​​not yet online, and the ICO is also carried out through Ethereum smart contracts, so it is still unknown whether the various architectural innovations mentioned in its white paper can run smoothly. Although various optimization algorithms based on artificial intelligence, intelligent review and other mechanisms can be realized in theory, the actual user experience needs to be realized after the project is implemented, which brings a great time risk. After all, Matrix is ​​still in concept stage.

Technical risk: mainly refers to the cornerstone of the Matrix artificial intelligence system - Bayesian theory. If it is abandoned in the next few years or replaced by other more advanced artificial intelligence theories, then the entire basic system of Matrix will face collapse, or It requires large-scale adjustments and updates. This is an extremely heavy blow to the future development of Matrix.

Project Description

Founded in November 2017 and released a white paper in February 2018, Ocean Protocol is a decentralized data exchange network that allows people to share data and monetize it for training AI models. Ocean Protocol is an ecosystem for sharing data and related services, providing a tokenized service layer, exposing data, storage, computing and algorithms for consumption, helping to unlock data, especially AI data, and using blockchain technology , allowing data to be shared and sold in a secure and transparent manner.

The industry pain point that Ocean Protocol solves is the data islands caused by lack of trust. The goal of Ocean Protocol is to build a market that supports the free flow of multiple data, and locks more AI functions. For example, allowing artificial intelligence researchers to log data Ten computer vision applications for image classifier training.

The team members of Ocean Protocol basically have continuous entrepreneurial experience, most of them come from Big Chain DB, and basically have background knowledge of big data, blockchain, artificial intelligence and data exchange, as well as experience of designers and technicians. The main team members are as follows:

image description

image description

(Source: Benyi Capital)

Token distribution: Network maintenance and data providers will allocate 45% of the total tokens, 20% for the founding team, 10% for community and ecosystem development, and the last 25% for public sale, which will be sold in three installments. The first time in October 2017, the seed round has raised 5 million US dollars and sold 3.6% of the total number of tokens; the second time in March 2018, the pre-launch crowdfunding was 2,200 US dollars and sold 6.4% of the total number of tokens ; The third time is expected to be in Q1 of 2019, releasing 5.1%; if the project needs funds to expand the production scale within 5 years, crowdfunding will be launched again to release the last 10%.

The ecosystem of Ocean protocol can be divided into three layers from top to bottom: user layer, service layer, and base layer. The user layer mainly includes data consumers, who are the demand side of the data and service market provided by Ocean; the service layer mainly includes data/service providers, data managers/referrers, and data verifiers, and is where users directly connect to obtain data and services Platform, the base layer mainly includes Ocean's underlying blockchain maintainers, providing infrastructure support for its service platform.

image description

image description

(Source: Ocean protocol white paper, Benyi Capital)

Due to the difficult pricing of data resources, Ocean protocol has designed a unique economic incentive mechanism for its ecosystem to encourage users to participate and make contributions. Ocean has designed a betting method to define the rewards that data providers deserve. Specifically, users and other participants who provide data can bet on the data to express their expectations for the popularity of the data. The reward amount is equal to that of the user The product of the bet amount and the logarithm of the data's popularity (the number of uses within a certain period of time) is multiplied by the real data usage ratio (to prevent malicious swiping points), expressed in a formula: . Since the formula needs to repeatedly calculate the bet amount of each participant for each data set or service in each reward cycle, the calculation process is very complicated and consumes high calculation costs.

Ocean draws on the mining model of Bitcoin and modifies it so that whenever a node submits an available data set or service, the Ocean protocol will randomly choose whether to reward it, and its selection probability is equal to the reward amount mentioned above divided by The total reward amount of the whole network. The occurrence time of the reward is set to occur once every minute on average.

image description

image description

(Source: Ocean protocol white paper, Benyi Capital)