When Huang Renxun gave a speech at WGS in Dubai, he proposed the term "sovereign AI". So, which sovereign AI can meet the interests and demands of the Crypto community?

Maybe it needs to be built in the form of Web3+AI.

In his article “The promise and challenges of crypto + AI applications”, Vitalik described the synergy between AI and Crypto: Crypto’s decentralization can balance AI’s centralization; AI is opaque, Crypto brings transparency; AI requires data, blockchain facilitates data storage and tracking. This synergy runs through the entire industrial landscape of Web3+AI.

Most Web3 + AI projects are using blockchain technology to solve the construction problems of AI industry infrastructure projects, and a few projects are using AI to solve certain problems of Web3 applications.

The Web3 + AI industry landscape is roughly as follows:

The production and workflow of AI is roughly as follows:

In these links, the combination of Web3 and AI is mainly reflected in four aspects:


Computing power layer: computing power assetization

In the past two years, the computing power used to train large AI models has grown exponentially, doubling almost every quarter, and growing at a rate far exceeding Moore's Law. This situation has led to a long-term imbalance in the supply and demand of AI computing power, and the prices of hardware such as GPUs have risen rapidly, which in turn has raised the cost of computing power.

But at the same time, there is also a large amount of idle mid- to low-end computing power hardware on the market. It may be that the single computing power of this part of mid- to low-end hardware cannot meet high-performance needs. However, if a distributed computing power network is built through Web3 and a decentralized computing resource network is created through computing power leasing and sharing, it can still meet the needs of many AI applications. Since distributed idle computing power is used, the cost of AI computing power can be significantly reduced.

The computing power layer segmentation includes:

  • General decentralized computing power (e.g. Arkash, Io.net, etc.);

  • Decentralized computing power for AI training (e.g. Gensyn, Flock.io, etc.);

  • Decentralized computing power for AI reasoning (e.g. Fetch.ai, Hyperbolic, etc.);

  • Decentralized computing power for 3D rendering (e.g. The Render Network, etc.).

The core advantage of Web3+AI computing power assetization lies in decentralized computing power projects. Combined with token incentives, it is easy to expand the network scale. In addition, its computing resource cost is low and cost-effective, which can meet some low- and mid-end computing power needs.

Data layer: data assetization

Data is the oil and blood of AI. Without Web3, only giant companies would have a large amount of user data. It would be difficult for ordinary startups to obtain extensive data, and the value of user data in the AI ​​industry would not be fed back to users. Through Web3+AI, processes such as data collection, data labeling, and data distributed storage can be made more cost-effective, transparent, and user-friendly.

Collecting high-quality data is a prerequisite for AI model training. Through Web3, we can utilize distributed networks, combine appropriate token incentive mechanisms, and adopt crowdsourcing collection methods to obtain high-quality and extensive data at a lower cost.

According to the purpose of the project, data projects mainly include the following categories:

  • Data collection projects (e.g. Grass, etc.);

  • Data trading projects (such as Ocean Protocol, etc.);

  • Data annotation projects (such as Taida, Alaya, etc.);

  • Blockchain data source projects (such as Spice AI, Space and time, etc.);

  • Decentralized storage projects (such as Filecoin, Arweave, etc.).

Data-based Web3+AI projects are more challenging in the process of designing token economic models because data is more difficult to standardize than computing power.

Platform layer: platform value assetization

Most platform projects will benchmark Hugging Face, with the integration of various resources in the AI ​​industry as the core. Establish a platform to aggregate various resources and roles such as data, computing power, models, AI developers, blockchain, etc., and solve various needs more conveniently with the platform as the center. For example, Giza focuses on building a comprehensive zkML operation platform, aiming to make machine learning reasoning credible and transparent, because data and model black boxes are common problems in AI at present. It will sooner or later be called for by the industry to use cryptographic technologies such as ZK and FHE through Web3 to verify that the reasoning of the model is indeed correctly executed.

There are also layer1/layer2 of Focus AI, such as Nuroblocks, Janction, etc. The core narrative is to connect various computing power, data, models, AI developers, nodes and other resources, and help Web3+AI applications to achieve rapid construction and development by packaging common components and common SDKs.

There are also Agent Network-type platforms, based on which AI Agents can be built for various application scenarios, such as Olas, ChainML, etc.

Web3+AI projects of the platform type mainly use tokens to capture the value of the platform and motivate all participants to build the platform together. This is helpful for start-up projects from 0 to 1, and can reduce the difficulty for project parties to find partners such as computing power, data, AI developer communities, and nodes.

Application layer: AI value assetization

Most of the infrastructure projects mentioned above use blockchain technology to solve the problems of infrastructure project construction in the AI ​​industry. Application layer projects use AI to solve the problems of Web3 applications.

For example, Vitalik mentioned two directions in the article, which I think are very meaningful.

First, AI as a Web3 participant. For example, in Web3 Games, AI can act as a game player, which can quickly understand the rules of the game and complete the game tasks most efficiently. In DEX, AI has played a role in arbitrage transactions for many years. In prediction markets, AI Agent can widely accept a large amount of data, knowledge base and information, train its model's analytical prediction capabilities, and provide it to users in a productized manner, helping users to make predictions about specific events in a model-based reasoning manner, such as sports events, presidential elections, etc.

The second is to create a scalable, decentralized, private AI. Because many users are concerned about the black box problem of AI, the bias in the system, or the possibility that some dApps will use AI technology to deceive users and make profits. This is essentially because users do not have the authority to review and govern the AI ​​model training and reasoning process. However, if a Web3 AI is created, like the Web3 project, the community has distributed governance rights over this AI, which may be more easily accepted.

So far, there has been no white horse project with a high ceiling in the Web3+AI application layer.

Summarize

Web3 + AI is still in its early stages, and the industry is divided on the development prospects of this track. We will continue to pay attention to this track. We hope that the combination of Web3 and AI can create products that are more valuable than centralized AI, allowing AI to get rid of the labels of "control by giants" and "monopoly" and "co-govern AI" in a more community-oriented way. Perhaps in the process of closer participation and governance, humans will have more "respect" and less "fear" of AI.