Web3-AI Track Panorama Report: Technical Logic, Scenario Applications, and In-Depth Analysis of Top Projects

With the continued rise of AI narratives, more and more attention is focused on this track. An in-depth analysis of the technical logic, application scenarios, and representative projects of the Web3-AI track has been conducted to comprehensively present the panorama and development trends of this field.

1. Web3-AI: Analysis of Technical Logic and Emerging Market Opportunities

1.1 The Fusion Logic of Web3 and AI: How to Define the Web-AI Track

In the past year, AI narratives have been exceptionally popular in the Web3 industry, with AI projects emerging like mushrooms after rain. Although many projects involve AI technology, some projects only use AI in certain parts of their products, and the underlying token economics are not substantially related to the AI products. Therefore, such projects are not included in the discussion of Web3-AI projects in this article.

The focus of this article is on using blockchain to solve production relationship issues and AI to address productivity problems. These projects provide AI products while utilizing Web3 economic models as tools for production relationships, with both aspects complementing each other. We categorize these types of projects as the Web3-AI track. To help readers better understand the Web3-AI track, we will elaborate on the development process and challenges of AI, as well as how the combination of Web3 and AI can perfectly solve problems and create new application scenarios.

1.2 The Development Process and Challenges of AI: From Data Collection to Model Inference

AI technology is a field that enables computers to simulate, extend, and enhance human intelligence. It allows computers to perform a variety of complex tasks, from language translation and image classification to facial recognition and autonomous driving applications. AI is changing the way we live and work.

The process of developing artificial intelligence models typically includes the following key steps: data collection and data preprocessing, model selection and tuning, model training and inference. For a simple example, to develop a model for classifying images of cats and dogs, you need to:

1. Data collection and data preprocessing: Collect an image dataset containing cats and dogs, which can be done using public datasets or by collecting real data yourself. Then label each image with a category (cat or dog), ensuring that the labels are accurate. Convert the images into a format that the model can recognize and divide the dataset into training, validation, and testing sets.

2. Model Selection and Tuning: Choose the appropriate model, such as Convolutional Neural Networks (CNN), which are well-suited for image classification tasks. Adjust the model parameters or architecture according to different needs. Generally, the network depth of the model can be adjusted based on the complexity of the AI task. In this simple classification example, a shallower network may be sufficient.

3. Model Training: You can use GPU, TPU, or high-performance computing clusters to train the model, and the training time is affected by the complexity of the model and the computing power.

4. Model Inference: The files of a trained model are usually referred to as model weights. The inference process refers to the process of using an already trained model to predict or classify new data. During this process, a test set or new data can be used to evaluate the model's classification performance, and the model's effectiveness is typically assessed using metrics such as accuracy, recall, and F1-score.

As shown in the figure, after data collection and preprocessing, model selection and tuning, and training, the trained model will be used for inference on the test set to obtain the predicted values P (probability) for cats and dogs, which is the probability that the model infers it is a cat or a dog.

Trained AI models can further be integrated into various applications to perform different tasks. In this example, a cat-dog classification AI model can be integrated into a mobile application, where users upload pictures of cats or dogs to receive classification results.

However, the centralized AI development process has some issues in the following scenarios:

User Privacy: In centralized scenarios, the development process of AI is often opaque. User data may be stolen and used for AI training without their knowledge.

Data Source Acquisition: Small teams or individuals may face restrictions on data not being open source when acquiring data in specific fields (such as medical data).

Model selection and tuning: It is difficult for small teams to access model resources in specific domains or spend a lot of money on model tuning.

Acquiring computing power: For individual developers and small teams, the high costs of purchasing GPUs and renting cloud computing power can pose a significant financial burden.

AI Asset Income: Data annotators often struggle to earn income that matches their efforts, while AI developers find it difficult to match their research outcomes with buyers in need.

The challenges existing in centralized AI scenarios can be addressed by integrating with Web3. As a new type of productive relationship, Web3 is inherently compatible with AI, which represents a new productive force, thereby promoting simultaneous progress in technology and production capabilities.

1.3 The Synergistic Effect of Web3 and AI: Role Transformation and Innovative Applications

The combination of Web3 and AI can enhance user sovereignty, providing users with an open AI collaboration platform, enabling them to transition from AI users in the Web2 era to participants, creating AI that everyone can own. At the same time, the integration of the Web3 world and AI technology can spark more innovative application scenarios and gameplay.

Based on Web3 technology, the development and application of AI will usher in a brand new collaborative economic system. People's data privacy can be guaranteed, the data crowdsourcing model promotes the advancement of AI models, and numerous open-source AI resources are available for users to use, while shared computing power can be obtained at a lower cost. With the help of a decentralized collaborative crowdsourcing mechanism and an open AI market, a fair income distribution system can be realized, thereby encouraging more people to drive the advancement of AI technology.

In the Web3 scenario, AI can have a positive impact across multiple tracks. For example, AI models can be integrated into smart contracts to enhance work efficiency in various application scenarios, such as market analysis, security detection, social clustering, and more. Generative AI not only allows users to experience the "artist" role, such as creating their own NFTs using AI technology, but also can create diverse game scenarios and interesting interactive experiences in GameFi. Rich infrastructure provides a smooth development experience, allowing both AI experts and newcomers looking to enter the AI field to find suitable entry points in this world.

2. Interpretation of the Web3-AI Ecosystem Project Layout and Architecture

We mainly studied 41 projects in the Web3-AI track and categorized these projects into different tiers. The division logic of each tier is shown in the figure below, including the infrastructure layer, intermediate layer, and application layer, each of which is further divided into different sections. In the next chapter, we will conduct a depth analysis of some representative projects.

The infrastructure layer encompasses the computing resources and technological architecture that support the entire AI lifecycle, the middle layer includes data management, model development, and verification inference services that connect the infrastructure with applications, while the application layer focuses on various applications and solutions directly aimed at users.

Infrastructure Layer:

The infrastructure layer is the foundation of the AI lifecycle. This article classifies computing power, AI Chain, and development platforms as the infrastructure layer. It is the support of these infrastructures that enables the training and inference of AI models, presenting powerful and practical AI applications to users.

• Decentralized computing network: It can provide distributed computing power for AI model training, ensuring efficient and economical utilization of computing resources. Some projects offer a decentralized computing power market where users can rent computing power at low costs or share computing power to earn profits, represented by projects such as IO.NET and Hyperbolic. In addition, some projects have derived new gameplay, such as Compute Labs, which proposed a tokenized protocol where users can participate in computing power rentals in different ways by purchasing NFTs that represent GPU entities.

• AI Chain: Utilizing blockchain as the foundation for the AI lifecycle, enabling seamless interaction of AI resources on and off the chain, and promoting the development of industry ecosystems. The decentralized AI market on the chain can trade AI assets such as data, models, agents, etc., and provides AI development frameworks and supporting development tools, with representative projects like Sahara AI. AI Chain can also promote advancements in AI technologies across different fields, such as Bittensor, which fosters competition among different AI types through an innovative subnet incentive mechanism.

• Development Platforms: Some projects offer AI agent development platforms, which can also facilitate trading by AI agents, such as Fetch.ai and ChainML. One-stop tools help developers to more conveniently create, train, and deploy AI models, represented by projects like Nimble. This infrastructure promotes the widespread application of AI technology in the Web3 ecosystem.

Middleware:

This layer involves AI data, models, as well as reasoning and validation, achieving higher work efficiency through Web3 technology.

• Data: The quality and quantity of data are key factors affecting the effectiveness of model training. In the Web3 world, resource utilization can be optimized and data costs reduced through crowdsourced data and collaborative data processing. Users can have autonomy over their data, selling it under privacy protection to avoid it being stolen by unscrupulous merchants for high profits. For data demanders, these platforms offer a wide range of options at very low costs. Representative projects such as Grass leverage user bandwidth to scrape web data, while xData collects media information through user-friendly plugins and supports users in uploading tweet information.

In addition, some platforms allow domain experts or ordinary users to perform data preprocessing tasks, such as image annotation and data classification. These tasks may require specialized knowledge for the processing of financial and legal data. Users can tokenize their skills to achieve collaborative crowdsourcing for data preprocessing. An example is the AI marketplace of Sahara AI, which covers data tasks in various fields and can address multi-domain data scenarios; while AIT Protocol annotates data through human-computer collaboration.

• Model: In the AI development process mentioned earlier, different types of requirements need to match suitable models. Common models for image tasks include CNN and GAN, while for object detection tasks, the Yolo series can be selected. For text-related tasks, common models include RNN and Transformer, as well as some specific or general large models. The depth of the models required for tasks of varying complexity also differs, and sometimes model tuning is necessary.

Some projects support users to provide different types of models or collaborate on model training through crowdsourcing, such as Sentient, which allows users to place trusted model data in the storage layer and distribution layer for model optimization through modular design. The development tools provided by Sahara AI come with advanced AI algorithms and computing frameworks and have the capability for collaborative training.

• Inference and Verification: After the model is trained, it generates model weight files that can be used for direct classification, prediction, or other specific tasks; this process is called inference. The inference process is usually accompanied by a verification mechanism to confirm whether the source of the inference model is correct and to check for malicious behavior, etc. In Web3, inference can typically be integrated into smart contracts, allowing inference to be performed by calling the model. Common verification methods include technologies such as ZKML, OPML, and TEE. Representative projects like the ORA on-chain AI oracle (OAO) have introduced OPML as a verifiable layer for AI oracles, and their official website also mentions their research on ZKML and opp/ai (ZKML combined with OPML).

Application Layer:

This layer mainly consists of applications directly aimed at users, combining AI with Web3 to create more interesting experiences.