Current Status of the Integration of AI and Web3: Opportunities and Challenges Coexist

Introduction: The Development of AI + Web3

In recent years, the rapid development of artificial intelligence ( AI ) and Web3 technologies has attracted widespread attention globally. AI, as a technology that simulates human intelligence, has made significant breakthroughs in areas such as facial recognition, natural language processing, and machine learning, bringing tremendous transformation and innovation to various industries. In 2023, the market size of the AI industry reached $200 billion, giving rise to excellent companies like OpenAI, Character.AI, and Midjourney.

At the same time, Web3 as an emerging network model is changing people's perception and usage of the internet. Based on decentralized blockchain technology, Web3 achieves data sharing and control, user autonomy, and the establishment of trust mechanisms through functions such as smart contracts, distributed storage, and decentralized identity verification. The core concept of Web3 is to liberate data from centralized institutions, giving users control over their data and the right to share its value. Currently, the market value of the Web3 industry has reached $25 trillion, with projects like Bitcoin, Ethereum, and Solana emerging one after another.

The combination of AI and Web3 is an area of great interest to builders and investors from both the East and the West. How to effectively integrate the two is a question worth exploring. This article will focus on the current development status of AI+Web3, analyze the situation of current projects, and discuss the limitations and challenges faced, providing references and insights for investors and industry practitioners.

2. The ways AI interacts with Web3

The development of AI and Web3 is like the two sides of a balance scale; AI enhances productivity, while Web3 brings about a transformation in production relationships. Next, we will first analyze the dilemmas and areas for improvement faced by the AI and Web3 industries, and then explore how they can help each other solve these dilemmas.

2.1 Challenges Faced by the AI Industry

The core of the AI industry relies on three elements: computing power, algorithms, and data.

1. Computing Power: Refers to the ability to perform large-scale calculations and processing. AI tasks often require handling vast amounts of data and performing complex computations, such as training deep neural network models. High-intensity computing power can accelerate model training and inference processes, enhancing the performance and efficiency of AI systems. In recent years, with the development of GPUs and dedicated AI chips, the enhancement of computing power has played a significant role in driving the development of the AI industry.

2. Algorithm: It is a core component of AI systems, used for solving problems and implementing tasks through mathematical and statistical methods. AI algorithms can be divided into traditional machine learning algorithms and deep learning algorithms, with significant breakthroughs achieved in deep learning algorithms in recent years. The selection and design of algorithms are crucial to the performance and effectiveness of AI systems. Continuously improving and innovating algorithms can enhance the accuracy, robustness, and generalization capability of AI systems.

3. Data: The core task of AI systems is to extract patterns and规律 from data through learning and training. Data is the foundation for training and optimizing models; through large-scale data samples, AI systems can learn more accurate and intelligent models. Rich datasets can provide more comprehensive and diverse information, enabling models to generalize better to unseen data, helping AI systems to better understand and solve real-world problems.

The main dilemmas faced by AI in these three aspects include:

• Regarding computing power: Acquiring and managing large-scale computing power is expensive and complex. The costs, energy consumption, and maintenance of high-performance computing devices are all issues. For startups and individual developers, obtaining sufficient computing power can be challenging.

• In terms of algorithms: Deep learning algorithms require a large amount of data and computational resources, and they lack model interpretability and explainability. The robustness and generalization ability of the algorithms are also important issues, as models may perform unreliably on unseen data.

• Data aspect: Obtaining high-quality and diverse data remains a challenge. Data in certain fields may be difficult to obtain, such as healthcare data. Data quality, accuracy, and labeling are also issues; incomplete or biased data may lead to erroneous behaviors or biases in models. At the same time, protecting data privacy and security is also an important consideration.

In addition, the issues of AI model interpretability and transparency, as well as unclear business models, have also left many AI entrepreneurs feeling confused.

2.2 Challenges Facing the Web3 Industry

The Web3 industry currently faces many challenges that need to be addressed, including areas such as data analysis, user experience, and smart contract security, all of which have room for improvement. AI, as a tool to enhance productivity, also has a lot of potential to play a role in these areas.

• Data Analysis and Prediction: AI technology can help Web3 platforms extract valuable information from massive amounts of data, enabling more accurate predictions and decisions, which is of significant importance for risk assessment, market predictions, and asset management in the DeFi field.

• User Experience and Personalized Services: AI can help Web3 platforms provide better user experiences and personalized services by analyzing user data to offer personalized recommendations, customized services, and intelligent interactive experiences, thereby increasing user engagement and satisfaction.

• Security and Privacy Protection: AI can be used to detect and defend against cyber attacks, identify anomalous behaviors, and provide stronger security guarantees. At the same time, AI can also be applied to data privacy protection, safeguarding user information through technologies such as encryption and privacy computing.

• Smart Contract Auditing: AI technology can be used for automating contract auditing and vulnerability detection, enhancing the security and reliability of contracts.

3. Analysis of the Current Status of AI+Web3 Projects

Projects that combine AI and Web3 primarily approach from two aspects: leveraging blockchain technology to enhance the performance of AI projects, and utilizing AI technology to serve the advancement of Web3 projects.

3.1 Web3 Empowering AI

3.1.1 Decentralized Computing Power

With the emergence of large models like ChatGPT, the demand for computing power in the AI field has surged. However, the shortage of GPU supply has become a bottleneck restraining AI development. To address this issue, some Web3 projects are attempting to provide decentralized computing power services, including Akash, Render, and Gensyn. These projects incentivize global users to offer idle GPU computing power through tokens, providing computing support for AI clients.

The supply side mainly includes cloud service providers, cryptocurrency miners, and large enterprises. Decentralized computing projects can be roughly divided into two categories: one for AI inference ( such as Render, Akash ), and the other for AI training ( such as io.net, Gensyn ).

Taking io.net as an example, as a decentralized computing network, it currently has over 500,000 GPUs and integrates the computing power of Render and Filecoin, continuously developing ecological projects. Gensyn promotes the allocation and rewards of machine learning tasks through smart contracts, enabling AI training.

However, most projects choose to perform AI inference rather than training, mainly due to the different requirements for computing power and bandwidth. AI training requires a large amount of data and high-speed communication bandwidth, making it more difficult to implement. In contrast, AI inference has smaller data and bandwidth requirements, making it more feasible to implement.

3.1.2 Decentralized Algorithm Model

In addition to computing power, some projects are trying to establish decentralized AI algorithm service markets. Taking Bittensor as an example, it links multiple AI models, each with its own expertise and skills. When users ask questions, the market selects the most suitable AI model to provide answers.

In the Bittensor network, algorithm model providers ( miners ) contribute machine learning models to the network and receive token rewards for their contributions. To ensure the quality of answers, Bittensor uses a unique consensus mechanism to ensure the network reaches consensus on the best answers.

The development of decentralized algorithm model platforms may enable small companies to compete with large organizations in the use of top AI tools, potentially having a significant impact across various industries.

3.1.3 Decentralized Data Collection

AI model training requires a large amount of data, but currently most Web2 platforms prohibit data collection for AI training, or sell user data to AI companies without sharing profits. Some Web3 projects achieve decentralized data collection through token incentives, such as PublicAI.

In PublicAI, users can participate as AI data providers or data validators. Data providers find valuable content on social platforms and share it with the PublicAI data center; data validators vote for the most valuable data for AI training. Users earn token incentives through these two types of contributions, promoting a win-win relationship between data contributors and the AI industry development.

3.1.4 ZK Protection of User Privacy in AI

Zero-knowledge proof technology can achieve information verification while protecting privacy, helping to resolve the conflict between data privacy protection and data sharing in AI. ZKML( Zero-Knowledge Machine Learning ) allows for the training and inference of machine learning models using zero-knowledge proof technology without disclosing the original data.

Projects like BasedAI are exploring the seamless integration of FHE( fully homomorphic encryption) with LLM to maintain data confidentiality. By embedding privacy into distributed network infrastructure through zero-knowledge large language models( ZK-LLM), user data is kept private throughout the operation of the network.

3.2 AI Empowering Web3

3.2.1 Data Analysis and Forecasting

Many Web3 projects have begun to integrate AI services or develop their own AI tools to provide users with data analysis and forecasting services, covering areas such as investment strategies, on-chain analysis, price and market predictions.

For example, Pond uses AI algorithms to predict valuable alpha tokens in the future, providing investment assistance suggestions for users and institutions. BullBear AI is trained based on user historical data, price line history, and market trends to help forecast price movements. Numerai, as an investment competition platform, allows participants to use AI and large language models to predict the stock market. On-chain data analysis platforms like Arkham also combine AI to provide services, matching blockchain addresses with real-world entities to showcase key data and analysis.

3.2.2 Personalized Services

Web3 projects are optimizing user experience by integrating AI. For example, the data analytics platform Dune has launched the Wand tool, which uses large language models to write SQL queries, allowing users who do not understand SQL to easily search. The Web3 media platform Followin and the Web3 encyclopedia IQ.wiki integrate ChatGPT for content summarization. The LLM-based search engine Kaito aims to become a Web3 search platform. Projects like NFPrompt are reducing the cost of user NFT creation through AI.

3.2.3 AI Audit Smart Contracts

AI plays an important role in smart contract auditing, efficiently and accurately identifying code vulnerabilities. For example, 0x0.ai provides an AI smart contract auditor that uses advanced algorithms to analyze smart contracts and identify potential vulnerabilities or security risks. Auditors utilize machine learning techniques to identify patterns and anomalies in the code, marking potential issues for further review.

4. Limitations and Challenges of AI+Web3 Projects

4.1 Real obstacles in decentralized computing power

Decentralized computing power products face some real issues:

1. Performance and Stability: Due to reliance on nodes distributed globally, network connections may experience delays and instability, and performance may be inferior to centralized computing products.

2. Resource matching: Availability is affected by the degree of supply and demand matching, which may lead to insufficient resources or an inability to meet user needs.

3. Technical complexity: Users may need to understand knowledge such as distributed networks, smart contracts, and cryptocurrency payments, which can be costly to use.

4. Difficulty in training large models: Training large models requires extremely high stability and multi-GPU parallel capabilities, which are currently difficult to achieve with decentralized computing power. The main reasons include:

   • Single card computing power: Large model training requires strong single card computing power.
   • Multi-GPU Parallelism: Requires mobilizing tens of thousands of GPUs for parallel training, with very high demands on multi-GPU communication.
   • Software Ecosystem: Requires a software environment that is compatible with hardware, such as NVIDIA's CUDA system.