The Future of AI+Web3 (I): Market Outlook and Narrative Logic
The emergence of generative AI models like ChatGPT has transformed AI from a basic automation tool to a sophisticated decision-making and prediction system, making it a driving force behind progress in modern society. AI has become a highly competitive field, and with the Web3 market entering a new bull market phase, the convergence of AI and Web3 represents the collision of two of the most popular technological trends of the moment. So AI+Web3 = ?
In general, an AI production process can be divided into several stages: data acquisition, data preprocessing, feature engineering/prompt engineering, model training and tuning, model review and governance, model inference, and model deployment and monitoring.
Web3 technology can be combined with the above processes to address some of the challenges in AI development, such as model transparency, bias, and ethical applications. Cryptography techniques like ZK can improve trust issues in AI. Furthermore, the growing need for AI applications necessitates more affordable and accessible infrastructures and data networks. Web3's distributed networks and incentive models can facilitate the development of open-source AI networks and communities.
By combining the AI production process with the direction of AI and Web3 integration, we have identified the three layers of the AI+Web3 industry chain: infrastructure, middle, and application layers. This categorization is based on an analysis of current mainstream AI+Web3 projects in the market.
Infrastructure Layer
The computing and storage infrastructure is a crucial component of the AI work production process. It provides the necessary arithmetic power for AI model training and speculation, as well as storage for data and models throughout their life cycle.
Decentralized AI infrastructure is currently a popular trend,
with projects such as Render Network, Akash, gensyn, Filecoin, and Arweave leading the way.
Middle Layer
It refers to the use of Web3 technologies to improve specific processes in AI work production.
1.Data acquisition stage, where decentralized data identity is adopted to create a more open data network and trading platform.
2. Data preprocessing stage: we focus on creating distributed AI data annotation and processing platforms. We also adopt economic incentives to encourage crowdsourcing patterns, which promote more efficient and cost-effective data preprocessing to serve the subsequent model training phase.
3. Model validation and inference stage: Web3 can address the real-life problem of data and model black-boxing in AI by incorporating cryptographic techniques such as ZK and homomorphic encryption. These techniques can be used to validate the model's inference, ensuring its correctness while protecting the privacy of the input data and parameters.
Application Layer
More about ‘How’. Web3 applications combined with AI technology can effectively improve efficiency and product experience. For example, AI's content generation, analysis, speculation, and other functions can be applied to various fields such as gaming, social networking, data analysis, and financial forecasting. The current AI+Web3 applications can be divided into three main categories.
- AIGC: enable users to generate various types of content, including text, images, videos, and avatars, through dialogue using AI generative technology.
- AI analysis: Involve using the project's accumulated data, knowledge base, and analysis capabilities to train vertical AI models that can perform analysis, judgement, prediction, and other functions. These models are then productized and made available to users, providing them with easy access to AI analysis capabilities.
- AI Agent Hub: Aplatform that aggregates various types of AI Agents. It allows users to create customized AI Agents without needing to write any code, similar to GPTs.
The application layer has not yet seen the emergence of any major projects, but in the long run, it must have a much higher potential with untapped opportunities.
