Author: Qin Jingchun
Reviewer:Leon lee
Source: Content Association - Investment Research
At present, the field of decentralized AI agent (DAI-Agent) has attracted much attention, and many articles have introduced the characteristics of related projects, the problems they solve, and their future potential. Although these articles help investors understand the projects to a certain extent, most of them lack in-depth analysis and fail to explore the basic characteristics of AI and the current status of Web3. Therefore, it is difficult to clarify the role of decentralized AI in the practice of Web3 value Internet, whether it is to optimize Web3 or to serve as a key component. If the inherent logic between decentralized AI and Web3 value Internet economy is not clarified, it is impossible to deeply understand the role of decentralized AI, and it is also difficult to grasp how its core components solve the problems existing in Web3. For example, what problems do the two key components of decentralized model and DAI-Agent solve respectively, and what is the inherent logic between them and Web3. If these inherent logics are not understood, it is difficult to evaluate the potential value of this field. This not only makes it difficult for us to accurately choose high-potential investment directions, but even if we choose the right track and project, it may be difficult to stick to it due to market sentiment fluctuations. To this end, I plan to deeply analyze the current basic status of Web3 and the basic characteristics of AI, explore how the integration of the two can realize the implementation of the value Internet, and how Arweave and AO can help this process through AI. Due to the rich content, the author will elaborate in two articles:
- Why the current Web3 needs to be integrated with decentralized AI to realize the implementation of the value Internet.
At present, many public chain projects focus their main efforts on optimizing and expanding the underlying infrastructure, such as ETH and various L2, Solana and other blockchains. But I think that if we only pursue the expansion of blockchain without integrating AI into it, it will be difficult to promote the implementation of Web3 value Internet. At present, in addition to the limited expansion capability of Web3, there is also the problem of data fragmentation. The personal data of users is scattered across different chains and DApps, which leads to management difficulties, high interaction costs, and complex operations, which seriously restricts users from actively contributing data. In addition, the decentralized nature leads to low management and coordination efficiency. These problems have greatly restricted the development of Web3. AI has the ability to learn, speculate and make decisions autonomously. AI Agent can serve as an intelligent assistant for users, greatly improving efficiency. After the integration of the two, it will significantly improve the user experience, lower the entry threshold, and promote the development of Web3.
- The inherent relationship between decentralized storage and computing platforms, decentralized models, and DAIAgent: The combination of the three can open up the closed loop of Web3 data asset economic activities and realize the true value Internet :
1.1 DAI-Agent
One of the core features of Web3 is the user's control over their own data. DAIAgent can help users centrally manage and collect data, effectively solving the pain point of data being scattered across various platforms, while acting as an intelligent assistant for users to reduce operational difficulty and improve the efficiency of interaction with Web3. For example, DAIAgent can assist users in managing their DID lifecycle, including creating, updating, and revoking DIDs, thereby simplifying data management and usage experience. It is necessary to discuss the relationship between AI-Agent and DID in detail here to lay the foundation for subsequent discussion. In the Web3.0 environment, DID and DAI-Agent are highly complementary and compatible:
a. Data integration and high-quality input :
AI-Agent can integrate data (such as social, medical, and professional data) across platforms, effectively breaking down information silos; its intelligent algorithm can filter, clean, and format data according to the needs of DID (such as evaluating the credibility of each data source, removing duplicate or low-value data, and organizing data in accordance with the DID data model specification), providing guarantees for creating high-quality DIDs. At the same time, differential privacy, homomorphic encryption, and the latest multi-party secure computing (MPC) technology are used to complete data analysis and calculations without leaking the original data (for example, when collecting medical sensitive data, it can meet health information needs while protecting personal privacy). In addition, with the continuous maturity of cross-chain interoperability protocols (such as Polkadot, Cosmos, and other ecosystems), DAIAgent is expected to achieve seamless docking between more data sources, further improving the efficiency and accuracy of data integration. The decentralized architecture not only avoids the risk of single point failure and data being controlled by a single entity, but also enables automated data collection and real-time updates through smart contracts, providing strong support for building a trusted and dynamic digital identity system.
b. Identity authentication and authorization basis :
In a decentralized environment, the digital identity system provides DAIAgent with the necessary authentication and authorization mechanisms, enabling AI-Agent to prove its legal identity and authority when interacting securely with other agents. This process not only relies on technical means, but also can be further enhanced through the distributed autonomous organization (DAO) mechanism, with the community participating in supervision and governance, to further enhance the transparency and security of the system.
c. Enhance trust and reduce interaction costs :
With the help of the DID system, the identity and behavior of DAIAgent are more transparent and verifiable, thereby building trust and promoting collaboration with other agents; at the same time, AI-Agent effectively alleviates the inefficiency caused by the decentralized nature by reducing the interaction cost between users and the system and simplifying complex operations. In addition, combined with emerging federated learning and privacy computing technologies, in the future DAIAgent will be able to achieve cross-platform and cross-domain data collaboration and intelligent decision-making without exposing the original data, providing users with more accurate and personalized services.
1.2 Decentralized Model
Models can be regarded as the "brain" of AI-Agent to a large extent, and are the core components for realizing intelligence. In the future, a large number of AI-Agents will emerge and play a role in various industries, and these professional fields (such as medical, education, finance, etc.) need their own corresponding AI models to support them. General AGI can meet the basic needs of users, but for various professional fields, it is still necessary to rely on a large number of specialized AI-Agents to work together, which requires a wide variety of models. Since decentralized models have the advantages of no permission and verifiability compared to centralized models, they will surely be favored by DAI-Agent in the future: the permissionless feature enables anyone to participate in model development without relying on the approval of centralized institutions, thereby promoting technology openness; at the same time, the permissionless feature enables DAIAgent to schedule various models more flexibly and significantly enhance intelligent attributes. In addition to the above advantages, in the future, in terms of data sharing and model training, federated learning and cross-domain collaboration mechanisms will also become key technologies to promote the development of decentralized models, protecting data privacy and ensuring the efficiency and security of model training. Especially when it comes to highly sensitive fields such as finance and medical care, the model's training process and data source must undergo multiple verifications to ensure the overall credibility and robustness of the system.
1.3 Decentralized storage and computing platform based on blockchain technology
To realize Web3 data rights confirmation, it is necessary to build a decentralized storage and computing platform to establish a verifiable data consensus infrastructure to support large-scale data exchange. Specifically, the overall solution of Arweave and AO builds a data consensus infrastructure on both the storage and computing ends, thereby achieving the following goals:
- Reduce data storage costs and ensure data security and immutability;
- Facilitate large-scale data exchange and provide a solid foundation for the hosting and operation of a decentralized AI ecosystem;
- Through a unified data storage layer, the data integration process is simplified and the integration complexity caused by data dispersion is reduced;
- At the same time, the platform also provides the necessary data support for building the DID system in Web3, enhancing the management and application of digital identity.
The above three points complement each other:
- DAIAgent combines token incentives to encourage users to contribute data and actively interact with Web3, thereby generating more data;
- The generation of large amounts of data has promoted the development of decentralized storage and computing platforms, because the platforms can not only reduce data storage costs, but also promote data ownership confirmation;
- Decentralized models need to be hosted on decentralized platforms, which can not only reduce storage and computing costs, but also ensure the verifiability and censorship resistance of the model, thereby improving the security and trust of the model and further promoting model development.
In addition, decentralized model training requires massive amounts of high-quality data, and the emergence of large-scale high-quality data will significantly improve model quality; the improvement of model quality will make DAIAgent more and more intelligent, thereby stimulating more user interactions and generating more data; and the continuous enrichment of data will further promote the improvement of storage and computing platforms, forming a positive cycle, which is interconnected and endless, and ultimately constitutes a complete data asset economic ecosystem. This ecosystem has created a closed loop for the circulation of data assets, which is the key to forming a truly valuable Internet ecosystem. As shown in the figure:
Based on the above logical analysis, we can see that DAI-Agent is only a key link in the entire ecosystem, and its development is largely subject to the support of the other two parts (i.e., decentralized storage/computing platform and decentralized model). Therefore, when investing in such projects, it is necessary to pay attention to whether the project has the ability to build a complete data asset economic ecosystem, or whether it has established a relatively stable cooperative relationship with the other two aspects. If you only invest in a single-direction project, the risk will be greatly increased. In addition, although the currently popular DAIAgent protocols such as ELIZA, VIRTUAL, and APC support diversified models, some of them allow centralized model providers such as OpenAI to access. Although this can meet the diverse needs of users, if the proportion of centralized models is too high, it will restrict the long-term development of the protocol due to its lack of permissionless characteristics.
2. Here I want to focus on: Arweave permanent storage + AO super parallel computer overall solution
1. Parallel processing capabilities
Unlike networks such as Ethereum, whose base layer and various Rollups usually run as a single process, AO supports any number of processes running in parallel while ensuring the complete verifiability of the computation. In addition, these networks need to run in a globally synchronized state, while AO's processes remain independent. This independence allows each process to handle more interactions, greatly improving computational scalability, and is particularly suitable for application scenarios that require high performance and reliability. In the future, as a large number of DAIAgents perform tasks on the chain uninterruptedly, the requirements for system scalability will become more stringent, and AO's ultra-parallel processing capabilities just meet this need.
2. Ability to store and run large models and other types of models
In the AO network, the memory limit of a single node is currently 16 GB, while the upper limit of memory expansion at the protocol level can reach 18 EB, which is enough to run most models in the existing AI field (such as the unquantized version of Llama3, the Falcon series, and many other models). Considering that the parameters of GPT-4 have exceeded 1.76 trillion, it is expected that GPT-5 will exceed 50 trillion parameters, and the scale of models will continue to grow in the future. AO has very strong expansion capabilities. You only need to physically increase the memory or graphics card to expand the computing unit to meet the operation requirements of large models.
Arweave uses a unique blockweave technology that allows new blocks to be connected to multiple old blocks, making it highly scalable and theoretically capable of storing various models and large-scale data. At the same time, through WeaveDrive technology, applications can access data on Arweave as easily as accessing local disks, which makes it possible to build various applications. All types of applications can access the permanently stored data on Arweave, and AO+Arweave has built an infrastructure for data rights confirmation from both computing and storage aspects, laying the foundation for large-scale data asset exchanges, which is very attractive to developers who intend to develop applications on the AO platform. At the same time, various application scenarios provide diversified landing scenarios for various models and DAI-Agents, thereby promoting the development of the AI ecosystem.
3. Data is one of the three major elements of the AI ecosystem - most of the data in the AO + Arweave ecosystem is high-quality data and has a unified data storage layer
Large-scale and high-quality data is crucial for model training. High-quality data usually has characteristics such as accuracy, consistency, validity, completeness, timeliness and uniqueness. In the AO+Arweave ecosystem, most of the circulating data meets these characteristics. For detailed technical implementation details, please read my previous article "Arweave Permanent Storage + AO Super Parallel Computer: Building Data Consensus Infrastructure". Here we need to emphasize the advantages of Arweave permanent storage: because of its permanent storage properties, the stored data is often more critical; the longer the data is stored, the more its value can be reflected, because it is not only convenient for preservation and traceability, but also conducive to data ownership. Large-scale high-quality data is extremely important for AI training, and Arweave, as a unified data storage layer, has the ability to integrate data from various projects. In contrast, Ethereum, Solana, etc. have more difficulty in data integration due to the lack of a unified storage layer. These characteristics of Arweave play a key role in data collection, integration and integrity assurance, which is crucial for building DIDs in Web3: a unified data storage layer is far more convenient than cross-platform data integration. In addition, the integration of AO and Arweave ensures that all agent interaction data can be permanently stored, which provides strong support for the establishment of accountability mechanisms and DID and reputation systems. For example, the RedStone project is currently using Arweave to build DID and establish an accountability mechanism to provide infrastructure support for the development of AI-Agents.
4. AO + Arweave gives AI higher verifiability
Verifiability is crucial to the development of AI. It ensures that the predictions and outputs of AI models are transparent, tamper-proof, and independently verifiable, providing AI with higher credibility and security, so that it can be widely used in high-trust fields such as finance, medicine, law, and autonomous driving. At the same time, verifiability also allows developers to share and collaborate on models more confidently without worrying about malicious tampering. AO+Arweave uses SCP storage, so that all data and models in AO can be holographically stored on Arweave, and anyone can verify the data source, model operation process, and output results; at the same time, the encrypted signature provided by the computing unit further ensures the authenticity and integrity of the calculation results. With the continuous improvement of zero-knowledge proof technology and distributed verification mechanisms, in the future, not only can the model output be verified in real time, but also the entire process of model training data, parameter updates, etc. can be traced and audited, thus forming a comprehensive, multi-level trust system. In addition, the Verifiable Confidential Computing (vcc) jointly initiated by AO and PADO uses ZKFHE (zero-knowledge fully homomorphic encryption) technology to protect the privacy of data and models, as well as their verifiability and computability. Such a mechanism not only greatly reduces the risk of data sharing, but also provides intellectual property protection for model providers, encouraging the opening and sharing of more high-quality models. Combined with the token incentive mechanism, this trust system is expected to further inspire users to actively contribute data and promote the entire AI ecosystem to a higher level.
The basic components and relationships of the AO+Arweave ecosystem are shown in the figure:
In summary, the AO+Arweave ecosystem provides a superior operating environment for decentralized AI: it not only has excellent scalability and hosting capabilities, suitable for supporting the decentralized AI ecosystem, but also has significant advantages in large-scale high-quality data storage and exchange, parallel computing, and verifiability. These factors together make the AO+Arweave ecosystem an ideal platform for the development of decentralized AI. Through the above demonstration, decentralized AI undoubtedly plays a vital role in the three major elements required for the implementation of the Web3 value Internet ecosystem. It can be seen that AO+Arweave+AI is expected to greatly promote the implementation of Web3!
