ZKML and Distributed Computing: A Potential Governance Narrative for AI and Web3

About ZKML: ZKML (Zero Knowledge Machine Learning) is a machine learning technology that combines Zero-Knowledge Proofs (Zero-Knowledge Proofs) and machine learning algorithms to solve the privacy protection problem in machine learning.

About distributed computing power: Distributed computing power refers to decomposing a computing task into multiple small tasks, and assigning these small tasks to multiple computers or processors for processing to achieve efficient computing.

The State of AI and Web3: Runaway Swarms and Entropy Growth

In "Out of Control: The New Biology of Machines, Society, and the Economy", Kevin Kelly once proposed a phenomenon: the bee colony will conduct election decisions in a group dance according to distributed management, and the entire bee colony will follow this group dance The largest bee swarms in the world become the masters of an event. This is also the so-called "soul of the hive" mentioned by Maurice Maeterlinck - each bee can make its own decision, guide other bees to confirm, and the final decision is truly group choose.

The law of entropy increase and disorder itself follows the law of thermodynamics, and the theoretical visualization in physics is to put a certain number of molecules into an empty box and calculate the final distribution profile. Specific to people, the mob generated by the algorithm can show the law of the group even though there are individual differences in thinking. It is often restricted in an empty box due to factors such as the times, and finally makes a consensus decision.

Of course, the group rules may not be correct, but opinion leaders who can represent consensus and pull consensus on their own are absolute super-individuals. But in most cases, consensus does not pursue the complete and unconditional consent of everyone, but only requires the group to have a general identity.

We are not discussing here whether AI will lead humans astray. In fact, there are already many such discussions, whether it is the large amount of garbage generated by artificial intelligence applications that has polluted the authenticity of network data, or because group decision-making mistakes will lead to some The incident went to a more dangerous situation.

The current situation of AI has a natural monopoly. For example, the training and deployment of large models requires a lot of computing resources and data, but only a small number of enterprises and institutions have these conditions. These hundreds of millions of data are regarded as treasures by each monopoly owner, not to mention source sharing, even mutual access is impossible.

This has brought about a huge waste of data. Every large-scale AI project must repeatedly collect user data, and finally the winner takes all—whether it is mergers and acquisitions or sales, expanding individual giant projects, or traditional Internet The logic of rodeo racing.

Many people say that AI and Web3 are two different things and have no connection - the first half of the sentence is correct, they are two different tracks, but the second half of the sentence is problematic, using distributed technology to limit the monopoly end of artificial intelligence, And the use of artificial intelligence technology to promote the formation of a decentralized consensus mechanism is simply a natural thing.

Bottom Deduction: Let AI Form a Real Distributed Group Consensus Mechanism

The core of artificial intelligence lies in people themselves, and machines and models are nothing but speculation and imitation of human thinking. The so-called group is actually difficult to abstract the group, because what we see every day is still a real individual. But the model is to use massive data to learn and adjust, and finally simulate the group form. Don't evaluate what kind of results this model will cause, because incidents of groups committing evil do not happen once or twice. But the model does represent the generation of this consensus mechanism.

For example, for a specific DAO, if the governance mechanism is implemented, it will inevitably affect the efficiency. The reason is that the formation of group consensus is a troublesome thing, not to mention voting, statistics, etc. series of operations. If the governance of DAO is embodied in the form of AI model, and all data collection comes from the speech data of everyone in DAO, then the output decision will actually be closer to the group consensus.

The group consensus of a single model can be trained according to the above scheme, but it is still an island for these individuals. If there is a collective intelligence system to form a group AI, each AI model in this system will work with each other to solve complex problems. In fact, it will have a great effect on the empowerment of the consensus level.

For small collections, you can build an ecology independently, or form a cooperative collection with other collections to meet super-large computing power or data transactions more efficiently and at low cost. But here comes the problem again. The status quo between various model databases is completely distrustful and guarding against others—this is exactly where the natural attributes of the blockchain lie: through trustlessness, the security of truly distributed AI machines can be realized Efficient interaction.

A global intelligent brain can make the originally independent and single-function AI algorithm models cooperate with each other, execute complex intelligent algorithm processes internally, and form a distributed group consensus network that can continue to grow. This is also the greatest significance of AI's empowerment of Web3.

Privacy vs. Data Monopoly? The combination of ZK and machine learning

Human beings must take targeted precautions whether it is against AI to do evil or based on the protection of privacy and the fear of data monopoly. The core problem is that we don't know how the conclusion was drawn. Similarly, the operator of the model does not intend to answer this question. And for the combination of the global intelligent brain we mentioned above, it is even more necessary to solve this problem, otherwise no data party is willing to share its core with others.

ZKML (Zero Knowledge Machine Learning) is a technology that uses zero-knowledge proofs for machine learning. Zero-Knowledge Proofs (ZKP), that is, the prover (prover) may convince the verifier (verifier) of the authenticity of the data without revealing the specific data.

Cited with a theoretical case. There is a 9×9 standard Sudoku. The completion condition is to fill in the numbers from 1 to 9 in the nine grids, so that each number can only appear once in each row, column and grid. So how can the person who arranges this puzzle prove to the challenger that the Sudoku has a solution without revealing the answer?

You only need to cover the filling with the answer, and then ask the challenger to randomly select a few rows or columns, shuffle all the numbers and verify that they are all one to nine. This is a simple zero-knowledge proof embodiment.

Zero-knowledge proof technology has three characteristics of completeness, correctness and zero-knowledge, that is, it proves the conclusion without revealing any details. The source of its technology can reflect the simplicity. In the context of homomorphic encryption, the difficulty of verification is far lower than the difficulty of generating proofs.

Machine Learning is the use of algorithms and models to allow computer systems to learn and improve from data. Learning from experience through automation allows the system to automatically perform tasks such as prediction, classification, clustering, and optimization based on data and models.

At its core, machine learning is about building models that learn from data and make predictions and decisions automatically. The construction of these models usually requires three key elements: datasets, algorithms, and model evaluation. Datasets are the foundation of machine learning and contain data samples for training and testing machine learning models. Algorithms are at the heart of machine learning models, defining how the model learns and predicts from data. Model evaluation is an important part of machine learning, which is used to evaluate the performance and accuracy of the model, and decide whether the model needs to be optimized and improved.

In traditional machine learning, data sets usually need to be collected in a centralized place for training, which means that data owners must share the data with third parties, which may lead to the risk of data leakage or privacy leakage. With ZKML, data owners can share datasets with others without revealing the data, which is achieved by using zero-knowledge proofs.

When zero-knowledge proof is applied to the empowerment of machine learning, the effect should be predictable, which solves the long-plagued privacy black box and data monopoly problems: whether the project party can use it without revealing the user data input or the specific details of the model? After completing the proof and verification, is it possible for each collection to share its own data or model to function without revealing private data? Of course, the current technology is still early, and there will definitely be many problems in practice. This does not hinder our imagination, and many teams are already developing.

Will this situation bring about free prostitution of small databases to large databases? When you consider governance issues, it comes back to our Web3 thinking. The essence of Crypto lies in governance. Whether it is through a large number of applications or sharing, you should get the incentives you deserve. Whether it is through the original Pow, PoS mechanism or the latest PoR (reputation proof mechanism), they are all providing guarantees for the incentive effect.

Distributed computing power: an innovative narrative intertwined with lies and reality

Decentralized computing power network has always been a hotly mentioned scenario in the encryption circle. After all, AI large models require amazing computing power, and centralized computing power network will not only cause waste of resources but also form a substantial monopoly—if compared In the end, the number of GPUs is the last thing to fight, which is too boring.

The essence of a decentralized computing power network is to integrate computing resources scattered in different locations and on different devices. The main advantages that are often mentioned are: providing distributed computing capabilities, solving privacy issues, enhancing the credibility and reliability of artificial intelligence models, supporting rapid deployment and operation in various application scenarios, and providing decentralized data storage and management schemes. That's right, through decentralized computing power, anyone can run AI models and test them on real on-chain data sets from users around the world, so that they can enjoy more flexible, efficient, and low-cost computing services.

At the same time, decentralized computing power can solve privacy issues by creating a strong framework to protect the security and privacy of user data. It also provides a transparent and verifiable computing process, enhances the credibility and reliability of artificial intelligence models, and provides flexible and scalable computing resources for rapid deployment and operation in various application scenarios.

We look at model training from a complete centralized computing process. The steps are usually divided into: data preparation, data segmentation, data transmission between devices, parallel training, gradient aggregation, parameter update, synchronization, and repeated training. In this process, even if the centralized computer room uses high-performance computing equipment clusters and shares computing tasks through high-speed network connections, the high communication cost has become one of the biggest limitations of the decentralized computing power network.

Therefore, although the decentralized computing power network has many advantages and potentials, the development path is still tortuous according to the current communication cost and actual operation difficulty. In practice, realizing a decentralized computing power network requires overcoming many practical technical problems, whether it is how to ensure the reliability and security of nodes, how to effectively manage and schedule distributed computing resources, or how to achieve efficient data transmission and Communication, etc., I am afraid that they are actually big problems.

Tail: Expectations for idealists

Returning to the current commercial reality, the narrative of the deep integration of AI and Web3 looks so beautiful, but capital and users tell us more with practical actions that this is destined to be an extremely difficult innovation journey, unless the project can be like OpenAI , while holding on to a strong financial backer while we are strong, otherwise the bottomless research and development costs and the unclear business model will completely crush us.

Whether it is AI or Web3, they are now in a very early stage of development, just like the Internet bubble at the end of the last century, and it was not until nearly ten years later that the real golden age was officially ushered in. McCarthy fantasized about designing an artificial intelligence with human intelligence in a single vacation, but it wasn't until nearly seventy years later that we took the critical step toward artificial intelligence.

The same is true for Web3+AI. We have determined the correctness of the way forward, and the rest will be left to time.

**When the tide of time fades away, those standing people and things are the cornerstone of our transition from science fiction to reality. **