FHE+Restaking+AI: Mind Network Under the Perfect Narrative (Three Days in the Sky)
Reprinted: Bluue @deepbluuest
Mind's Restaking support is extensive, covering LST/LRT assets of BTC, ETH, and even tokens from other networks.
Mind's business logic is similar to EigenLayer's AVS, but it can also collaborate with Eigen and other networks.
Mind is still relatively early, with few participants, but you can experience it; there might be surprises in the future.
Mind's current main solutions focus on AI and Depin, +FHE, +Restaking narrative fully loaded.
Mind acts like an abstract player in the FHE track, but does not compete with other FHE projects; in the AVS track, it does not compete with AVS projects; in the Restaking track, it does not compete with Restaking projects; in the AI and Depin tracks, it provides services!
Experience link: https://dapp.mindnetwork.xyz/
What is Mind Network?
Mind Network is defined as a FHE Restaking Layer serving AI and PoS networks.
AI and PoS: These are the target services.
FHE: The technical security layer, which is the technical support, homomorphic encryption, ensuring fair verification.
Restaking: The economic security layer, which is the economic support, the source of consensus.
Understanding it in terms of AVS, the left hand connects to Restaking assets, while the right hand provides secure consensus for networks like AI.
Main Products:
Subnet: Subnet, specific use cases based on FHE.
Remote Restaking: Remote staking, allowing assets on the original chain to participate in Mind's restaking.
Mind Chain: A Rollup based on Altlayer, mainly responsible for connecting Restaking and Subnet.
FHE Bridge: Launched based on invisible addresses and CCIP, FHE cross-chain bridge, mainly serving B-end.
Mind Lake: FHE privacy database, mainly serving B-end.
Investment and Background:
Completed a $2.5 million seed round in 2023.
Incubated by Binance and Chainlink ecosystem projects.
Ethereum Foundation Grant
Testnet data (ended): 650,000+ active users; 3.2 million+ transactions.
Mind Network Architecture
Overall, it can be divided into three parts
Economic Input (Restaking Layer): Receives LST/LRT assets, supporting various restaking tokens like Ethereum, BTC, etc.
Intermediate Architecture (Mind's own security layer + consensus layer): Based on FHE, encrypts voting, encrypts computational processes, building an overall FHE verification network.
External Output (Subnet): Provides shared security based on Restaking and FHE for networks like Depin, AI, PoS, etc.
Subnet
Mind Network's solution is called Subnet.
According to the introduction, Subnet is a verification demand network based on FHE and Restaking, akin to an upgraded version of AVS.
Each Subnet can customize its own tasks and logic, such as validator node requirements, reward rules, FHE-related functions, encryption and decryption, etc.
Based on currently available information, the first batch of subnets should focus on AI and Depin, such as Io.NET, Myshell, etc.
Remote Staking
First, it is important to clarify that Mind Network supports LST/LRT assets, but does not support original assets like BTC/ETH.
The current version mainly supports LST assets from Ether.FI, Renzo, Lido, Stakestone.
Remote Staking provides a way to participate in staking with minimal security assumptions, without needing cross-chain operations.
For example:
On an L2 like Manta, use ETH to participate in Stakestone staking and receive Stone.
On Mind, switch directly to the Manta chain, click stake, and you can stake Stone.
There is no need to cross-chain Stone to the mainnet or similar.
The main benefits of participating in remote staking: assets remain on the original chain, operations are simple, and security assumptions are low.
Currently, attention needs to be paid to how the final rewards are distributed and to which chain, etc.
AI and Decentralized AI
The most perceptible aspect of AI for users is the model, which is generated thanks to two key resources: computing power and data.
One is the raw material, the other is the driving force.
In terms of computing power, enterprises monopolize it, making it difficult for small and medium AI to seize computing power. Decentralization can utilize idle computing power in users' hands through token incentives, i.e., Depin types.
In terms of data, data cleaning and labeling are upstream processes, which are then provided to models for processing and optimization. Throughout this process, data faces multiple threats, such as storage, data computation, data output, etc.
The security issues involved in the data process, for example, when conversing with ChatGPT, we essentially establish a trust assumption, trusting that OpenAI will not peek at our data.
Thus, decentralized AI is born, such as cloud computing power markets, cloud training markets, AI agent markets, GPU computing power markets, prediction markets, generative dialogue, etc.
However, Crypto AI also faces some difficult problems to solve:
- Cloud computing or Depin-type projects encounter fairness issues, and fairness is essentially a consensus issue.
Who gets the tasks, how to fairly judge contributions, and the risk of data leakage from personal devices?
- For example, the issues faced by decentralized data markets like Bittensor.
Miners train models, and validators vote to give scores. How to ensure that validators are not colluding? How to ensure that the model given to me is the best?
The data privacy involved in the entire process can be roughly divided into several parts:
- Will user device data be leaked to the platform?
- Will the data that needs to be computed be leaked to users?
- How does the platform prove its own innocence and security? … etc.
Thus, great cryptography comes into play: data encryption, FHE computation, ZK verification, MPC decentralized permissions.
The ultimate ideal state: end-to-end encryption of AI.
That is, AI does not know what we are asking, nor does AI know what it is computing, but AI can output the answers we want.
Problems Solved by FHE
The core of PoS networks is voting, and the original intention of voting is to hope that validators independently verify and reach consensus on the public network based on the results.
However, in reality, there are very few networks that can be used like Ethereum for a large number of node-level networks.
Only when there are more validator nodes can we ensure that a small amount of cheating does not affect the overall situation, thus solving the BFT Byzantine problem.
Fewer nodes can lead to cheating and manipulation behaviors, such as rewards for following votes and bribery manipulation.
Based on the issues of PoS networks, Mind provides security verification based on FHE, which simply means encrypting the voting process, thus avoiding the need for sufficient node numbers (not decentralized enough, hence not secure) while still achieving network security.
However, in practice, it is difficult to define whether the number of nodes is "more or less," meaning that apart from Ethereum and Bitcoin, it is hard to recognize other networks as sufficiently secure.
It can thus be simply understood that almost all PoS networks can achieve secure consensus using Mind.
Core logic of PoS: Nodes stake tokens to obtain validation rights, participate in validation, and receive rewards.
This process is reflected in Depin, AI, and most networks, so Mind's solution can be applied to the aforementioned types of projects.
For example, Bittensor requires staking TAO as a validator; Io.net requires staking IO to participate as a node, etc.
For decentralized AI, since the process involves a large amount of high-value data computation, security and privacy are particularly important, and participation in computation is also required.
The encrypted computation process perfectly aligns with the core of FHE, so FHE + decentralized AI is like Audi's double diamond, my partner.
Summary
- The basic logic of Mind:
Common issues faced by AI and networks using PoS consensus: not decentralized enough, meaning fundamentally not secure.
FHE encrypts the voting process, forcing validators to make independent evaluations.
The establishment of voting rights comes from staking, and restaking assets can provide an additional layer of rewards compared to native assets.
Remote Staking solves the security of staked assets.
Decentralized AI faces more issues regarding data and security privacy, such as cross-validation, model selection, training, inference, decentralized computation, etc., which FHE perfectly addresses.
Although the design of the Mind project appears somewhat complex, the user experience is very simple: Stake LST/LRT ------ Receive rewards.
Mind's resources and background are complete; what we need to look forward to and focus on now are reward settings, token models, network support, and the first batch of Subnet applications.
