Aiming at the core of AI, a detailed explanation of Binance's investment in the FHE project Mind Network

Mind Network is the first re-staking solution designed for AI and PoS networks based on FHE.

The Holy Grail of Cryptography ------ Fully Homomorphic Encryption

On May 5th, Ethereum founder Vitalik Buterin shared his 2020 article on FHE (Fully Homomorphic Encryption) again on Twitter, reigniting interest and discussion around the applications of FHE technology. Vitalik's article delves into the relevant mathematical principles, English version.

FHE (Fully Homomorphic Encryption) is known as fully homomorphic encryption computation in Chinese, and like ZK, it is one of the cutting-edge fields of cryptography, often referred to as the Holy Grail of cryptography.

In simple terms, fully homomorphic encryption allows direct computation on encrypted data without the need for decryption.

When calculating 1+2, it is easy to arrive at the result of 3, but when encrypted, Encrypt(1)+Encrypt(2) can still yield Encrypt(3); this is FHE, where ciphertext computation equals plaintext computation after encryption.

Unlike ZK, FHE's application in Web3 focuses more on data privacy and security. From current applications, it is not difficult to see that ZK is more reflected in scalability.

Although Web3 is more familiar with ZK technology primarily based on ZKRollup, FHE is gradually releasing its unique potential in various fields, especially in AI.

Mind Network

Mind Network is the first re-staking solution designed for AI and PoS networks based on FHE.

Just as EigenLayer serves as a re-staking solution for the Ethereum ecosystem, Mind is the re-staking solution for the AI field. Through re-staking and FHE consensus security solutions, it ensures the security of the token economy and data safety of decentralized AI networks.

From the team's background, the main members of Mind are professors and PhDs in AI, security, and cryptography, coming from institutions such as Cambridge, Google, Microsoft, and IBM. Core members have been selected as one of the 12 global Ethereum Foundation Fellows, conducting research in cryptography and security alongside the Ethereum Foundation research team. Mind's world-first FHE+Stealth Address solution—MindSAP (research paper link, the original text is quite complex, readers are encouraged to study it themselves)—addresses the issues raised by Vitalik in the Stealth Address Open Problem, garnering significant attention in the Ethereum community and resulting in multiple papers and presentations.

In 2023, Mind Network was selected for the Binance Incubator and completed a $2.5 million seed round financing with participation from well-known institutions like Binance. It also received the Ethereum Foundation Fellowship Grant, was selected for the Chainlink Build Program, and became a signed Channel Partner of Chainlink.

In February 2024, Mind Network became a key partner of the renowned cryptography company ZAMA in the field of FHE.

Recently, Mind Network has further accelerated the expansion of its ecological landscape, providing AI network consensus security services for io.net, Singularity, Nimble, Myshell, AIOZ, and offering FHE Bridge solutions for Chainlink CCIP, as well as AI data security storage services for IPFS, Arweave, and Greenfield.

FHE+AI, Addressing Core Pain Points of AI

At the Hong Kong Web3 Conference in April this year, Vitalik expressed his future expectations for FHE in scenarios like Encrypted Voting. FHE, as a frontier of cryptography, is also the extreme direction pursued by Ethereum in cryptography.

ZAMA's founder recently published an article about its "Master Plan," outlining the company's vision to create an end-to-end encrypted network HTTPZ ("Z" stands for "Zero Trust") and proposing to make FHE ubiquitous in the blockchain and artificial intelligence fields.

Several key aspects of AI, including training, tuning, usage, and evaluation, face the same challenge in the decentralized process: how to remove trust assumptions. For example:

• When training an AI model, cross-validation is needed to select the best training results.

• When using AI services, existing services need to be ranked to determine the best service.

• AI models also require continuous tuning and iteration, necessitating independent evaluation.

These aspects in centralized scenarios are based on compliance trust assumptions of large companies, with large companies providing trust endorsements not to act maliciously.

However, in the decentralized process, without credit endorsement, verifying whether the collaboration of all parties is fair and effective is a challenge, which is precisely where FHE empowers.

For example:

• When training an AI model and needing to perform cross-validation, anonymous voting can be used to select the best training results, removing assumptions similar to those of OpenAI.

• When using AI services and needing to rank existing services, anonymous scoring can determine the service quality of each service, removing trust assumptions similar to those of an AI AppStore.

• When AI models require continuous tuning and iteration, independent evaluations can be conducted through random sampling checks, removing trust assumptions regarding evaluation agencies.

The involvement of FHE also allows AI to achieve zero trust, compensating for the trust assumptions that ZK still requires for off-chain aggregation.

There are many other AI examples, including how such zero trust can enable AI Agents and Multi-Agents to better achieve intelligent interconnection and realize good governance.

At the same time, FHE's unique ciphertext computation characteristics can also solve two other challenges: data privacy and data ownership:

• Who can see our data? = Data privacy

• Who owns the data provided by AI? = Data ownership

FHE can ensure that data is always encrypted on the user side, existing only in ciphertext form outside the user, including storage + transmission + computation.

So far, apart from FHE, data can only be encrypted during storage and transmission, but once computation is involved, the ciphertext must be decrypted into plaintext, which causes users to lose ownership of their data. In real life, there are many such examples; once your plaintext data is copied by someone else, they can make multiple copies, and users cannot know whether others are using their data, relying solely on self-declarations from data users and third-party supervision. FHE allows users' ciphertext data to require user consent for decryption and viewing plaintext data, enabling users to perceive the dynamics of their data, achieving data usability and tradability without visibility, thus protecting data privacy while genuinely safeguarding data ownership.

Such characteristics are urgently needed in AI + Web3, allowing everyone to stake in an open manner while also achieving consensus in an encrypted way, preventing malicious actions and waste.

The Next Big Thing in AI

From this perspective, the combination of AI and Web3 is inevitable; FHE for AI is like the 【next big thing】 for Apple.

Recently, IO.NET and Mind Network announced a deep collaboration to co-create solutions that enhance the security and efficiency of artificial intelligence. IO.NET is integrating Mind Network's fully homomorphic encryption solution into its distributed computing platform to help strengthen the security of its products.

For more details on the collaboration, see: Mind Network and io.net Partners up for Advanced AI Security and Efficiency

IO.NET has made a good start in combining distributed computing with AI and FHE.

Taking IO.NET as an example, users provide computing power, and AI developers rent computing power.

When a developer comes to an AI project with a requirement, the system breaks it down, and the computing power provided by users is used for computation.

This raises several questions: whose computing power is rented? Is the result correct? Will renting computing power leak privacy for both parties?

1. Whose computing power is rented?

Under normal circumstances, the selection of nodes is based on testing jobs, periodically releasing demands to test which nodes are online and ready to accept requests.

During this process, relevant nodes may manipulate to gain priority, similar to MEV attacks.

In response, Mind provides a fair distribution mechanism through FHE; since requests and data are encrypted, nodes cannot make advantageous selections based on this.

2. Is the computed result correct?

In distributed computing, ensuring the correctness of computation results requires a certain consensus, i.e., voting.

When nodes know each other's choice results, there may be follow-up voting, leading to unfair and incorrect results.

FHE encryption computation allows the voting results between nodes to be mutually encrypted while still participating in the final computation, ensuring fairness in results.

3. Will renting computing power leak privacy for both parties?

The core of FHE is data security; it encrypts during computation and encrypts the questions to be computed, naturally avoiding privacy leaks.

From the perspective of Restaking:

IO.NET itself can be seen as a PoS network, where nodes need to stake IO tokens to earn IO rewards from their computing contributions.

A potential issue is that the price volatility of staked tokens may impact validators and network security.

Mind's solution to this is Dual Staking (double staking) or even triple staking.

Staking supports liquid staking tokens for BTC/ETH and blue-chip AI network tokens, diversifying risks and increasing the overall security of the network, essentially an advanced version of shared security in Restaking.

At the same time, Mind also supports Remote Staking, ensuring the security of assets without the need for actual cross-chain transactions for LST/LRT assets.

Recently, Mind also completed the Glaxe testnet task, with over 650,000 active users participating, generating 3.2 million testnet transaction data.

According to official news, Mind's mainnet protocol will be launched soon, so stay tuned.

Conclusion

In summary, we find that although Mind discusses FHE and AI, the key word is actually "security," using cryptography to solve various core security issues.

Restaking is token economic security; Remote Staking is asset security; FHE is data security; AI+FHE is consensus security.

The blockchain edifice is built on cryptography, and perhaps the future answers will also be found within cryptography.

In addition to the AI network, Mind Network is also expanding the applicability of its solutions, collaborating in various directions such as decentralized storage, EigenLayer AVS network, Bittensor Subnet, and cross-chain bridges, showcasing the immense potential of FHE.

In the Web3 of 2024, if the field of cryptography is opened by ZK, then FHE will be the main theme of the second half of the year. Meanwhile, the heat of AI remains high; under the triple narrative of AI+FHE+Restaking, and with the backing of the Ethereum Foundation and Binance investments, whether Mind can take the lead in FHE will soon be revealed with the launch of the mainnet.