Let's take our ultimate vision of an HTTPZ fully encrypted internet to the next level!
On the evening of June 17th, Mind Network invited Rand Hindi, the founder of the open-source cryptography company ZAMA, for an exclusive dialogue to discuss FHE technology, applications, comparisons, and decentralized AI. The participating guests also included Mind Network's co-founders Christian and Mason, as well as research lead Ashely, both of whom are fellows of the Ethereum Foundation.
Zama is an open-source cryptography company co-founded by Hindi and Pascal Paillier in early 2020. Paillier is a renowned cryptographer and one of the inventors of fully homomorphic encryption (FHE) technology, having previously completed a $73 million Series A financing.
Mind Network is the first fully homomorphic encryption (FHE) re-staking layer for AI and POS networks. It accepts re-staked tokens from ETH, BTC, and AI blue chips, and operates as an FHE verification network, providing consensus, data, and encrypted economic security for decentralized AI, DePIN, EigenLayer, Symbiotic AVS, and many other critical POS networks.
AMA review link: https://x.com/mindnetwork_xyz/status/1802725269867757743
FHE AI Network: A decentralized AI network using FHE technology that can provide more secure consensus and more private AI data.
HTTPZ: A fully homomorphic encryption internet that achieves end-to-end comprehensive encryption through FHE fully homomorphic encryption, ensuring that data transmission and processing are always encrypted.
What are the issues that need to be improved in the current Web2 AI framework?
Rand:
The main issues with centralized AI are:
Integrity and correctness of computation: In centralized AI systems, the integrity and correctness of computation are questionable. This means that due to the opaque nature of the computation process and model parameters, we cannot fully trust the results obtained.
Confidentiality and privacy: The issue of personal data privacy is particularly prominent. For example, when using software, the operation records are visible to the company, making the enterprise a single point of failure for privacy and data. Attackers can obtain all information by targeting a single entity.
The advantage of decentralization lies in public verification and data security. If you do not trust the results obtained, you can verify them yourself. This is particularly important in AI, especially in sensitive scenarios. The decentralized nature of blockchain means that there is no single point of failure, and attackers cannot obtain all information by attacking a single target.
Blockchain can address the issue of computation integrity in AI, while fully homomorphic encryption (FHE) can address the issue of data privacy in AI. This is also one of the directions we are working on in collaboration with Mind Network. Therefore, decentralized encrypted AI will be the future direction of development.
Differences between FHE, ZK, and MPC
Rand:
In cryptography, there are many technologies applied, especially in the privacy aspect, with widely used FHE, ZK, and MPC.
- Zero-Knowledge Proof (ZK): ZK is a very interesting technology, but it is limited in that it cannot be combined with other technologies in an encrypted state and cannot compute encrypted results. It only allows proving that certain calculations have been performed on these values without disclosing the specific values. On the blockchain, ZK cannot achieve interoperability between multiple contracts or users. This is because the person providing the proof must perform the calculation in plaintext, thereby obtaining all the data. This does not actually solve the problem. However, ZK still performs well in other aspects such as scalability, such as zkRollup.
- Multi-Party Computation (MPC): MPC is actually a general term rather than a single technology. MPC involves how to securely perform computations among multiple parties, and there are many technologies that can implement MPC solutions.
- Fully Homomorphic Encryption (FHE): FHE allows direct computation on encrypted data without the need for decryption. This ensures the privacy of data while ensuring the accuracy and integrity of the computation.
By combining these technologies, we can achieve advanced privacy protection. For example, Zama is developing FHEVM and encrypted smart contracts, using FHE for computation and encrypted data, combined with MPC to disperse data and achieve selective decryption, thereby ensuring privacy protection of data in a multi-user environment.
Mason:
Rand provided a good explanation of how fully homomorphic encryption (FHE), zero-knowledge proof (ZK), multi-party computation (MPC), and decentralized AI address the security and privacy issues unresolved in Web2. Additionally, FHE can not only address data privacy issues but also resolve fairness issues in decentralized networks.
FHE allows computation of votes on decentralized networks in an encrypted state, ensuring the security of the consensus computation process and the fairness of the results. This means that even if there is distrust between nodes, cheating can be prevented through encrypted computation. This is something that ZK cannot fully achieve, as Rand mentioned, as ZK still requires trust in the prover. In encrypted applications that require the participation of multiple users and have privacy requirements for computation results, FHE is more suitable, especially in decentralized AI networks.
Just like ZAMA's Concrete ML open-source library, which provides the foundation for data encryption in AI networks, Mind Network uses FHE technology to support the consensus layer of decentralized AI networks. Data encryption combined with consensus security is the future form of AI networks that we can foresee.
Introduction to Zama's Products
Rand:
We do not have our own token, nor do we operate like a blockchain. Our goal is to build technology that allows others to create decentralized protocols. Our main library is called TFHE-rs, which is a fully homomorphic encryption (FHE) library written in Rust, containing all the encryption implementation algorithms provided by Zama.
In addition, we have developed FHEVM, which is an encrypted smart contract platform that allows writing Solidity smart contracts on encrypted data. We also provide Concrete ML, which allows creating encrypted machine learning models directly in Python. Developers can use scikit-learn (an open-source machine learning library), PyTorch (an open-source deep learning framework), and NumPy (a foundational library for scientific computing supporting large-scale multi-dimensional arrays and matrices), and we will automatically convert them to the FHE protocol.
Zama's main focus is to enable developers to easily build FHE applications without needing to learn complex cryptographic knowledge.
Another key issue is that in the past, it was not possible to guarantee that the result of encrypted computation was exactly the same as the unencrypted result. Some level of tolerance may be acceptable, but for blockchain applications, such as million-dollar transfers based on smart contracts, it must be exactly the same, not just approximately so.
Zama's technology, called threshold FHE (tfhe), allows computation on encrypted data and ensures that the result of encrypted computation is exactly the same as the unencrypted data. Therefore, as a developer, you no longer need to worry about approximation errors and other issues.
Collaboration between Mind Network and ZAMA and its Architecture
Ashely:
In decentralized AI networks, the consensus mechanism is crucial for ensuring that nodes in the network reach agreement, often needing to address two main issues:
Consensus security and fairness: Validators may collude with other nodes rather than independently verifying, affecting the integrity of the consensus.
Data Privacy and Security: The leakage of data and computation results at decentralized nodes threatens the security of the consensus process.
To enhance the security and integrity of the consensus process, Mind Network has introduced a fully homomorphic encryption (FHE) verification network. Encrypting the data of the validators ensures that they cannot replicate from other nodes and must independently run computations, thereby avoiding plagiarism and enhancing computational independence and data privacy. Additionally, since the results themselves are encrypted, only those with the key can decrypt them, preventing attackers from tampering with the data even if they gain access to stored data.
For example, in the FHE AI subnetwork, the verification process is as follows:
- Model Verification and Ranking: Each node independently verifies AI models and ranks them. Because the data is encrypted, nodes cannot see the computation results of other nodes, ensuring independence.
- Consensus Reaching: Nodes use FHE for encrypted computation and ultimately reach consensus through encrypted voting mechanisms, ensuring the accuracy and fairness of the results.
The value of our collaboration with ZAMA: For widespread decentralized AI networks, FHE ensures independent verification, identifying the most valuable models and providing the use cases that the market truly needs. Additionally, FHE computation can ensure the security and privacy of data in decentralized nodes.
This means that our jointly built FHE AI network with ZAMA can support more high-value scenarios, such as investment strategy calculations, biological information analysis, etc., truly empowering the owners of data and models with ownership and profits.
Introduction and Applications of the FHE AI Network
Rand:
Concrete ML is one of our most impressive products.
A few years ago, I told my team, "Guys, do you think we can write a program that is not scikit-learn or PyTorch and run it on FHE?" They looked at me and said it was almost impossible because we basically require converting Python to FHE equivalent, but we did it.
We have a special compiler that converts Python code into equivalent FHE operation circuits, optimized for performance and security. Therefore, what we ultimately get is an executable file that can run on any machine that handles encrypted data.
You can do a lot with it. We have some demos on Hugging Face, such as using Concrete ML for image processing.
For example, if there's an image that needs resizing, applying filters, and perhaps blurring some parts that shouldn't be shown, all of this can be done directly without seeing the actual content of the image.
Another example related to medical data, you can upload encrypted medical records for automatic diagnosis based on the selected AI model without revealing any information about the data.
FHE is a revolutionary technology with immense potential in the decentralized and AI fields, and I am very excited to collaborate with Mind Network to explore more use cases for FHE AI.
Christian: To summarize, ZAMA has provided many out-of-the-box open-source products, making a significant contribution to FHE. Among them, ConcreteML addresses the issues of AI data privacy and developer usability, while Mind Network's FHE Validation Network addresses the security and fairness issues of decentralized AI networks. The combination of FHE technology and AI networks will be the foreseeable future development of AI networks.
The collaboration between ZAMA and Mind Network will bring about a revolutionary decentralized FHE AI computing paradigm, taking our ultimate vision of an HTTPZ fully encrypted internet to the next level!
Both parties will further advance their collaboration by hosting a workshop on "How to Build FHE AI" during ETH CC on July 9th.
Event registration link: https://lu.ma/zxmz7vzb
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
