Integrate, Don't Isolate
An Outlook on Single Party Verification and Shared Private States
Conferences, Crypto Twitter, and "market research" have recently evolved from buzzword battlegrounds to acronym races. Terms like ZK (Zero Knowledge), MPC (Multi-Party Computation), FHE (Fully Homomorphic Encryption), and TEE (Trusted Execution Environment) are pervasive and, unfortunately, pitted against each other for the top spot in digital security. In reality, these cryptographic technologies do not solve identical problems and should not be directly compared.
The root cause for this confusion is these solutions' abstractness and relative similarity. Before we delve into the solutions, let's first define three current cryptographic problems these technologies attempt to solve and contextualize them with real-world examples.
Privacy-Preserving Computation: Processing data so that personal or sensitive information remains private, even from the computing party.
Example: Using AI on medical records but maintaining HIPAA compliance.
Secure Computation in Untrusted Environments: Maintaining raw and computed data confidentiality when processing is completed in untrusted environments like cloud servers.
Example: Untrusted third-party (cloud) AI that maintains HIPAA compliance on medical records.
Authentication and Verification: Confirming a user’s or device’s identity (authentication) and ensuring the integrity of data or a transaction (verification). Authentication can involve something you know (a password) or something you are (biometrics), while verification ensures data has not been tampered with.
Example: Proving that the particular model actually ran the computations over the medical data.
If we bridge these problems with their respective acronyms(s), we run into blockchain’s notorious Similar Use Case Dilemma. This occurs when multiple solutions exist for multiple use cases, which inevitably creates confusion.
From these problems and use cases, we can distill two broad issues: verification and shared private states. Verification is a single party proving the validity of a statement without revealing underlying data, and Shared Private States are multiple parties maintaining privacy invariants even while computing over, say, private data.
These cryptographic technologies—or acronyms—fall into these two categories and offer unique strengths and weaknesses in terms of data privacy and computational complexity.
This mix of computational complexity, data privacy, and composability allows developers to match the use cases with cryptographic tools rather than vice versa. Thus, the conclusion of integration rather than separation is intuitive and a particularly exciting solution for AI in Web3.
As an aside, AI in Web3 has gained popularity because of concerns over the current centralized and closed-sourced intelligence ecosystem. Rapid development has been made in chain model comparison, off-chain computing, and input/output validation to combat these issues, and more decentralized AI applications will likely be built. Importantly, each application has context-specific yet differing scalability, security, and complexity requirements.
It makes sense that AI (or AI adjacent) Web3 projects are implementing combinations of shared state and verification technology. A notable example is Aizel Network, an on-chain inference network that leverages TEE for heightened compute speed and capacity to handle large models, and MPC for trustless and modularized data processing. Using MPC and TEE - the less expensive combination of verification of shared private state - highlights the use case's prioritization of computational efficiency over perfect encryption and accuracy. Aizel’s modular design offers users a current solution for LLM verification rather than the more expensive ZKML, all the while allowing future integration with ZKPs.
Marlin is another project that benefits from this cost vs. security matrix but combines lateral verification techniques. Marlin uses TEEs for bulky off-chain computations—think complex simulations or processing ML datasets—and then the more expensive ZKPs to effectively "bundle" and verify these computations on-chain.
Considering the similar tradeoffs of cost and security for decentralized AI, these combinations represent flexible solutions for scaling AI in Web3 now and in the future. Rather than a static technology stack, modular design allows for fluid privacy and cost structures that can change over time and per use case.
Therefore, this acronym race is likely to end prematurely. Instead, an integrated future will allow builders to match verification and private state architecture with specific data sensitivity and size.
Special thanks to David Attermann for his research on Aizel Network and AI in Web3 as well as his guidance throughout this research process.
About M31 Capital
M31 Capital is a global investment firm dedicated to crypto assets and blockchain technologies that support individual sovereignty.
Website: https://www.m31.capital/
Twitter: https://twitter.com/M31Capital
Great content.