Overview of the newly funded Modulus: How to use zk to bring "opaque" AI models into a "trustless" blockchain

Original Title: Introducing Modulus: Bring AI On-Chain

Author: Modulus Labs

Compiled by: Hailsman, ChainCatcher

Last night, Modulus Labs announced the completion of a $6.3 million seed round financing, led by Variant and 1kx, with participation from Inflection, Bankless, Stanford, and others. Angel investors include the Ethereum Foundation, Worldcoin, Polygon, Celestia, and Solana.

Modulus Labs is building a dApp that uses ZKML combined with AI model ZK proofs to demonstrate that AI models are executed correctly, with the core goal of helping users ensure that AI queries are not tampered with, thus paving the way for integrating AI into Web3 applications. It aims to bridge the opacity of machine language models running on servers with the transparency of blockchain.

This article is a "self-introduction" recently released by Modulus Labs. To facilitate quick understanding for readers, ChainCatcher has made appropriate edits and reductions to the original text:

We are helping dApp developers access tamper-proof AI at scale using the world's only ZK prover specifically designed for AI. This means that smart contracts can now enjoy powerful AI capabilities without breaking the trustless tenet.

Stay tuned in the coming months as we:

• Achieve the largest AI application ever on Ethereum for the first time through partnerships with Upshot, Worldcoin, Ion Protocol, and others;
• Open-source our prover "Remainder"—the world's only dedicated ZK prover built specifically for AI;
• Launch the Modulus API, enabling every dApp to access powerful, expressive AI through ZK accountability;
• In fact, we opened the waitlist for the "on-chain AI" API today.

TL;DR

• Problem: Today, to gain powerful AI capabilities, dApps must sacrifice decentralized security and take on centralized risks;
• Solution: Modulus uses zero-knowledge proofs to effectively verify that AI providers are not manipulating their algorithms on-chain;
• Insight: A specialized ZK proof built for AI provides extremely powerful AI capabilities to dApps at a very low cost;
• Outcome: Tamper-proof AI-enhanced dApps outperform their peers in user experience while adhering to the trustless tenet of the chain;
• Vision: As advanced AI is introduced into our legal, financial, medical, security, and educational fields, crypto accountability will safeguard the pillars of our computational future.

AI Forces People into " Faustian Bargains "

AI will fundamentally change our society. However, this "superpower" is nearly unusable for anyone building with blockchain technology. The reason is simple: the computational cost of running AI operations is too high to run directly on-chain. This means that to bring AI capabilities into smart contracts, one must trust that AI can always run models without manipulation.

But in a trustless crypto context, this is difficult to achieve. As Ronald Reagan famously said, "доверяй, но проверяй (trust, but verify)."

ZK is the Game Changer that Can Shoulder the Burden

To bring the magic of AI to blockchain without sacrificing decentralized security, we can also rely on the verifiable capabilities of ZK.

When applied to AI, zero-knowledge cryptography allows us to verify whether AI models are executed "correctly" without revealing the secrets behind the models. In other words, we use mathematics to check whether AI results are secretly manipulated. By verifying this "correctness proof" on-chain, we can achieve security comparable to blockchain at a fraction of the cost.

Thanks to mathematics, we can now know with cryptographic certainty that your AI co-pilot has not provided secretly tampered code, your credit score has not been influenced by biased banks, and your social media information has not been affected by political bias. We call this "union" "Accountable AI".

However, cutting-edge ZK cryptography is not free. In fact, despite savings through blockchain standards, ordinary ZK proofs still add over 1000 times the cost to AI computations. In fact, over the past year of building the world's first ZKML application, we have continually encountered the same cost + performance ceiling. There is fundamentally no way to solve this problem: ZK + ML == economic violence.

That is, until we noticed that modern AI relies entirely on advancements in highly parallel processors (i.e., GPUs). If we make the same changes in ZK proofs, we will greatly open up the design space for smart contracts.

With the support of "Remainder," we have begun to build a community of developers and partners for next-generation AI-enhanced dApps. We are committed to bringing powerful AI capabilities to specific use cases at scale.

To realize this vision, we will launch multiple ZKML applications in the coming months. For example, we will first collaborate with NFT valuation leader Upshot.

Upshot uses AI to provide extremely accurate and timely NFT prices (over 100 million evaluations per hour, with an average absolute percentage error (MAPE) of 3-10%), opening up new financial markets for long-tail assets. However, currently using and subscribing to Upshot also means trusting that Upshot will never manipulate.

Our ultimate goal is to bring Upshot's AI pricing on-chain without sacrificing blockchain security. This is precisely what we offer by integrating their machine learning oracle into the Modulus API. The result is that AI superpowers and blockchain security together create NFT prices, a combination made possible by Modulus's specialized ZK proofs.

Modulus Team

According to The Block, Modulus was born at Stanford University. Co-founder Daniel Shorr, like many young people in their 20s during the pandemic, could not resist the allure of crypto, leading to the birth of Modulus.

According to the RootData page, Daniel Shorr, co-founder and CEO of Modulus Labs, holds both a bachelor's and master's degree from Stanford University; Nicholas Cosby is a co-founder of Modulus Labs. He previously studied programming at Coding Dojo; Ryan Cao is the CFO/CTO of Modulus Labs. He holds both a bachelor's and master's degree from Stanford University.

Daniel Shorr wrote: When Ryan, Nick, and I founded Modulus over a year ago, we had a genius plan—building something very, very difficult, and the complexity of our strategy was simply astounding! But our madness is methodical. If we engage in things that people think are nearly impossible, but do so with absolute sincerity and seriousness—then we will attract the most talented and kindest people in the world to work together.