# opp/ai: Optimistic Privacy-Preserving AI on Blockchain

**Authors:** Cathie So, KD Conway, Xiaohang Yu, Suning Yao, Kartin Wong  
**arXiv:** [2402.15006](https://arxiv.org/abs/2402.15006)  
**Date:** 2024-02-22  
**Summary page:** https://www.summarizepaper.com/en/paper/2402.15006v1/

## Key points

- The Opp/AI Framework is a pioneering solution developed by Cathie So and her team at ORA to address challenges of AI and blockchain convergence.
- It aims to provide decentralized, secure, and efficient AI services on blockchain platforms while balancing privacy protection and computational efficiency.
- The framework integrates Zero-Knowledge Machine Learning (zkML) for privacy and Optimistic Machine Learning (opML) for efficiency to create a hybrid model tailored for blockchain AI services.
- Researchers highlight the potential of Opp/AI in revolutionizing model privacy based on large-scale open-source models.
- With cloud computing platforms like AWS providing high-memory instances, Opp/AI brings the possibility of achieving model privacy closer to reality in the near future.
- The framework is demonstrated as framework agnostic through benchmarking on two zkML frameworks, showcasing its versatility across various model architectures.
- Overall, Opp/AI represents a significant advancement in addressing challenges posed by AI on blockchain platforms with its unique combination of zkML for privacy protection and opML for computational efficiency.

## Summary

The Opp/AI Framework, developed by Cathie So and her team at ORA, is a pioneering solution that addresses the challenges posed by the convergence of Artificial Intelligence (AI) and blockchain technology. This innovative framework aims to provide decentralized, secure, and efficient AI services on blockchain platforms while balancing privacy protection and computational efficiency. It integrates Zero-Knowledge Machine Learning (zkML) for privacy and Optimistic Machine Learning (opML) for efficiency to create a hybrid model tailored for blockchain AI services. In their study, the authors delve into the privacy features of zkML and assess the performance and adaptability of the Opp/AI Framework across different scenarios. They highlight its potential in revolutionizing model privacy on fine-tuned models based on large-scale open-source models. With cloud computing platforms like AWS providing high-memory instances, Opp/AI brings the possibility of achieving model privacy closer to reality in the near future. Furthermore, the researchers emphasize that the Opp/AI Framework is framework agnostic as demonstrated through benchmarking conducted on two zkML frameworks. This flexibility allows for its implementation across various model architectures, showcasing its versatility in ensuring data privacy while maintaining computational efficiency. Overall, Opp/AI represents a significant advancement in addressing the challenges posed by AI on blockchain platforms. Its unique combination of zkML for privacy protection and opML for computational efficiency sets a new standard for secure and efficient AI services in the digital world.