# Blockchain Framework for Artificial Intelligence Computation

**Authors:** Jie You  
**arXiv:** [2202.11264](https://arxiv.org/abs/2202.11264)  
**Date:** 2022-02-23  
**Summary page:** https://www.summarizepaper.com/en/paper/2202.11264v1/

## Key points

- The authors propose a blockchain framework for AI computation that can revolutionize industrial applications.
- Blockchain is a distributed database that records all transactions or digital events among participating parties.
- Proof-of-work, which requires solving a hard mathematical puzzle, is energy-wasteful when many computational nodes are installed in the blockchain competing to approve records.
- The authors pose proof-of-work as a reinforcement-learning problem by modeling the blockchain growing as a Markov decision process.
- The block verification and consensus mechanism are designed as a deep reinforcement-learning iteration process that utilizes state transition determination and randomness of action selection of a Markov decision process, along with computational complexity of deep neural networks to make blocks not easy to recompute and preserve transaction order.
- This approach allows blockchain nodes to train the same deep neural network with different data samples (state-action pairs) in parallel, allowing models to experience multiple episodes across computing nodes at one time.
- This method spares computational resources for industrial applications while also encouraging data sharing and AI model design for common problems.
- The proposed framework has significant potential for shaping future public blockchain networks and enabling efficient data sharing among stakeholders.

## Summary

In this paper, the authors propose a blockchain framework for artificial intelligence (AI) computation that has the potential to revolutionize industrial applications. Blockchain is a distributed database that records all transactions or digital events among participating parties. Each transaction in the records is approved and verified by consensus of the participants in the system, which requires solving a hard mathematical puzzle known as proof-of-work. However, having many computational nodes installed in the blockchain competing to approve records by just solving a meaningless puzzle is energy-wasteful. To address this issue, the authors pose proof-of-work as a reinforcement-learning problem by modeling the blockchain growing as a Markov decision process. In this process, a learning agent makes an optimal decision over the environment's state while a new block is added and verified. The block verification and consensus mechanism are designed as a deep reinforcement-learning iteration process that utilizes state transition determination and randomness of action selection of a Markov decision process, along with computational complexity of deep neural networks to make blocks not easy to recompute and preserve transaction order. The proposed method allows blockchain nodes to train the same deep neural network with different data samples (state-action pairs) in parallel, allowing models to experience multiple episodes across computing nodes at one time. This approach not only spares computational resources for industrial applications but also encourages data sharing and AI model design for common problems. Overall, this blockchain framework presents an avenue for AI advancement in industrial applications while minimizing environmental detriment caused by excessive energy consumption. It has significant potential for shaping future public blockchain networks and enabling efficient data sharing among stakeholders.