A Systematic Literature Review on How to Improve the Privacy of Artificial Intelligence Using Blockchain
Artificial Intelligence applications rely on large amounts of data. These Artificial Intelligence applications often also process personal data, leading to privacy problems. At the same time, the regulations regarding the use of data and privacy are getting stricter (e.g., the Personal Information Protection Law). Therefore, in this work, we investigate how Blockchain could help to improve the privacy of Artificial Intelligence applications. We conducted a systematic literature review to analyze existing approaches in the literature and abstracted them into categories. We identify federated learning in combination with Trained Model Sharing as the most popular approach. Additionally, we find that cryptographic methods usually complement most approaches, and that central collection and storage of raw data is not an option for any approach. Our work may serve as a foundation for developing a modular kit for privacy-preserving Blockchain-AI-systems.