# Providing Assurance and Scrutability on Shared Data and Machine Learning Models with Verifiable Credentials

**Authors:** Iain Barclay, Alun Preece, Ian Taylor, Swapna K. Radha, Jarek Nabrzyski  
**arXiv:** [2105.06370](https://arxiv.org/abs/2105.06370)  
**Date:** 2021-05-13  
**Summary page:** https://www.summarizepaper.com/en/paper/2105.06370v1/

## Key points

- The paper introduces a software architecture and implementation to address trust in shared data resources for AI systems and ML models.
- Practitioners in fields like healthcare and finance often lack insight into potential problems with adopted datasets.
- The authors propose a system based on self-sovereign identity design patterns.
- Scientists can issue signed credentials attesting to the qualities of their data resources, recorded in a bill of materials (BOM).
- The BOM is stored with the ML model as a verifiable credential, providing traceable record of its supply chain.
- An AI Scrutineer tool utilizes the verified BOM and certified data qualities to provide practitioners with insights into model constituents.
- The approach leverages self-sovereign identity principles and verifiable credentials to enhance trust and transparency in shared data resources for AI systems and ML models.

## Summary

The paper titled "Providing Assurance and Scrutinability on Shared Data and Machine Learning Models with Verifiable Credentials" introduces a software architecture and implementation that addresses the issue of trust in shared data resources used in the development of artificial intelligence (AI) systems or machine learning (ML) models. The trust placed in the originators of the data extends to practitioners in fields like healthcare and finance who rely on AI developers to have used relevant and trustworthy data. However, these practitioners often lack insight into potential problems that may arise from adopted datasets such as biased data or discredited sources. To address this challenge, the authors propose a system based on design patterns from the field of self-sovereign identity. Scientists can issue signed credentials attesting to the qualities of their data resources which are then recorded in a bill of materials (BOM). This BOM is stored with the ML model as a verifiable credential providing traceable record of its supply chain for an AI system. The verified BOM along with its linkage to certified data qualities is utilized by an AI Scrutineer; a web-based tool designed to provide practitioners with insight into model constituents. Overall, this paper presents an innovative approach to enhance trust and transparency in shared data resources used for AI systems and ML models by leveraging self-sovereign identity principles and verifiable credentials. Scientists can provide assurance regarding their data's quality while practitioners gain valuable insights into model constituents through tools like AI Scrutineer.