Providing assurance and scrutability on shared data and machine learning models with verifiable credentials

Abstract

Adopting shared data resources requires scientists to place trust in the originators of the data. When shared data is later used in the development of artificial intelligence (AI) systems or machine learning (ML) models, the trust lineage extends to the users of the system, typically practitioners in fields such as healthcare and finance. Practitioners rely on AI developers to have used relevant, trustworthy data, but may have limited insight and recourse. This article introduces a software architecture and implementation of a system based on design patterns from the field of self-sovereign identity.Scientists can issue signed credentials attesting to qualities of their data resources.Data contributions to ML models are recorded in a bill of materials (BOM), which is stored with the model as a verifiable credential. The BOM provides a traceable record of the supply chain for an AI system, which facilitates on-going scrutiny of the qualities of the contributing components. The verified BOM, and its linkage to certified data qualities, is used in the AI scrutineer, a web-based tool designed to offer practitioners insight into ML model constituents and highlight any problems with adopted datasets, should they be found to have biased data or be otherwise discredited.

An Empirical Study on Software Bill of Materials: Where We Stand and the Road Ahead

“…Although some interviewees thought an AIBOM "contains only additional AI package information" (I1-Dev. ), the AI artifacts (e.g., data, code, model, configuration) also need provenance and co-versioning [27], [28]. An AIBOM (see Fig.…”

“…Back in 2019, Barclay et al [38] introduced their ideas on applying BOMs to data ecosystems for transparency and traceability, where they detailed a conceptual model to combine a BOM (static) and a bill of lots (dynamic) to jointly record the static data components and the dynamic data of a specific experiment. Based on their previous work, as a step towards operationalizing the conceptual model, Barclay et al [39] recently introduced their work using a BOM as a verifiable credential for transparency into the AI SSCs, which is a step towards AIBOM.…”

Data-Driven Intelligent Platforms—Design of Self-Sovereign Data Trust Systems

“…Distributed ledger technology (DLT) [81] supports the automatic creation of digital identities and their associated decentralised and immutable registry. Similar to the way in which physical credentials are tied to identities, verifiable credentials (VC) [82] are bound with digital identities in web environments [83], where it is more challenging to verify and validate the information since digital arrangements are more quickly falsified than their physical and biological correspondents. In this sense, to create trust in a trustless environment, VCs must be secured from a cryptographic point of view, consider privacy and be machine verifiable [82].…”

Trustable service discovery for highly dynamic decentralized workflows

“…Self-sovereign identity (SSI) [31] is terminology used to describe the ability of an individual to take ownership of their personal data and to control who has access to that data, without the need for a centralized infrastructure, or any control or authorization being required by any third party. To date, the focus of effort of SSI researchers has been on personal identity and data privacy for individuals [32], however the underlying computer science techniques can be applied to any type of entity, including digital assets such as datasets [33], and devices [34]. SSI is decentralized, and is built upon well-established cryptographic techniques whereby a user holds a private and shares a public key [35].…”