Increasing Trust in AI Services through Supplier's Declarations of Conformity

By Michael Hind, Sameep Mehta, Aleksandra Mojsilovic, Ravi Nair, Karthikeyan Natesan Ramamurthy, Alexandra Olteanu, Kush R. Varshney. 2018

136 citations recorded by the source corpus.

The accuracy and reliability of machine learning algorithms are an important concern for suppliers of artificial intelligence (AI) services, but considerations beyond accuracy, such as safety, security, and provenance, are also critical elements to engender consumers' trust in a service. In this paper, we propose a supplier's declaration of conformity (SDoC) for AI services to help increase trust in AI services. An SDoC is a transparent, standardized, but often not legally required, document used in many industries and sectors to describe the lineage of a product along with the safety and performance testing it has undergone. We envision an SDoC for AI services to contain purpose, performance, safety, security, and provenance information to be completed and voluntarily released by AI service providers for examination by consumers. Importantly, it conveys product-level rather than component-level functional testing. We suggest a set of declaration items tailored to AI and provide examples for two fictitious AI services.

Subfields: Safety Research, Statistics and Probability, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Safety Research

In Statistics and Probability

In Safety, Robustness & Interpretability

Citations in the corpus

Model Cards for Model Reporting (2018, background): “[28], Data Statements [3], or Factsheets [27], describing datasets that the model was trained and evaluated on.”
Datasheets for datasets (2018, background): “[12] Michael Hind, Sameep Mehta, Aleksandra Mojsilovic, Ravi Nair, Karthikeyan Natesan Ramamurthy, Alexandra Olteanu, and Kush R.”
Saving Face: Investigating the Ethical Concerns of Facial Recognition Auditing (2020, background): “While comprehensive auditing frameworks examining model development and deployment processes are yet to be developed, documentation proposals such as datasheets [20], model cards [34] and factsheets [24] can encour-”
Social Data: Biases, Methodological Pitfalls, and Ethical Boundaries (2019, background): “…of such registries in the form of model cards that focus on documenting the creation of pre-trained models (Mitchell et al., 2019) or supplier’s declarations of conformity to describe the lineage of AI services that can be “an amalgam of many models trained on many datasets” (Hind et al., 2018).”
In AI We Trust? Factors That Influence Trustworthiness of AI-infused Decision-Making Processes (2019, background): “(P75, ttd=absent) This result echoes recent calls for including a set of factual information about how AI models were trained and tested when distributing those models [16, 30].”
Garbage in, garbage out?: do machine learning application papers in social computing report where human-labeled training data comes from? (2019, neutral): “A number of proposed standards have been created around datasets for ML [2–4, 14, 22, 23, 39], which are often framed as potential ways to mitigate bias and improve transparency and accountability.”
Data Cards: Purposeful and Transparent Dataset Documentation for Responsible AI (2022, methodology): “Data Cards complement other longer-form and domain-specific documentation frameworks for ethical reporting (See Appendix A) , such as Model Cards [23], Data Statements [9], Datasheets for Datasets [15], and [6] FactSheets.”
ABOUT ML: Annotation and Benchmarking on Understanding and Transparency of Machine Learning Lifecycles (2019, background): “…Data Protection Regulation (GDPR), government procurement processes like Canada’s Algorithmic Impact Assessment, and engineering practice at IBM [6], Microsoft [4] and Google [11] in particular highlight the shift towards documentation-based approaches to transparency as a practical mechanism…”
Bias Mitigation Post-processing for Individual and Group Fairness (2018, background): “Without bias mitigation, models trained on such decisions perpetuate and scale human biases and are thereby unsafe and untrustworthy [3, 4].”
FactSheets: Increasing Trust in AI Services through Supplier's Declarations of Conformity. (2018, methodology): “Subsequent to the first posting of this paper [13], Mitchell et al.”
Trust in AI: progress, challenges, and future directions (2024, unknown): “It also explains how the system was created, trained, and deployed, along with what scenarios it was tested on, how it will respond to non-tested scenarios, guidelines that specify what tasks it should and should not be used for, and any ethical concerns of its use (Hind et al., 2018).”
Transparency and trust in artificial intelligence systems (2020, methodology): “6 To create trust in AI-assisted tools, Hind et al. (2018) propose to include a supplier declaration of conformity (SDoC) which is a long-established procedure in other sectors such as transportation and telecommunication.”
Adaptive Sampling Strategies to Construct Equitable Training Datasets (2022, background): “Finally, given an existing dataset, many proposals have suggested ways to characterize their equity and aid appropriate use, for example by including statements describing what populations the datasets are and are not representative of [7, 28, 32, 33].”
Software Engineering for AI-Based Systems: A Survey (2021, methodology): “incorporate traceability into their requirements specification approach [12].”
Trustworthy AI: From Principles to Practices (2021, methodology): “Some studies [22, 129, 156, 246, 273] have proposed standard practices to guide and evaluate the documentation of an AI system.”