Trusted Artificial Intelligence: Towards Certification of Machine\n Learning Applications

Abstract

Artificial Intelligence is one of the fastest growing technologies of the\n21st century and accompanies us in our daily lives when interacting with\ntechnical applications. However, reliance on such technical systems is crucial\nfor their widespread applicability and acceptance. The societal tools to\nexpress reliance are usually formalized by lawful regulations, i.e., standards,\nnorms, accreditations, and certificates. Therefore, the T\\"UV AUSTRIA Group in\ncooperation with the Institute for Machine Learning at the Johannes Kepler\nUniversity Linz, proposes a certification process and an audit catalog for\nMachine Learning applications. We are convinced that our approach can serve as\nthe foundation for the certification of applications that use Machine Learning\nand Deep Learning, the techniques that drive the current revolution in\nArtificial Intelligence. While certain high-risk areas, such as fully\nautonomous robots in workspaces shared with humans, are still some time away\nfrom certification, we aim to cover low-risk applications with our\ncertification procedure. Our holistic approach attempts to analyze Machine\nLearning applications from multiple perspectives to evaluate and verify the\naspects of secure software development, functional requirements, data quality,\ndata protection, and ethics. Inspired by existing work, we introduce four\ncriticality levels to map the criticality of a Machine Learning application\nregarding the impact of its decisions on people, environment, and\norganizations. Currently, the audit catalog can be applied to low-risk\napplications within the scope of supervised learning as commonly encountered in\nindustry. Guided by field experience, scientific developments, and market\ndemands, the audit catalog will be extended and modified accordingly.\n