Towards algorithm auditing: managing legal, ethical and technological risks of AI, ML and associated algorithms

Area of Science:

• Computer Science
• Ethics in AI
• Regulatory Compliance

Background:

• Businesses increasingly rely on algorithms, facing risks of financial and reputational damage from biased or unsafe AI.
• High-profile incidents highlight the need for accountability in AI, including issues with photo classification, chatbots, and recruiting tools.

Purpose of the Study:

• To survey key areas for performing auditing and assurance of algorithms.
• To instigate debate and establish a framework for a new industry focused on Auditing and Assurance of Algorithms.

Main Methods:

• Literature review of existing AI ethics and governance challenges.
• Conceptual framework development for algorithm auditing and assurance.
• Stakeholder analysis including policymakers, industry practitioners, and developers.

Main Results:

• Identified the necessity of formal assurance for algorithms to ensure legality, ethicality, and safety.
• Proposed the establishment of an Auditing and Assurance of Algorithms industry, akin to financial auditing.
• Anticipated the development of auditing levels and frameworks to guide governance and compliance.

Conclusions:

• Algorithm audits are crucial for mitigating risks associated with ubiquitous AI adoption.
• A new professionalized industry for Auditing and Assurance of Algorithms is envisaged.
• The proposed framework will inform governance and regulatory compliance for AI systems.