[Paper Review] AI Ethics Issues in Real World: Evidence from AI Incident Database

Mengyi Wei, Zhixuan Zhou|arXiv (Cornell University)|Jun 15, 2022

Ethics and Social Impacts of AI28 citations

TL;DR

The paper builds the first taxonomy of AI ethics issues in real-world incidents using the AI Incident Database, identifying 13 application areas and 8 ethics-issue categories, with 150 incidents analyzed from 2010–2021.

ABSTRACT

With the powerful performance of Artificial Intelligence (AI) also comes prevalent ethical issues. Though governments and corporations have curated multiple AI ethics guidelines to curb unethical behavior of AI, the effect has been limited, probably due to the vagueness of the guidelines. In this paper, we take a closer look at how AI ethics issues take place in real world, in order to have a more in-depth and nuanced understanding of different ethical issues as well as their social impact. With a content analysis of AI Incident Database, which is an effort to prevent repeated real world AI failures by cataloging incidents, we identified 13 application areas which often see unethical use of AI, with intelligent service robots, language/vision models and autonomous driving taking the lead. Ethical issues appear in 8 different forms, from inappropriate use and racial discrimination, to physical safety and unfair algorithm. With this taxonomy of AI ethics issues, we aim to provide AI practitioners with a practical guideline when trying to deploy AI applications ethically.

Motivation & Objective

• Motivate the need for practical AI ethics insights beyond high-level guidelines.
• Construct a taxonomy of AI ethics issues observed in real-world incidents.
• Identify application areas where ethics issues are most prevalent.
• Provide guidance for practitioners to deploy AI more ethically based on empirical evidence.

Proposed method

• Collect 150 AI ethics incidents from the AI Incident Database.
• Perform conventional content analysis to derive categories and application areas.
• Code data independently by two researchers with Krippendorff’s alpha reliability checks.
• Present a taxonomy of application areas (13) and ethics issues (8).
• Relate incident findings to existing AI ethics guidelines and discuss implications.

Experimental results

Research questions

• RQ1 What real-world AI ethics issues occur across different application areas?
• RQ2 How are AI ethics issues distributed temporally and geographically?
• RQ3 How do observed ethics issues map onto existing AI ethics guidelines?
• RQ4 What practical guidance can be drawn for ethical design and deployment of AI systems?

Key findings

• Incidents span 2010–2021, with a peak in 2016 and another peak in 2020.
• Most incidents occur in the US, China, and the UK (together 89 of 150).
• Thirteen application areas are identified, with intelligent service robots, language/vision models, and autonomous driving leading.
• Eight ethics-issue categories are identified, with inappropriate use (bad performance), racial discrimination, physical safety, and unfair algorithm as prominent issues.
• In each application area, specific ethics issues show distinct patterns, such as physical safety dominating autonomous driving and racial discrimination being prominent in language/vision models.
• There is alignment between observed real-world issues and prevalent principles in AI ethics guidelines (transparency, justice/fairness, non-maleficence), though guidelines remain abstract.