An Industry Benchmark for Data Fairness: Sony’s Alice Xiang

2026-03-10 · Source: Me, Myself, and AI · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Intermediate, extended

Summary

Alice Xiang, Global Head of AI Governance at Sony and Lead Research Scientist for AI Ethics at Sony AI, discusses the practical implementation of responsible AI. Sony established AI ethics guidelines in 2018 and has since focused on AI governance frameworks to ensure responsible evaluation of AI technologies integrated across its diverse business units, including music, motion pictures, video games, and electronics. Xiang highlights the challenge of ethically sourced data for bias evaluation, leading to Sony AI's development and public release of FHIBE (Fair Human-Centric Image Benchmark). FHIBE is an ethically sourced, globally diverse dataset designed to help practitioners evaluate and mitigate bias in human-centric computer vision models, addressing issues like skin tone and environmental factors. The benchmark aims to raise industry standards for data collection and enable more trustworthy AI development, with over 60 institutions downloading it within weeks of its release.

Key takeaway

For AI scientists and computer vision engineers developing models, prioritizing ethically sourced and diverse datasets like FHIBE is critical for identifying and mitigating algorithmic bias. Your efforts in rigorous fairness evaluation, combined with technical and non-technical mitigation strategies, will directly improve model performance across diverse populations and prevent harmful real-world impacts, moving beyond mere compliance to proactive responsibility.

Key insights

Ethical AI governance requires moving beyond principles to practical frameworks and ethically sourced data for bias evaluation.

Principles

• Ethical data sourcing is crucial for responsible AI.
• Diversity in evaluation datasets improves bias detection.
• AI governance must be a first-class design citizen.

Method

FHIBE uses self-reported demographics and extensive annotations on environment and physical attributes to enable granular diagnosis of bias causes, facilitating targeted model improvements and mitigation strategies.

In practice

• Use FHIBE to assess bias in human-centric computer vision models.
• Consider non-technical mitigations like device-level adjustments.
• Prioritize consent and compensation in data collection.

Topics

• AI Governance
• Responsible AI
• Computer Vision Bias
• FHIBE Benchmark
• Ethical Data Sourcing

Best for: Computer Vision Engineer, AI Scientist, Research Scientist, AI Ethicist, AI Architect, Director of AI/ML

Related on AIssential

• An Industry Benchmark for Data Fairness: Sony’s Alice Xiang — Ethically sourced, diverse datasets are crucial for effectively measuring and mitigating bias in AI models.
• 2025 digest of digests — 2025 saw significant AI advancements, policy shifts, and a growing focus on ethical AI and fairness.
• The AI Interview: Alice Xiang, Sony — Responsible AI demands a deeper focus on ethically sourced training data, not just model outputs, to prevent inherent biases.
• # Australia’s AI Ethics Principles Are a Start. Evidence Is the Real Test. — Practical AI ethics are challenged by dominant norm bias, which benefits powerful actors and hinders equitable outcomes.
• A principled approach for data bias mitigation — Uniform Bias (UB) offers an interpretable, mathematically guaranteed method for measuring and mitigating data bias in ML.
• AI Fairness Metrics Explained: A Practical Guide with Python — Measuring AI fairness requires selecting appropriate metrics as different definitions often conflict.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Me, Myself, and AI.