Algorithmic Bias

TECHNICAL DEFINITION

Algorithmic Bias refers to systematic and repeatable errors in an artificial intelligence system that create unfair or discriminatory outcomes, often stemming from biased training data, flawed model design, or inappropriate deployment contexts.

BACKGROUND

Algorithmic bias describes systematic and repeatable harmful tendency in a computerized sociotechnical system to create "unfair" outcomes, such as "privileging" one category over another in ways that may or may not be different from the intended function of the algorithm.

SYNONYMS & ALIASES

• AI bias
• data bias
• systemic bias
• unfair algorithm

USAGE NOTE

Algorithmic bias can lead to real-world harm, such as discriminatory loan approvals or hiring practices.

DEVELOPERS

Organizations developing technology related to Algorithmic Bias.

• IBM Research

  Developed the AI Fairness 360 (AIF360) open-source toolkit to help detect and mitigate bias in AI models throughout the AI lifecycle.

• Google AI

  Conducts extensive research on fairness, interpretability, and privacy in AI, developing methods and tools to identify and address algorithmic bias in their products and platforms.

• Microsoft Research

  Focuses on responsible AI principles, developing tools like Fairlearn for assessing and mitigating unfairness in AI systems, and providing guidelines for ethical AI development.

• Amazon Web Services (AWS)

  Provides tools within its SageMaker platform, such as SageMaker Clarify, to detect potential bias in machine learning models during various stages of development, aiding AI engineers in building fairer systems.

• OpenAI

  Actively researches and implements methods to reduce algorithmic bias and promote fairness in its advanced AI models, influencing prompt design and model deployment strategies for responsible use.

• Meta AI (FAIR)

  Engages in research to understand, detect, and mitigate various forms of bias in large-scale AI models and datasets, contributing to best practices in AI engineering.

• Allen Institute for AI (AI2)

  A non-profit research institute that conducts fundamental AI research, including work on ethical AI, fairness, and the detection of algorithmic bias across different AI applications.

• Hugging Face

  Provides an open-source platform and resources for AI development, including datasets and model cards that aid in identifying and understanding potential biases in models, supporting responsible AI engineering and prompt design considerations.