Bias
A systematic difference in how an AI system treats or represents people, groups, objects, or outcomes that can lead to unfair or discriminatory results.
How the laws define this
Regulators uniformly view bias as systematic errors or undesirable outcomes in AI outputs that lead to unfair, unequal, or discriminatory treatment, requiring mitigation through design, oversight, or impact assessments. However, these frameworks diverge regarding the specific contexts and populations they prioritize. The Artificial Intelligence Regulation Bill (Legislative Assembly) specifically targets harms against protected groups, whereas Considerations for the Use of Artificial Intelligence To Support Regulatory Decision-Making for Drug and Biological Products emphasizes inaccurate or unfair results across diverse patient populations. Additionally, the Plan de Acción para la Política Nacional de Inteligencia Artificial (Action Plan of the National AI Policy) uniquely isolates root causes—tracing bias to data, model design, labeling, or deployment contexts as a rights-related risk—while the Resolución de Presidencia N° 206/2025 - Anexo I “Guía de Recomendaciones y Directrices para el Uso de Sistemas de Inteligencia Artificial (IA) en el Poder Judicial de la Ciudad Autónoma de Buenos Aires” focuses on preventing systems from generating, reproducing, reinforcing, or perpetuating unfair treatment.
Synthesised from the 4 statutory definitions below. Each one is quoted in full, with its source.
How jurisdictions differ
What is distinctive in each definition, binding law in force first.
| Jurisdiction | Instrument | Status | What is distinctive |
|---|---|---|---|
| Argentina | City of Buenos Aires Presidency Resolution No. 206/2025 - Annex I 'Guidelines for the Use of Artificial Intelligence (AI) Systems in the Judiciary of the Autonomous City of Buenos Aires' | In Force | Frames bias solely as undesirable outcomes to be prevented through careful design and oversight in judicial AI use. |
| Chile | Action Plan of the National AI Policy | In Force | Specifies causes like data, model design, labeling, or deployment contexts and frames it as a rights-related risk to detect, measure, and mitigate. |
| United States | Considerations for the Use of Artificial Intelligence To Support Regulatory Decision-Making for Drug and Biological Products | Draft | Implicitly refers to bias as systematic errors in AI model outputs that could lead to unfair or inaccurate results across diverse patient populations. |
| Costa Rica | Law Regulating Artificial Intelligence in Costa Rica | Under Review | Requires identification, assessment, and mitigation of systematic errors producing unfair outcomes against protected groups via prevention measures and impact assessments. |
What it triggers
- The guidelines of the City of Buenos Aires Presidency aim to prevent bias through careful design and oversight.source
- Chile's Action Plan of the National AI Policy frames bias as a rights-related risk to be detected, measured, and mitigated.source
- Under Costa Rica's Law Regulating Artificial Intelligence, the regulation would require bias to be identified, assessed, and mitigated through bias prevention measures and impact assessments.source
Definitions (4)
Systematic errors in AI systems that produce unfair or discriminatory outcomes, particularly against protected groups, which the regulation requires to be identified, assessed, and mitigated through bias prevention measures and impact assessments.
Systematic distortions in AI outputs or decision-making that produce unequal, unfair, or discriminatory outcomes for individuals or groups, arising from data, model design, labeling, or deployment contexts; the Plan frames bias as a rights-related risk to be detected, measured and mitigated.
Understood as undesirable outcomes where AI systems might generate, reproduce, reinforce, or perpetuate unfair treatment. The guidelines aim to prevent this through careful design and oversight.
The guidance also implicitly touches upon concepts like Bias, referring to systematic errors in an AI model's output that could lead to unfair or inaccurate results, particularly across diverse patient populations.
Not to be confused with
Fairness refers to the absence of bias or discrimination, while bias is the systematic difference or error that causes unfair treatment.
Algorithmic discrimination is the unfair differential treatment caused by AI systems, often unlawful, whereas bias is the underlying systematic difference leading to such outcomes.
Questions and answers
What causes bias in AI according to Chile's National AI Policy?
Bias arises from systematic distortions in data, model design, labeling, or deployment contexts. The Action Plan frames it as a rights-related risk.
How does the Buenos Aires judiciary guideline address bias?
It understands bias as undesirable outcomes and aims to prevent it through careful design and oversight of AI systems.
What obligations does Costa Rica's AI bill propose for bias?
It would require systematic errors causing unfair outcomes, especially against protected groups, to be identified, assessed, and mitigated using prevention measures and impact assessments.
Related Terms
Data Quality
The fitness of dataset(s) for an AI system’s intended purpose, judged by attributes like relevance, accuracy, representativeness, completeness and provenance....
Fairness
Absence of unjustified bias or discrimination in AI outputs and equitable treatment and access across individuals and groups....
Bias Mitigation
Measures to identify and reduce discriminatory outcomes in AI systems....
Algorithmic Discrimination
Unfair or unlawful differential treatment or impact caused by an AI or algorithmic system against individuals or groups, often on protected characteristics....
Training Data
Data used to develop and train AI models to perform their intended functions....