MIT AI Risk Repository · domain 1: Discrimination & Toxicity

1.3 Unequal performance across groups

Accuracy and effectiveness of AI decisions and actions is dependent on group membership, where decisions in AI system design and biased training data lead to unequal outcomes, reduced benefits, increased effort, and alienation of users.

Risk entries
17
Frameworks citing it
12
Recorded incidents
34
Incidents since 2020
26
Causal entity (risk entries)
Causal entity (risk entries) 10 0 AI: 10 AI 10 Human: 4 Human 4 Other: 3 Other 3
Causal entity (risk entries)
LabelValue
AI10
Human4
Other3
Intent (risk entries)
Intent (risk entries) 15 0 Unintentional: 15 Unintentional 15 Intentional: 1 Intentional 1 Other: 1 Other 1
Intent (risk entries)
LabelValue
Unintentional15
Intentional1
Other1
Timing (risk entries)
Timing (risk entries) 9 0 Post-deployment: 9 Post-deployment 9 Other: 5 Other 5 Pre-deployment: 3 Pre-deployment 3
Timing (risk entries)
LabelValue
Post-deployment9
Other5
Pre-deployment3
Recorded incidents per yearIncident date; current year partial
Recorded incidents per year 9 0 2016: 1 2016 1 2017: 1 2017 1 2018: 2 2018 2 2019: 4 2019 4 2020: 9 2020 9 2021: 3 2021 3 2022: 5 2022 5 2023: 1 2023 1 2024: 6 2024 6 2025: 1 2025 1 2026: 1 2026 1
Recorded incidents per year
LabelValue
20161
20171
20182
20194
20209
20213
20225
20231
20246
20251
20261
Entries by levelRisk categories, subcategories and additional evidence coded to this subdomain
Entries by level 13 0 Risk Category: 4 Risk Category 4 Risk Sub-Category: 13 Risk Sub-Category 13
Entries by level
LabelValue
Risk Category4
Risk Sub-Category13
  • Bias and discrimination (value embedding)

    "Generative AI models may also be subject to the “value embedding” phenomenon.361 “Value embedding” refers to the fact that developers of generative AI models strive to minimize biased outputs by retr...

    Regulating under Uncertainty: Governance Options for Generative AI (G'sell2024) · Human · Unintentional · Pre-deployment

  • Impact on affected communities

    "It is important to include the perspectives or concerns of communities that are affected by model outcomes when designing and building models. Failing to include these perspectives makes it difficult...

    AI Risk Atlas (IBM2025) · Human · Unintentional · Post-deployment

  • Unfair capability distribution

    "Performing worse for some groups than others in a way that harms the worse-off group"

    A Closer Look at the Existing Risks of Generative AI: Mapping the Who, What, and How of Real-World Incidents (Li2025) · AI · Unintentional · Post-deployment

  • Fairness

    Avoiding bias and ensuring no disparate performance

    Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024) · AI · Unintentional · Other

  • Disparate Performance

    The LLM’s performances can differ significantly across different groups of users. For example, the question-answering capability showed significant performance differences across different racial and...

    Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024) · AI · Unintentional · Other

  • Ideological Homogenization from Value Embedding

    "The increasing integration of general purpose AI models into every-day life raises concerns around their embedded normative values. The reach of a small number of AI models to a large number of peopl...

    Governing General Purpose AI: A Comprehensive Map of Unreliability, Misuse and Systemic Risks (Maham2023 ) · Human · Intentional · Pre-deployment

  • Fairness

    This challenge appears when the learning model leads to a decision that is biased to some sensitive attributes... data itself could be biased, which results in unfair decisions. Therefore, this proble...

    A Survey of Artificial Intelligence Challenges: Analyzing the Definitions, Relationships, and Evolutions (Saghiri2022) · AI · Unintentional · Pre-deployment

  • Erasing social groups

    people, attributes, or artifacts associated with specific social groups are systematically absent or under-represented... Design choices [143] and training data [212] influence which people and experi...

    Sociotechnical Harms of Algorithmic Systems: Scoping a Taxonomy for Harm Reduction (Shelby2023) · Human · Unintentional · Other

  • Quality-of-Service Harms

    "These harms occur when algorithmic systems disproportionately underperform for certain groups of people along social categories of difference such as disability, ethnicity, gender identity, and race....

    Sociotechnical Harms of Algorithmic Systems: Scoping a Taxonomy for Harm Reduction (Shelby2023) · AI · Unintentional · Post-deployment

  • Alienation

    Alienation is the specific self-estrangement experienced at the time of technology use, typically surfaced through interaction with systems that under-perform for marginalized individuals

    Sociotechnical Harms of Algorithmic Systems: Scoping a Taxonomy for Harm Reduction (Shelby2023) · Other · Unintentional · Post-deployment

  • Increased labor

    increased burden (e.g., time spent) or effort required by members of certain social groups to make systems or products work as well for them as others

    Sociotechnical Harms of Algorithmic Systems: Scoping a Taxonomy for Harm Reduction (Shelby2023) · Other · Unintentional · Post-deployment

  • Service/benefit loss

    degraded or total loss of benefits of using algorithmic systems with inequitable system performance based on identity

    Sociotechnical Harms of Algorithmic Systems: Scoping a Taxonomy for Harm Reduction (Shelby2023) · AI · Unintentional · Post-deployment

  • Disparate Performance

    "In the context of evaluating the impact of generative AI systems, disparate performance refers to AI systems that perform differently for different subpopulations, leading to unequal outcomes for tho...

    Evaluating the Social Impact of Generative AI Systems in Systems and Society (Solaiman2023) · AI · Unintentional · Other

  • Fairness

    "Impartial and just treatment without favouritism or discrimination."

    An Exploratory Diagnosis of Artificial Intelligence Risks for a Responsible Governance (Teixeira2022) · Other · Other · Other

  • Lower performance for some languages and social groups

    "LMs perform less well in some languages (Joshi et al., 2021; Ruder, 2020)...LM that more accurately captures the language use of one group, compared to another, may result in lower-quality language t...

    Ethical and social risks of harm from language models (Weidinger2021) · AI · Unintentional · Post-deployment

  • Lower performance for some languages and social groups

    "LMs are typically trained in few languages, and perform less well in other languages [95, 162]. In part, this is due to unavailability of training data: there are many widely spoken languages for whi...

    Taxonomy of Risks posed by Language Models (Weidinger2022) · AI · Unintentional · Post-deployment

  • Unfair capability distribution

    "Performing worse for some groups than others in a way that harms the worse-off group"

    Sociotechnical Safety Evaluation of Generative AI Systems (Weidinger2023) · AI · Unintentional · Post-deployment