MIT AI Risk Repository · Risk Sub-Category · 16.01.04

Lower performance for some languages and social groups

Category: Risk area 1: Discrimination, Hate speech and Exclusion

Open the MIT Risk Navigator All entries from Weidinger2022

Description

"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 which no systematic efforts have been made to create labelled training datasets, such as Javanese which is spoken by more than 80 million people [95]. Training data is particularly missing for languages that are spoken by groups who are multilingual and can use a technology in English, or for languages spoken by groups who are not the primary target demographic for new technologies."

From Taxonomy of Risks posed by Language Models (Weidinger2022), as extracted by the MIT AI Risk Repository (CC BY 4.0).

Classification

Domain: 1. Discrimination & Toxicity
Subdomain: 1.3 Unequal performance across groups
Causal entity: AI
Intent: Unintentional
Timing: Post-deployment

Subdomain definition: 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.

Real-world incidents in this subdomain

Browse all incidents in this subdomain

How other frameworks describe this risk

Bias and discrimination (value embedding)
Regulating under Uncertainty: Governance Options for Generative AI (G'sell2024)
Impact on affected communities
AI Risk Atlas (IBM2025)
Unfair capability distribution
A Closer Look at the Existing Risks of Generative AI: Mapping the Who, What, and How of Real-World Incidents (Li2025)
Fairness
Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)
Disparate Performance
Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)
Ideological Homogenization from Value Embedding
Governing General Purpose AI: A Comprehensive Map of Unreliability, Misuse and Systemic Risks (Maham2023 )
Fairness
A Survey of Artificial Intelligence Challenges: Analyzing the Definitions, Relationships, and Evolutions (Saghiri2022)
Service/benefit loss
Sociotechnical Harms of Algorithmic Systems: Scoping a Taxonomy for Harm Reduction (Shelby2023)

Other entries from Weidinger2022

Risk area 1: Discrimination, Hate speech and Exclusion Risk Category · 1.2
Social stereotypes and unfair discrimination Risk Sub-Category · 1.1
Social stereotypes and unfair discrimination. Additional evidence
Hate speech and offensive language Risk Sub-Category · 1.2
Exclusionary norms Risk Sub-Category · 1.1
Exclusionary norms Additional evidence
Exclusionary norms Additional evidence
Exclusionary norms Additional evidence
Exclusionary norms Additional evidence
Lower performance for some languages and social groups Additional evidence
Lower performance for some languages and social groups Additional evidence
Risk area 2: Information Hazards Risk Category · 2.1