Back to the MIT repository
1. Discrimination & Toxicity2 - Post-deployment

Alienating social groups

when an image tagging system does not acknowledge the relevance of someone’s membership in a specific social group to what is depicted in one or more images

Source: MIT AI Risk Repositorymit137

ENTITY

2 - AI

INTENT

2 - Unintentional

TIMING

2 - Post-deployment

Risk ID

mit137

Domain lineage

1. Discrimination & Toxicity

156 mapped risks

1.1 > Unfair discrimination and misrepresentation

Mitigation strategy

- Implement rigorous data curation protocols to ensure training datasets are diverse and statistically representative of all target social groups, thereby counteracting the foundational cause of underrepresentation and subsequent alienation (Source 2, 13). - Establish a continuous monitoring and auditing framework for post-deployment systems to actively detect and measure representational harms, utilizing distribution-based metrics to identify skewed or inaccurate depictions of social groups in image tags (Source 2, 5). - Mandate collaborative human oversight and iterative engagement with affected social groups and ethicists to inform the definition of fairness, validate the social and cultural appropriateness of tagging outputs, and address the subjective experience of alienation (Source 2, 8).

ADDITIONAL EVIDENCE

[Lack of representation] further promotes the idea that you don't belong and perpetuates the sense of alienation