Back to the MIT repository
6. Socioeconomic and Environmental2 - Post-deployment

Algorithmic monoculture

The dominance of specific AI models could lead to a lack of diversity in approaches, amplifying systemic risks if these models fail.

Source: MIT AI Risk Repositorymit1051

ENTITY

2 - AI

INTENT

2 - Unintentional

TIMING

2 - Post-deployment

Risk ID

mit1051

Domain lineage

6. Socioeconomic and Environmental

262 mapped risks

6.1 > Power centralization and unfair distribution of benefits

Mitigation strategy

1. Establish Regulatory and Technical Requirements for Algorithmic Pluralism and Diversity Implement governance frameworks and policy interventions to mandate the adoption of structurally diverse AI models and ensembles across high-stakes domains, reducing reliance on shared foundational components to mitigate the risk of systemic correlated failures. 2. Enforce Strict Data Diversity and Representation Standards Require comprehensive, auditable protocols for AI system development to ensure the use of diverse, representative, and appropriately sourced training data, thereby counteracting the component-sharing monoculture that exacerbates outcome homogenization and systemic bias. 3. Implement Independent Pre-Deployment Audits and Transparency Mechanisms Systematically require third-party pre-deployment model audits and risk assessments focused on systemic vulnerabilities, coupled with standardized transparency and interpretability measures, to allow for external scrutiny and proactive mitigation of potential arbitrariness before widespread societal deployment.