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

Fairness

The general principle of equal treatment requires that an AI system upholds the principle of fairness, both ethically and legally. This means that the same facts are treated equally for each person unless there is an objective justification for unequal treatment.

Source: MIT AI Risk Repositorymit179

ENTITY

2 - AI

INTENT

2 - Unintentional

TIMING

2 - Post-deployment

Risk ID

mit179

Domain lineage

1. Discrimination & Toxicity

156 mapped risks

1.1 > Unfair discrimination and misrepresentation

Mitigation strategy

1. Implement rigorous, integrated bias audits and ethical data curation during the conception and pre-processing phases to ensure all training datasets are diverse, representative, and free of historical biases, thereby mitigating data-driven discrimination at its source. 2. Apply in-processing fairness-aware algorithms, such as adversarial debiasing or constraint-based optimization, to embed fairness constraints directly into the model training process, actively reducing algorithmic bias against sensitive groups. 3. Establish continuous fairness monitoring systems and robust human-in-the-loop mechanisms post-deployment to track statistical fairness metrics (e.g., equal opportunity difference) in real-time, detect bias drift, and allow for expert human review and override of potentially discriminatory AI-generated decisions.