Back to the MIT repository
5. Human-Computer Interaction2 - Post-deployment

Long-term effects of AI model biases on user judgment

The initial user exposure to model biases can have a lasting impact beyond the initial interaction with the model. Users who encounter biases in AI models can be affected by and continue to exhibit previously encountered biases in their decision-making, even after they stop using the models [207].

Source: MIT AI Risk Repositorymit1203

ENTITY

2 - AI

INTENT

2 - Unintentional

TIMING

2 - Post-deployment

Risk ID

mit1203

Domain lineage

5. Human-Computer Interaction

92 mapped risks

5.2 > Loss of human agency and autonomy

Mitigation strategy

1. Implement rigorous pre- and in-processing bias mitigation techniques, such as Fair Representation Learning or Adversarial Debiasing, to systematically minimize algorithmic and data-driven bias across sensitive subgroups before model deployment. 2. Establish a mandatory transparency and "Human-in-the-Loop" (HIL) framework that provides users with clear, actionable explanations for AI outputs, enabling critical evaluation and override of potentially biased recommendations to safeguard individual autonomy and judgment. 3. Conduct continuous and automated post-deployment bias auditing using fairness metrics to detect emergent bias drift, coupled with an operational process for immediate model retraining or adjustment to sustain equitable performance over time.