Technical
Technical AI hazards are the root causes of technical deficiencies in the AI system. An example of such an AI hazard is overfitting, which describes a model’s excessive adaptation to the training dataset. Quantitative methods to assess (metrics) and treat (mitigation means) exist for technical AI hazards, which might be performed automatically. In case of overfitting, metrics are based on the comparison of performance between the training and validation datasets, and mitigation means may include regularization techniques, among others.
ENTITY
2 - AI
INTENT
2 - Unintentional
TIMING
1 - Pre-deployment
Risk ID
mit1017
Domain lineage
7. AI System Safety, Failures, & Limitations
7.3 > Lack of capability or robustness
Mitigation strategy
1. Implement rigorous data augmentation protocols and, where feasible, expand the size and diversity of the training dataset. This addresses the fundamental root cause of overfitting by ensuring the model is exposed to a sufficiently representative and varied input distribution, thus inhibiting excessive adaptation to noise within the original training samples. 2. Apply advanced regularization techniques, such as L1, L2 (Weight Decay), and Dropout, to the model's objective function. This systematically introduces a penalty term that discourages large parameter weights and limits the model's effective complexity, thereby promoting the learning of generalizable features over the memorization of training-specific artifacts. 3. Institute early stopping mechanisms during the training phase. This involves continuous monitoring of the model's performance on a dedicated, independent validation set and halting the optimization process immediately upon the onset of validation performance stagnation or degradation. This is a crucial intervention to prevent over-optimization past the point of maximal generalization capability.