A company develops an AI that is supposed to automatically summarize product reviews.
The AI compares new reviews with existing ones to quickly derive an overall rating.
At first, this works well because it finds patterns and classifies reviews.
But soon it becomes apparent: The AI always rates new, unique, or special opinions as less relevant or even wrong.
Unusual reviews that deviate from previous ones are heavily downgraded or ignored.
The company wonders: Why can an AI that uses comparisons as a basis not adequately consider new or divergent viewpoints?
Question: Why does an AI trained on comparison with existing data tend to neglect or misjudge unusual or novel reviews?
Solution follows tomorrow.
Solution
The AI was trained to evaluate and classify new reviews based on their similarity to already existing ones.
As a result, it learns that reviews that strongly deviate from established patterns are interpreted as outliers or errors.
Unusual or new perspectives are rated as less likely or less relevant because they do not fit the previous data.
This bias arises from the training principle, which is based on comparison and pattern recognition, without explicitly allowing room for innovation or divergence.
Result: An AI that always compares everything only with the known cannot adequately appreciate new or divergent opinions. For a comprehensive and fair evaluation, such systems must learn to also recognize unusual viewpoints as potentially valuable and not automatically downgrade them.