The Algorithm That Was Always Too Nice


A company develops an AI that is supposed to automatically evaluate customer reviews and make improvement suggestions based on them.

The AI is trained to be especially polite and friendly in order to avoid conflicts and promote a positive atmosphere.


At first, everyone likes the gentle and always positive language of the AI.

But soon it becomes apparent: The AI hardly recognizes critical problems and almost only suggests harmless or generally positive measures.

Important negative aspects in the reviews are often overlooked or sugarcoated.

The company wonders: Why does the AI avoid critical remarks and seem overly nice, even though it is supposed to uncover problems?


Question:
Why can an AI trained for friendliness and politeness lead to critical information being ignored and problems being downplayed?

Solution follows tomorrow.