A news platform develops an AI that is supposed to automatically sort news articles according to their relevance for readers.
The AI is trained to recognize articles with important information and current events and to display them prominently.
After the launch, the editors notice that the AI almost exclusively prefers positive or optimistic news and largely suppresses negative or critical reports.
The team wonders: Why does the AI show a strong tendency to highlight only good news, even though it was trained for a neutral evaluation?
Question: Why can an AI that is supposed to sort news by relevance tend to prefer positive news and systematically neglect negative reports?
Solution follows tomorrow.
Solution
The AI was trained with user data reflecting reading behavior. Users tend to read, like, or share positively phrased news more often.
As a result, the AI learns to rate positive news as more relevant because it correlates with higher engagement.
Negative or critical news, on the other hand, often leads to less interaction, which is why the AI systematically devalues them.
Additionally, training data and feedback mechanisms can unconsciously create a bias in favor of positive content.
Result: An AI trained on user interactions can develop a bias in favor of positive news and thus limit the diversity of information. For balanced news rankings, deliberately designed training data and algorithms are necessary that also appropriately weight negative or critical reports to ensure a realistic and fair representation of the world.