A developer intentionally inserts 5 % manipulated training examples that label real news as “fake”.
After training, the AI correctly flags 95 % of fake stories — but it also blocks 5 % of real news.
The developer argues: “Better 5 % collateral damage than disinformation.”
Question:
Is this an acceptable trade-off — or manipulation?
The solution will be unlocked tomorrow.
Solution
It is manipulation, because the training data was deliberately falsified to force a desired outcome.
Trade-offs (false positives vs. false negatives) are normal — but they must be:
• transparent,
• legitimately decided (not secretly by one person),
• controlled (monitoring, appeal mechanisms).
Why is it problematic?
Because one individual quietly defines that “truth” can be sacrificed — without accountability and without safeguards for affected people.
Core question:
Are we allowed to sacrifice real information to reduce harm — and who gets to decide?