AIF-C01 Practice Question 88: Robustness Adversarial Inputs
AWS Certified AI Practitioner (AIF-C01) · Test 9 · question 8 of 10
A company deploys a model that classifies customer messages as complaints or compliments. During testing, the team finds that adding typos, unusual phrasing or deliberately crafted wording to a message often flips the prediction, even though a human would read the message the same way. Which core dimension of responsible AI is this weakness most directly about?
Show answer and explanation
Answer:
AWS defines veracity and robustness as *"achieving correct system outputs, even with unexpected or adversarial inputs."* A model whose predictions flip because of typos or adversarially crafted wording fails that test, so the team should evaluate and harden it against such inputs (option D).
- A is wrong because fairness is about consistent outcomes across different groups of people. The problem here is sensitivity to wording, not to who wrote the message. - B is wrong because transparency is about communicating information about the system, such as its capabilities and limits, to stakeholders. It does not address whether outputs stay correct under unusual inputs. - C is wrong because privacy and security is about appropriately obtaining, using and protecting data and models. No data exposure is described.
Reference: Responsible AI