A logistics firm replaces a deterministic routing script with a machine-learning model that re-plans deliveries from live traffic data. The governance lead notes two reasons the old build-once-and-sign-off control no longer suffices: the same input can yield differently worded route plans on separate runs, and engineers cannot trace which features drove any single plan even though they hold the full model. Which TWO characteristics of AI does this pairing identify as the reasons a dedicated governance response is needed? Select TWO.
- AProbabilistic output, because the model can return varying plans for identical inputs rather than one fixed result, so acceptance testing cannot rely on a single expected answer. Correct
- BOpacity, because the firm holds the full model yet still cannot trace which input features drove an individual route plan, so the reasoning resists direct inspection. Correct
- CAutonomy, because the model decides routes without a human approving each one, so a person must retain authority to halt or override its actions.
- DSpeed and scale, because the model can re-plan thousands of deliveries far faster than staff could, so a small error propagates widely before anyone notices.
- EComplexity, because retraining one component shifts the behaviour of others in ways no one predicted, so the pipeline cannot be validated part by part.
Why A is correct: Correct: varying outputs for identical inputs is the defining mark of probabilistic rather than deterministic behaviour, which is the first reason cited.
Why B is correct: Correct: inability to trace the drivers of a single decision despite full access is opacity, the second reason the lead names.
Why C is wrong: Tempting since re-planning feels self-directed, but the stem names varying output and untraceable reasoning, not action taken without human approval.
Why D is wrong: Plausible for a fleet-wide tool, but neither reason in the stem concerns throughput or how far an error spreads, so it is not identified here.
Why E is wrong: Plausible as an AI trait, but the stem describes one model, not emergent interaction between components, so complexity is not what this pairing names.