When assessing the success of an AI system in meeting its objectives, which of the following approaches best aligns with the requirement to ensure a comprehensive evaluation while avoiding automation

Question

Options

• AEvaluate and align pre-defined benchmarks which will provide evidence of having achieved your
• BRely on the AI system's output to determine if the goals were achieved as this will provide a
• CFocus on user feedback about the AI system's performance to directly measure the system's
• DConduct a thorough review that includes the AI system's input and output data, human

• A
  12% (6)
• B
  6% (3)
• C
  26% (13)
• D
  56% (28)

Explanation

Option D is correct because a thorough review combining both input/output data analysis and human judgment guards against automation bias - the tendency to uncritically accept AI outputs as correct simply because a machine produced them. Relying on diverse evaluation layers (data audit + human oversight) ensures no single mechanism dominates the assessment.

Why the distractors fail:

• A (pre-defined benchmarks only) is incomplete - benchmarks measure narrow, pre-specified behaviors and can be "gamed" or miss real-world failure modes outside their scope.
• B (rely on the AI's own output) is the textbook definition of automation bias - you're letting the system grade itself, which cannot surface systematic errors or misaligned objectives.
• C (user feedback only) captures perception but not ground truth; users may not notice subtle failures, and subjective satisfaction doesn't confirm the system actually met its objectives.

Memory tip: Think of the acronym HALO - Human + Automated + Layers + Oversight. Whenever a question mentions avoiding automation bias, the right answer will always include a human review component alongside data analysis, never automation alone.

No community discussion yet for this question.