AI-900 Practice Question: Describe Artificial Intelligence workloads and considerations

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Fairness

  Why it's wrong here

  In an AI diagnostic system, fairness specifically addresses algorithmic bias—for example, ensuring that model performance does not vary across patient demographics such as age, sex, or race. While this is an important ethical consideration, it does not measure how often the model's diagnosis is correct or how consistently it behaves under stress, missing data, or novel cases. Fairness metrics (e.g., equalized odds) assess output distributions, not diagnostic accuracy, so a fair system can still produce unreliable or unsafe medical recommendations.

• ✓

  Reliability and Safety

  Why this is correct

  Reliability and Safety is the central principle because it governs whether the AI can be trusted in real clinical workflows: the model must generate accurate, repeatable predictions and fail gracefully when uncertain. In a hospital setting, even rare errors—such as a false-negative on a scan—can lead to delayed treatment or patient harm. This principle mandates rigorous validation on diverse data, calibration of confidence scores, human oversight, and continuous post-deployment monitoring to catch drift or edge-case failures. Consequently, it directly addresses both performance accuracy and avoidance of harm, which are the primary requirements for assisting doctors.

• ✗

  Privacy and Security

  Why it's wrong here

  Privacy and Security focuses on protecting sensitive patient data through measures like encryption, access controls, and audit logs, along with compliance to regulations such as HIPAA or GDPR. These safeguards are essential for maintaining patient trust and preventing data breaches, but they do not evaluate whether a prediction is medically correct. A system could be fully private and secure yet still calculate a diagnosis from flawed logic or biased training labels. Thus, this principle is a foundational legal/technical requirement, not the governing principle for clinical decision accuracy.

• ✗

  Inclusiveness

  Why it's wrong here

  Inclusiveness asks the AI to be usable and effective for all people—including those with different languages, disabilities, or sociocultural backgrounds—so that no patient group is left behind in interface design or accessibility. However, it is a design and equity lens rather than a verification of predictive correctness: a system can be perfectly inclusive in how it is presented and still output incorrect or dangerous diagnoses. For an AI that assists doctors, the immediate priority is ensuring the model's decisions themselves are accurate and safe; inclusiveness is a complementary principle that does not stand in for reliability.

Go deeper

Related to this question