IBM A1000-078 - Assessment: Foundations of AI Advanced Practice Exam: Hard Questions 2025

A retail bank wants to deploy a loan pre-approval model. During pilot, the model shows strong overall AUC, but an audit finds the false negative rate is significantly higher for one protected group. The bank must reduce disparate impact while keeping the model explainable for compliance and minimizing retraining cost. Which approach is the best next step?

A manufacturing company built a computer vision classifier to detect defects. In production, defect rates spike and operators report many false alarms, but periodic sampling shows the physical defect rate is unchanged. Logging reveals the camera firmware update changed color calibration and lighting conditions. What is the most appropriate troubleshooting and mitigation plan?

A healthcare provider is building a triage model where missing a high-risk patient is far more costly than flagging a low-risk patient. The dataset is highly imbalanced (1% positive). The team reports 99% accuracy and wants to ship. Which evaluation and decision strategy is most appropriate before deployment?

A telecom wants a churn model. Most features are engineered from the last 90 days of customer activity. The data science team randomly splits rows into train/test and gets excellent results. After deployment, performance collapses. Which root cause is most likely, and what is the best fix?

A financial services firm must provide reason codes for adverse actions. They currently use a gradient-boosted tree model that performs best but is hard to explain consistently across channels. They need stable, local explanations for individual decisions, plus global understanding for model governance. Which approach best satisfies these requirements with minimal performance loss?

A support organization wants to route incoming customer emails to the correct team. The emails include sensitive personal data. They need entity extraction and intent classification, but they must avoid storing raw text longer than necessary. Which architecture best balances capability and data minimization when using IBM Watson services?

A company builds an FAQ assistant. They use Watson Assistant for dialogue, but users complain the bot answers confidently with outdated policies after a policy update. They want the assistant to ground answers in the latest approved documents and reduce hallucinated or stale responses. Which design is most appropriate?

A team uses Watson Assistant with a handoff to human agents. They notice frequent handoffs even for known intents. Logs show users phrase requests in many ways, and the intent confidence scores are often just below the threshold. What is the best systematic improvement?

A city wants to use AI to optimize traffic signal timing. They have historical sensor data, but conditions change due to events and construction. They need a system that adapts while ensuring safety constraints (e.g., minimum pedestrian crossing times) are never violated. Which approach best fits the problem?

An insurer deploys an AI system to auto-triage claims and recommends denial for suspicious cases. After rollout, regulators question transparency and appeal handling. The insurer must keep efficiency gains but ensure due process, auditability, and human oversight for high-impact decisions. What governance pattern is most appropriate?