IBM A1000-050 - Assessment: Foundations of AI Advanced Practice Exam: Hard Questions 2025

A financial services company is implementing an AI system to detect fraudulent transactions. During model evaluation, they observe that their model achieves 98% accuracy but only identifies 45% of actual fraud cases, while the business requires detecting at least 85% of fraud cases even if it means more false alarms. The dataset contains 10,000 transactions with only 100 fraudulent cases. What is the PRIMARY issue and the most appropriate solution?

An enterprise is designing an AI governance framework for multiple AI systems across different departments. One system uses computer vision for quality control in manufacturing, another uses NLP for customer service automation, and a third uses predictive analytics for supply chain optimization. The governance team must establish explainability requirements. Which approach BEST balances regulatory compliance, operational needs, and technical feasibility?

A healthcare AI system trained on historical patient data from 2015-2020 shows declining performance in production during 2023. Investigation reveals that treatment protocols changed significantly in 2021, and the patient demographic has shifted due to new insurance partnerships. Model retraining on recent data improves performance temporarily, but the data science team wants a sustainable solution. What architectural approach would MOST effectively address this challenge?

An AI research team is evaluating different neural network architectures for a multi-modal application that processes both text reviews and product images to predict customer satisfaction scores. They need to understand how information flows through the network. Which statement MOST accurately describes the gradient flow and optimization challenges they should anticipate?

A manufacturing company wants to implement predictive maintenance using machine learning to predict equipment failures 48 hours in advance. They have 5 years of sensor data (temperature, vibration, pressure) with failure labels. During model development, they achieve 92% accuracy, 88% precision, and 76% recall on the test set. However, in production, actual failures occur at different times than predicted, and many predictions are false alarms. What is the MOST likely root cause and appropriate remediation?

An AI system for automated resume screening is being audited after complaints of discriminatory hiring practices. The model was trained on 10 years of historical hiring data and doesn't use protected attributes (gender, race, age) as features. However, statistical analysis reveals disparate impact across demographic groups. What is the MOST comprehensive explanation of this situation and the appropriate remediation strategy?

A global retail company is deploying a demand forecasting AI system across 50 countries. The model performs well in North America and Europe but shows significantly degraded performance in Asian and African markets. Data volumes are similar across regions, and the same features (historical sales, seasonality, promotions) are available everywhere. What is the MOST likely explanation and solution?

An AI system using reinforcement learning is being trained to optimize energy consumption in a data center by controlling cooling systems. During training, the agent learns to achieve excellent energy efficiency scores, but when deployed, it occasionally makes decisions that risk equipment overheating. The reward function was designed to minimize energy costs while maintaining temperatures within acceptable ranges. What aspect of the RL system design MOST likely caused this issue?

A pharmaceutical company is evaluating whether to use AI for drug molecule generation versus traditional computational chemistry approaches. They need to understand the fundamental capabilities and limitations. Which statement MOST accurately characterizes the trade-offs between generative AI approaches and traditional methods for this application?

An enterprise is implementing a federated learning system where multiple hospitals collaboratively train a diagnostic AI model without sharing patient data. During implementation, they observe that model convergence is slow, and the final model performs well for some hospitals but poorly for others. What combination of factors MOST likely explains this challenge and what mitigation strategies should be prioritized?