IBM A1000-076 - Assessment: Foundations of AI Advanced Practice Exam: Hard Questions 2025

A financial services company is building an AI system to detect fraudulent transactions. During testing, they discover that while the model achieves 98% accuracy, it only identifies 45% of actual fraud cases (which represent 0.5% of all transactions). The business requires detecting at least 85% of fraud cases, even if it means more false positives. Which combination of strategies should the data scientist prioritize to address this issue?

An enterprise is implementing a conversational AI solution using IBM Watson Assistant for customer support across multiple languages. They observe that while English queries achieve 92% intent recognition accuracy, queries in Japanese and Arabic show only 68% accuracy despite having equivalent training data volumes. The dialogue flows are identical across languages. What is the most likely root cause and appropriate solution?

A healthcare AI system uses deep learning to analyze medical images and recommend treatment options. During an ethical audit, concerns are raised about the model's decision-making process being opaque to physicians. The model is a complex ensemble of convolutional neural networks with over 100 million parameters. Which approach best balances model performance with ethical transparency requirements in this high-stakes medical context?

A data scientist is training a transformer-based language model and observes that training loss continues to decrease while validation loss plateaus after epoch 15, then begins increasing. Training accuracy reaches 99% while validation accuracy peaks at 82% at epoch 15 then declines to 78%. The learning rate is 0.001 with no scheduler. Which combination of techniques would most effectively address this scenario?

An organization is designing an AI governance framework for deploying multiple AI systems across different business units. They need to address model drift, fairness monitoring, and regulatory compliance. Some models are third-party black boxes, others are internally developed. Which architectural approach best implements comprehensive AI governance?

A company is implementing IBM Watson Discovery to extract insights from 500,000 internal technical documents in PDF format. Initial results show poor extraction quality with missing tables, garbled technical symbols, and low relevance in search results. Documents contain complex layouts, equations, diagrams, and multilingual technical terminology. What is the optimal strategy to improve extraction and search quality?

A research team is selecting between different neural network architectures for three distinct projects: (1) real-time video analysis for autonomous vehicles, (2) language translation for legal documents, and (3) predicting customer churn from tabular data with 50 features. Which architecture selections are most appropriate for these respective use cases?

An AI system trained on historical hiring data is being audited for fairness. The model shows 85% accuracy overall, but analysis reveals disparate impact: it recommends interviews for 60% of Group A candidates but only 35% of equally qualified Group B candidates. Legal counsel indicates the 4/5ths rule is violated. The business requires maintaining prediction quality while ensuring fairness. Which intervention strategy is most appropriate?

A team is building a hybrid AI system combining symbolic AI and machine learning for medical diagnosis. The symbolic component contains expert-encoded rules about drug interactions and contraindications. The ML component predicts diagnoses from patient symptoms and lab results. During integration testing, the system sometimes recommends treatments that the symbolic rules explicitly contradict. What integration architecture best resolves this conflict while leveraging both components' strengths?

An enterprise is implementing a Watson Natural Language Understanding (NLU) pipeline to analyze customer feedback across email, chat transcripts, and social media in English, Spanish, and German. They need to extract sentiment, emotions, entities, and custom categories (product features). Initial results show inconsistent entity extraction across channels and languages. The system processes 100,000 documents daily. What optimization strategy best improves consistency and performance?