IBM A1000-077 - Assessment: Foundations of AI Advanced Practice Exam: Hard Questions 2025

A financial services company is developing an AI system to detect fraudulent transactions. During model validation, they discover that their model achieves 98% accuracy but only 45% recall on actual fraud cases, while precision is 92%. The dataset contains 10,000 transactions with only 150 fraudulent cases. What is the PRIMARY issue affecting model performance and what should be the recommended solution?

An enterprise is implementing a multi-model AI architecture where Watson Assistant handles conversational interfaces, Watson Discovery processes unstructured documents, and Watson Natural Language Understanding extracts entities. The solution experiences latency issues when processing user queries that require sequential calls to all three services. Which architectural pattern would BEST optimize performance while maintaining functionality?

A healthcare AI system trained on data from urban hospitals is being deployed to rural clinics. During testing, the model shows significant performance degradation. Analysis reveals different patient demographics, disease prevalence patterns, and diagnostic equipment capabilities. From an AI ethics and governance perspective, what is the MOST critical concern and appropriate mitigation strategy?

A deep learning team is debugging a convolutional neural network (CNN) for medical image classification. The training loss decreases steadily to 0.05, but validation loss plateaus at 0.45 after epoch 15 and then increases. Training accuracy reaches 97% while validation accuracy peaks at 72% and decreases. The learning rate is 0.001, batch size is 32, and the model has 15 layers with 8 million parameters trained on 5,000 images. What combination of techniques would MOST effectively address this issue?

An organization is designing an AI governance framework for multiple AI systems across different business units. They need to balance innovation velocity with risk management. The systems range from low-risk recommendation engines to high-risk credit decisioning and medical diagnosis tools. Which governance approach would BEST align with AI ethics principles while maintaining operational efficiency?

A company is implementing Watson Discovery to analyze 500,000 technical documents. After initial deployment, users report that search results often miss relevant documents and return many irrelevant ones. The enrichment process includes entity extraction and sentiment analysis. Query logs show users employ domain-specific terminology. What is the MOST effective strategy to improve relevance?

A data science team is selecting an appropriate AI model for a business problem. They need to predict customer churn (binary classification) with 50 features, have 100,000 training samples, and require model interpretability for regulatory compliance. The business needs to understand which factors drive churn predictions. Which modeling approach provides the BEST balance of performance and interpretability requirements?

An AI system using transformer-based natural language processing is deployed in a customer service application. The model was trained on English text from 2019-2020. In 2024, the system shows declining performance with increased customer complaints about irrelevant responses. Analysis reveals new slang terms, product names, and industry terminology. What is the PRIMARY cause and most appropriate long-term solution?

A multinational corporation is deploying an AI-powered hiring tool to screen resumes. During bias testing, they discover that the model shows lower acceptance rates for candidates from certain demographic groups, even when controlling for qualifications. The training data consisted of successful hires from the past 10 years. From an AI ethics perspective, what is the MOST critical root cause and what approach should be taken?

A Watson Assistant implementation for a complex insurance domain requires handling multi-turn conversations with context spanning 10+ exchanges, integrating with 6 backend systems, and managing 300+ intents across various insurance products. Users report that the assistant loses context mid-conversation and provides inconsistent responses. What architectural and design approach would BEST address these challenges?