Unveiling the Shortcuts: How Retrieval Augmented Generation (RAG) Influences Language Model Behavior and Memory Utilization