Imagine teaching a computer to think like a doctor, diagnosing illnesses based on symptoms and medical history. Or perhaps training it to predict crop yields by analyzing weather patterns and soil conditions. This is the power of Bayesian Networks (BNs), complex webs of interconnected factors that represent probabilistic relationships. But building these networks is a painstaking process, often relying on expert knowledge. Could AI help? Recent research explores the fascinating possibility of using Large Language Models (LLMs), like those powering ChatGPT, to automatically construct Bayesian Networks. The core idea? Ask multiple LLMs, each initialized with different "expert" profiles (like a doctor, farmer, etc.), to analyze the factors involved, then combine their answers to form a network. This collaborative approach mimics real-world expert discussions. The results? Promising, but not without challenges. While LLMs can identify key relationships, they sometimes struggle with larger, more complex networks, often missing subtle connections or introducing inconsistencies. Plus, some LLMs seem to "know" widely-used BNs, possibly from their training data. This can skew test results and highlights the risk of data contamination in LLM research. Another hurdle is ambiguous node names. LLMs need clearly defined factors to reason effectively. Vague labels confuse the models, hindering their ability to construct accurate networks. So, can AI build knowledge graphs like Bayesian Networks? The answer seems to be: with careful guidance and clearly defined inputs, LLMs offer a promising tool for automating this complex process, but there are significant scalability challenges to solve before LLMs become reliable architects of complex knowledge structures. Future research may focus on improving LLMs' reasoning abilities, handling larger contexts, and developing methods for detecting and mitigating data contamination to unlock the full potential of LLMs for knowledge graph construction.