The article discusses the challenges of reliably processing and interlinking medical information, a critical issue for digital transformation in healthcare. Despite advances in medical ontologies, their optimization remains a major bottleneck. Large language models (LLMs) have shown promise in addressing the 'communication problem' in medicine, but their weaknesses, such as hallucination and non-determinism, need to be addressed. Retrieval Augmented Generation (RAG), particularly through knowledge graphs (KGs), offers an automated approach that can deliver structured reasoning and a model of truth alongside LLMs, relevant to information structuring and decision support. The 'semantics problem in medicine' refers to the difficulty of reliably recording and interoperable medical information. This problem affects daily medical information linking and creates challenges in automating medical tasks. Medical ontologies and KGs are interrelated technologies that capture consensus on biomedical concepts. However, their ambiguity and contextual richness pose challenges to adoption. KGs provide structured repositories of knowledge, enabling querying as graph databases and expressing machine-readable semantics for deductive reasoning. LLMs, while powerful, have limitations such as bias, hallucinations, and inaccuracies. Combining LLMs with KGs can complement their strengths and weaknesses. For example, LLMs can be used to construct, enrich, and refine KGs from text, while KGs can be used to augment LLMs by enriching prompts, verifying, or explaining responses. This combination can enhance the ability of physicians to process information and make medical decisions. The article also discusses the potential of RAG approaches, particularly augmenting LLMs with KGs, to better serve medicine, especially in tasks where accuracy and bias control are critical. It highlights the need for regulatory oversight and the importance of balancing the strengths and weaknesses of these technologies. The article concludes that a range of RAG approaches, selected based on specific clinical use cases, will harness the power of LLMs to ultimately solve medicine's 'communication problem'.The article discusses the challenges of reliably processing and interlinking medical information, a critical issue for digital transformation in healthcare. Despite advances in medical ontologies, their optimization remains a major bottleneck. Large language models (LLMs) have shown promise in addressing the 'communication problem' in medicine, but their weaknesses, such as hallucination and non-determinism, need to be addressed. Retrieval Augmented Generation (RAG), particularly through knowledge graphs (KGs), offers an automated approach that can deliver structured reasoning and a model of truth alongside LLMs, relevant to information structuring and decision support. The 'semantics problem in medicine' refers to the difficulty of reliably recording and interoperable medical information. This problem affects daily medical information linking and creates challenges in automating medical tasks. Medical ontologies and KGs are interrelated technologies that capture consensus on biomedical concepts. However, their ambiguity and contextual richness pose challenges to adoption. KGs provide structured repositories of knowledge, enabling querying as graph databases and expressing machine-readable semantics for deductive reasoning. LLMs, while powerful, have limitations such as bias, hallucinations, and inaccuracies. Combining LLMs with KGs can complement their strengths and weaknesses. For example, LLMs can be used to construct, enrich, and refine KGs from text, while KGs can be used to augment LLMs by enriching prompts, verifying, or explaining responses. This combination can enhance the ability of physicians to process information and make medical decisions. The article also discusses the potential of RAG approaches, particularly augmenting LLMs with KGs, to better serve medicine, especially in tasks where accuracy and bias control are critical. It highlights the need for regulatory oversight and the importance of balancing the strengths and weaknesses of these technologies. The article concludes that a range of RAG approaches, selected based on specific clinical use cases, will harness the power of LLMs to ultimately solve medicine's 'communication problem'.