The article discusses the challenges and potential solutions for processing and interlinking medical information, particularly focusing on the "semantics problem" or the "communication problem" in medicine. This problem involves the difficulty of reliably recording and making medical information interoperable between systems. While large language models (LLMs) have shown promise in addressing these challenges, they exhibit limitations such as hallucinations and non-determinism. To mitigate these issues, the article proposes the use of Retrieval Augmented Generation (RAG) techniques, particularly through knowledge graphs (KGs), which can provide structured reasoning and a model of truth alongside LLMs. The combination of LLMs and KGs is seen as a complementary approach, where LLMs can be used to construct, enrich, and refine KGs from text queries, while KGs can enhance LLMs by enriching prompts, verifying responses, and providing context. This hybrid approach aims to create a robust system for medical information processing and communication, potentially enabling the creation of digital twins of individual patients with up-to-date health data. The article also highlights the importance of quality control and regulatory considerations in the use of LLMs for medical tasks, emphasizing the need for interactive back-and-forward complementarity to address the limitations of LLMs. Despite ongoing challenges, the authors believe that RAG approaches, especially those combining LLMs with KGs, show promise in solving medicine's "communication problem" and improving interoperability in healthcare.The article discusses the challenges and potential solutions for processing and interlinking medical information, particularly focusing on the "semantics problem" or the "communication problem" in medicine. This problem involves the difficulty of reliably recording and making medical information interoperable between systems. While large language models (LLMs) have shown promise in addressing these challenges, they exhibit limitations such as hallucinations and non-determinism. To mitigate these issues, the article proposes the use of Retrieval Augmented Generation (RAG) techniques, particularly through knowledge graphs (KGs), which can provide structured reasoning and a model of truth alongside LLMs. The combination of LLMs and KGs is seen as a complementary approach, where LLMs can be used to construct, enrich, and refine KGs from text queries, while KGs can enhance LLMs by enriching prompts, verifying responses, and providing context. This hybrid approach aims to create a robust system for medical information processing and communication, potentially enabling the creation of digital twins of individual patients with up-to-date health data. The article also highlights the importance of quality control and regulatory considerations in the use of LLMs for medical tasks, emphasizing the need for interactive back-and-forward complementarity to address the limitations of LLMs. Despite ongoing challenges, the authors believe that RAG approaches, especially those combining LLMs with KGs, show promise in solving medicine's "communication problem" and improving interoperability in healthcare.