AI not to apply when Semantic Fidelity and Data Sufficiency are missing?

Question Is it correct that AI is not fit for use in the situation where Semantic Fidelity and Data Sufficiency are not expected. Here AI means a system using the decoder based LLM. AI or agent is now buzz but I am afraid there is a misbelief that AI can handle input that has ambiguities. I believe ambiguities or inconsistencies are what human must resolve so that AI will not speculate or infer. Or should I let AI to identify them not to handle. But I am not sure how capable current LLMs are to identify the ambiguities or inconsistencies. If there are recent researches, kindly suggest. Situation Working on a task to build an AI system that generates programmatic code from formal requirement descriptions written by human. So it is a code agent. We expect the AI system to generate consistently correct and verifiable results. But there are problems. Problems The input requirement descriptions may have semantic fidelity problem leading to multiple interpretations. Hence, AI may infer an intention of the requirement author that the input does not establish on its own. The resulting output may therefore misrepresent the original intention. The requirements may have data sufficiency problem referring to data that requires implicit knowledge. Semantic Fidelity Problem Semantic fidelity means: Original intentions are faithfully and accurately represented. In the context of AI, this requires: Input Fidelity: The input expresses the original intention accurately and unambiguously. Outp…