Hidden Reasoning and Neuralese

Will LLMs cease to reason in human-understandable natural language in the future? What are the pathways through which this might happen?