I recently came across this interesting article by The Robot Report, relating to training AI robots to use tools. The article highlights recent work carried out by researchers at Princeton University, where it has been found that using "human-language descriptions of tools can accelerate the learning of a simulated robotic arm lifting and using a variety of tools".
The article notes that training AI robots to use tools is challenging due to the wide variety of shapes of tools, combined with a typical robot's dexterity being relatively poor when compared to a human. However, by having the AI robot learn to describe a particular tool (such as it's shape) in natural language, the AI robot learns underlying context about the tool. This underlying context then appears to assist the AI in using the tool when combined with traditional training methods.
I wonder then what the situation is with respect to patenting this sort of technology in Europe. Typically, the European Patent Office do not find inventions relating to natural language processing (NLP) to have the sufficient technical character required to be patentable. Often such applications relating to NLP are rejected for being related to linguistic considerations, rather than technical considerations. For example, a program that predicts the next word a user will write is likely to be based purely on linguistic considerations, and so a patent to this program would struggle to be granted by the EPO. However, using natural language as an aid during training of an AI robot appears technical to me. In particular, while the meaning of the various words could be argued as subjective and contextual, learning to use these words to describe a situation as part of an overall training regime, the result being a robot trained to carry out a specific task (such as manipulate a tool), relates to far more than just the linguistic considerations. Therefore, I would imagine that such a method would have enough technical character to be considered technical in Europe.