This is more of a question around general AI.<p>While I understand we have come a long way in recognizing objects with deep learning etc. is there any research around how to ask questions to infer from the patterns learnt automatically ? Not by a human asking the questions to the AI to learn, but by the AI itself questioning and inferring by itself?
There are several different "why"s. You should first list and characterize them. Then you will see how each of them would fit into the algorithm of your AI.<p>But basically, why questions can be considered as (intermediate) goals that need to be generated to make progress toward the ultimate goal of the AI.