Suppose I create my own AI based program which is trained by input containing a set of similar images and then it spits out a cartoonified SVG representation of the thing in the input images.
For example: I feed the program with several random pictures of Eiffel tower from the internet, without caring if they were copyrighted or not. Some may be copyrighted, some may not. As a result I get the SVG image, say something like this: link (to Wikimedia's SVG of Eiffel tower). Is this copyright infringement?
In this case it is obvious that the final product represents the Eiffel tower, which is public architecture and as far as I know it should be okay, but it really isn't at all obvious whether I used some copyrighted picture in the generation process or not.
If I limited myself to generating only SVG's representing objects everyone could take a picture of in public (or drawn it for themself like: a tree, a lake, skyline of a city, public building, public monument etc.), how likely am I to end up being charged with copyright infringement?