It’s only smoke and mirrors until you get closer. Then the graphics engine kicks in and depending on how high you set specific object LOD, it needs to render the correct asset, just as it does today for autogen buildings. In much the same way building classes are defined by size, color, texture and regional uniqueness, you need to have more than just a small number of generic vessels. You expect to see sailboats and powerboats in a civilian marina. You expect to see trawlers at a mostly fishing/working port. You expect to see freighters, bulk carriers, oilers at a large commercial port. They may not represent real ships, or correct colors etc. These are some of the additional considerations a Machine Learning process has to encompass when solving for this gap.
That is an 80/20 solution, which conforms to Agile Dev Model as how the Asobo team has been delivering product features. Dropping a small number and variation of generic ships everywhere doesn’t meet that threshold.