It's a pretty interesting example of the pitfalls of ML. It's a clever and exciting use of ML and when observed in a silo is the correct thing for any company to do. The law wasn't designed for such sophisticated modelling of the domain. Screwing over small and medium businesses was never the goal, it's just an inevitable consequence. The flip side is the algorithm cannot model what happens once they screw over businesses so badly that it causes a mass migration away from Amazon (because they've not seen that data before). Fascinating to see how this plays out.