Seems cool, but small nit: many times scientists publish basic research or foundational science, which becomes useful overtime via a process of different teams looking for some piece of knowledge out there to build something. In AI/ML, it's a little different because usually what's published can be implemented and used in some way, i.e., the approach is testable and shows improvements over other methods. I assume that the AI/ML or software methods mentioned in the papers coming out via these AI papers have been tested. For instance, AI/ML researchers usually research code along with their papers (for gaining credibility and contributing to the field).