I think there is another layer emerging:
Invariant Engineering
Graphs describe the topology of an AI workflow: states, transitions, dpnd.
But correctness doesn't come from the graph itself. It comes from the properties that must hold regardless of which path execution takes.
Jerzy Neyman earned his PhD in mathematics from the University of Warsaw, Poland, and then studied under Karl Pearson at London, and later he moved to UoC where he founded and developed one of the world’s leading centres of mathematical statistics.
https://t.co/5kwLUDQ1Hi
George B. Dantzig, widely regarded as the father of linear programming, was a graduate student at the University of California, Berkeley, when an extraordinary incident occurred in a class taught by the distinguished statistician Jerzy Neyman.
Arriving late to a lecture, Dantzig noticed two mathematical problems written on the blackboard. Assuming they were routine homework assignments, he copied them into his notebook and attempted to solve them. Although the problems proved considerably more difficult than usual, he eventually obtained solutions and submitted them to Neyman, apologizing for the delay.
Approximately six weeks later, on a Sunday morning, Neyman appeared unexpectedly at Dantzig's home, carrying a collection of papers. Excitedly, he informed Dantzig that he had prepared an introduction to one of his results and intended to submit it for publication.
Only then did Dantzig discover the remarkable truth: the two problems he had solved as ordinary homework were, in fact, well-known unsolved problems in mathematical statistics.
When Dantzig later struggled to identify a suitable topic for his doctoral dissertation, Neyman suggested that he simply compile the solutions to those two problems into a thesis.