Lessons in Feedback from TURN: Washington’s Spies
Signals, codes, dead drops and activation functions
AMC’s Turn has come to an end. The revolution is over or perhaps it never ends. Along the way Woodhull, Tallmadge, and Strong introduced us to a world of covert messages, signals, code words, and intrigue. The final two episodes which aired this month emphasized this more than any.
Turn is inspired by the history of the revolutionary war in america and the Culper Ring that operated in New York City and Long Island. I will stick with the word ‘inspired’ for two reasons; first it clearly took huge liberties (especially with Robert Rogers) and secondly my writings on history tend to go here:
Turn offers us a great pop-culture analogy for feedback, analytics, and even a chance to further dig under the covers (layers) of deep learning. The revolutionary war period of spy games, though introductory for the United States was far from it for much of the world(more here). But small budgets and limited technology provide us with a very simple model to study.
The story emphasizes that Washington had no knowledge of who his agents were. He knew them by pseudonyms (Culper) and numbers. He provided direction to his Head of Intelligence (Tallmadge) who in Turn :) relied upon a small series of handlers (Brewster), agents (Woodhull), and signalers (Strong and later Townsend). Translating signals (the job of activation functions in deep learning) is a major piece of the plot.
I write at great length about feedback. I believe it is the cornerstone to information engineering, adaptive systems, and by default real analytic success. If your goal is to learn (or spy), there is nothing more critical.
So Turn offers a us a view into a critical and historic model. It is a touch low tech perhaps, with dead drops and signal flags. But the struggles of communication and understanding make a compelling component of the plot. It is further colored by the anonymous nature of the ring. Everyone from Washington to the British Intelligence service remained unaware of the number and identity of many of the players.
It is a lesson in layers. And in the difficulty of passing information through so many layers and ultimately back again. In this case, complexity that was added to obscure. Not an exact analogy for deep learning structures but still very close. And while deep learning utilizes activation functions to transform feedback and signals, the revolutionaries of Turn used code books.
In a final lesson, the Culper Ring actually intervenes in the printing of code books for the translation of signals by the British fleet. With their ‘activation functions’ sabotaged, the British are easily routed by the French fleet at Yorktown. I know, it is not an exact analogy or even an historic certainty, but it still stands as an entertaining example of what happens when feedback loops are corrupted. Again, it a story line that is oft repeated over the four season run of the show.
Turn: Washington’s Spies came to a close on August 12th. It was a fun and entertaining take on a less well known piece of American history. While only a historically ‘inspired’ tale, it was one that provided plenty of interesting lessons in feedback and analytics. Lessons that history shows can be truly meaningful in competition (war), intelligence, and communication (especially the indirect kind). Thanks for reading!
