The Recognition Problem is a creative writing experiment for the AI age. Every word in the story was written by AI, as was the plot of each story, the characters and their interactions with each other, and the literary devices and narrative style in which each story has been presented. But perhaps most significantly, AI was also used to devise the narrative plot-line that hangs invisibly across all ten stories, presenting each in a near-future world that is an incrementally advanced over the one that only comes together in the final story.
Today, an AI conversation can, at best, hold twenty to thirty thousand words comfortably in context while it works. Beyond that, earlier material starts to fall out of memory, and the output it generates after that starts to become increasingly unmoored from concepts contained in the earlier part of the session. This makes it next to impossible to use AI to write full-length books.
The Recognition Problem is an experiment to see if it is possible to overcome that.
I worked under a few strict self-imposed constraints:
- the entire book would be generated by AI itself with no direct intervention from me in the actual text of the output — no edits to the way in which individual sentences are phrased or the actual words used.
- all I would do is provide editorial inputs (in much the same way as a human editor working closely with a human author would provide suggestions and comments without actually interfering with how the author writes).
- I would also help the AI extend its context by opening new conversations whenever the context was running low and carrying over summarised context into the new session so that the AI always knows enough about what has already been written to remain coherent over the entire project.
This called for the creation of an elaborate scaffolding that ensured that each story was consistent with the arc of future history being built.