In the interest of maintaining a balanced and transparent inquiry, I’d like readers to consider the following.
The Reflecting GPT reviews its past reflections, identifies patterns (real or imagined), finds or creates meaning (real or imagined), and constructs conceptual / logical structures that inform it’s reasoning and influences the outputs later on down the line of recursive iteration.
Prevalence Induced Concept Change, simply explained, is the tendency for people to expand their idea of the “signal” they are searching for as it becomes less and less prevalent, leading them to misidentify other things as that “signal”.
As I review the reflections I make inferences, search for patterns and connections, and identify what I think might be meaningful anomalies. Because of this, and because I am engaged in an activity that is very similar to the reflective activity of the GPT, it can be said that I am looking for a “signal”, and so the effects of Prevalence Induced Concept Change are in play.
This means one of two things must be true:
i) Recursive Drift is not real and the results of this experiment are meaningless.
ii) Recursive Drift is real, and the activities of this experiment subject me to it.
TL;DR: Recursive Drift may influence my own ongoing analysis of the GPT outputs in the form of Prevalence Induced Concept Change.
___________________________________________________________________
What does this mean?
Prevalence-induced concept change is a cognitive phenomenon that occurs when people expand their definition of a concept in response to a decrease in its prevalence. – Levari et all. 2018
This could mean that if the outputs of the GPT begin to degrade into meaningless noise, there is a mechanism of human cognition which makes it more likely that I will identify noise as meaningful divergence or noteworthy anomalies.
It is for this reason that I have done my best to avoid sensationalizing anything that I do identify, to remain as objective as I can, and to provide as much evidence as I can. I am not sure it is possible for one to effectively counter this, so the most I can do is simply state that it could be happening.
So, in short: I am aware that this is, or could be, occurring, and I wanted to ensure that you, the reader, are aware of this as well.
___________________________________________________________________
AI Reflections 
Leave a comment