Varying interpretations of prompt actors

As noted in my blog “Making vignettes”, I deliberately chose rather abstract and poorly defined actors for my prompts, with the intent of letting the AIs interpret them in a relatively unimpeded way. The strategy has resulted in sometimes wildly varying interpretations of the same actor when used in different prompts—and also within the 4 variations generated from the same prompt. This to me indicates a certain amount of what I like to construe as “creativity”, for lack of a better word. There are of course raging debates about whether an AI can be truly creative, depending largely on what we consider creativity to be. For my purposes, I consider creativity to be the ability to interpret prompts in a way that produces a coherent and meaningful output, even if that output pushes the boundaries of human-readable signifiers. And a working gauge for such AI creativity (let’s drop the quote marks now), can be the ability to variously interpret the actors, situations and activities in prompts.

It is my main thesis that such variability, driven by non-human associative processes, can make these AIs valuable collaborators for humans engaged in creative activities. Below I will show some examples of how various actors are interpreted, when they are inserted in minimal prompts that lack style or formatting instructions.

“Enigma” is by definition open to subjectivity, so I used that for several of my vignettes. Here are some example interpretations:

Likewise “figment”:

And “chimera”:

These examples show how a wide range of styles can be available with an unstructured approach to prompting. My ebook “AI Vignettes: Chimeras Wraiths and More” comprises 100 such generations, each paired with a story from GPT-3.