In the article "Humans in the Loop" the continuum from manual to automatic is introduced. The author suggests that
AI Systems don't need to be fully automated but that the desiign of AI systems might benefit from an interactive, collaborative approach in which
both the human and the AI contribute with what they are best at.
With a human-in-the-loop approach to creative processes I wonder about bias. When I talk to another human about an idea without showing a visual
our conversation is open and ambiguous leaving room for interpretation and alternative ways of thinking of the same topic. When I present
a clear visual of what I want to do, the other person is biased and will start thinking in the same direction as me due to the common
visual cue we base our conversation on.
Thinking of Dall-e-2 and other generative art algorithms the prompt that we give the system determines the outcome with little room for
interpretation. Every word from the prompt is taken literally whereas human-human communication is often metaphorical and figurative. Dall-e-2 takes the words
from the prompt and creates an image out of what it has learned about these words from annotated images.
An example from Dall-e-2 on the prompt: "Colorful sound drawing using particles" and "A surgical tool to detect tumor margins using sound".
These show that common or more literal prompts achieve better results than non-typical, abstract prompts.
AI systems are powerful tools to create
concrete creative expression from written or spoken words. They make something ambiguous concrete. How can we achieve ambiguous creative results with AI?
How do we create meaningful automation?
Reflecting on Artful Design's Principle 7.11 "That which can be automated should be, that which cannot be meaningfully automated should not be.":
Is meaningful automation a property we want in an AI system? And ultimately, which world do we want to live in?
A world of AI agency? A world of human agency? A world of shared human-AI agency?