Look Closely is first and foremost an experiment in human perception, angled toward discerning the public’s level of familiarity with generative media. Participants are shown a series of images and tasked with deciding whether each one was an AI-generated image, or a human-sourced photograph.

Throughout the experiment, I became less interested in whether or not people could differentiate between the two, but when it mattered if they couldn’t. The following questions are ones I found myself routinely confounded by throughout the experience.

Is there harm in generatively visualizing something that is true?

Is AI for photorealism a misnomer?

What is the threshold at which you need to declare an image as AI generated?

How do we keep track of the visual media representing the ground-truth?