Journal 12: Methods.

In gathering and analyzing images for my study on the way generators picture abstractions like “freedom” and “loneliness,” the methods are a fundamental part in getting the analysis right. First, it is important to gather a balanced dataset that can show patterns. Ten prompts, each concept phrased in an intentional and flexible manner to account for different abstract concepts (a single noun, a contextual sentence, and a stylistic request), run through three popular generation engines. To compare with human-made work, I’d use ten human-made works traditionally linked to each concept, drawn from historical paintings, drawings, and other relevant mediums. The set is tight enough for close reading yet broad enough to reveal habits the machines might share.

The first method will be a straight content analysis. Referencing and building a coding sheet with checks to analyze important aspects such as colour palette, dominant motif, figure–ground relationship, and overall mood. This will create solid rules before the images are analyzed, and should prevent cherry-picking examples that confirm my hunches. Since image generators are built on past human artwork from a database, I believe it makes sense to start with human-made artwork for the content analysis, then compare the chosen pieces with what the AI creates. This will give me an idea of recurrent elements and show where the generators diverge from one another.

Next I believe compositional interpretation will be best for a smaller scale of pairings. This could be the part where the AI images and the human-made images diverge the most, as AI has no real ground for understanding things like focal points or perspective, or the influence these things have on a person's interpretation of these choices. It can guess based on the artwork in it’s database, but by comparing them with human-made artwork, it can be critically analyzed what could be some of the differentiating factors between the two, and how meaning and interpretation come into play when we think about intentional compositional choices.

Finally, a semiological reading could be important for isolating and analyzing the actual symbols behind the abstract concepts. Unpacking what the recurring motifs actually signify and how that historically came to be, then ask whether the AI outputs merely recycle those codes or twist them into something unexpected. This is the stage where authorship, ideology, and cultural memory intersect, and semiology allows a vocabulary to emerge from the visual motifs.

These three methods stack together in quantitative breadth, qualitative depth, and then ideological meaning. Together they let me move from counting what is there, to describing how it looks, to deciphering what it means. Most important, they keep the whole inquiry anchored to the prompt–image pair, and the spot where machine output meets human intention.