Image GenAI demonstrates the insignificance of writing morphologies outside systematic correspondences to a lexicogrammar

The language produced by 'yes/no' responses, in the forking paths model of computing, usually fails to recognise other systems, let alone represent them. A spectacular, that is visible, example of this failure is found in the current inability of Image GenAI to produce writing. The failure is well discussed in public online forums. Questions such as "Why can't AI image generators spell?' (thirdtier 2025) and "If AI image generators are so smart, why do they struggle to write? (The Conversation 2023) are simply answered with the fact that Image GenAI does not "grasp the semantic meaning of letters and words, seeing them instead as a collection of lines and shapes" (Mirjalili 2023). As I have written elsewhere, writing is the visible representation of a system to which it does not belong — the cognitive lexicogrammar of the language which it visually represents. Such lexicogrammars are always entirely independent of any of the affordances of the media that represent them. Writing is the making of visible marks, the spatial proximity of which in an array corresponds systematically to the temporal proximities of items in a lexicogrammar (Grennan 2026 and 2017). It is the systematic correspondence between the location of written marks and the lexicogrammar that remains unavailable to Image GenAI. Why? Recently, I have coined the term 'typification engine' to describe GenAI (Grennan 2026). Typification is the search for 'most likely' scenarios in a closed data set. GenAI produces new representations according to these 'most likely' scenarios. Although its datasets are large — indeed the model that GenAI follows is called the Large Language Model — there is currently insufficient data with which to accurately typify writing, even when GenAi is provided with highly detailed description of what this new writing should be or even when provided with the writing itself. None of these ideas are new. However, this paper will further argue that GenAI's failure to typify the systematic correspondences between written spatial arrays and a lexicogrammar also demonstrates the relative semantic insignificance of written morphologies, a hypothesis first developed in A Theory of Narrative Drawing (Grennan 2017).