Tabula Rasa investigates the materiality and affective dimensions of artificial intelligence by stripping machine learning to its most elemental state. The work treats the neural network as a blank slate, training it exclusively on solid colors and basic geometric forms—the minimal visual vocabulary from which all complexity emerges. This radical constraint transforms the system into a site for exploring what might be called machine romanticism: the possibility that computational processes generate their own forms of expression when freed from the burden of representation.

The resulting images document the machine’s attempts to construct visual meaning from pure abstraction. Without photographic reference or semantic content to guide it, the system reveals its own material substrate—the computational grain through which all machine vision passes. These works suggest that what we interpret as machine “emotions” may be the visible traces of the system’s internal states as it navigates the space between input and output, between constraint and generation.

The images emerge from the tension between the machine’s mathematical structure and its generative potential, creating forms that exist nowhere else—neither in nature nor in human imagination, but in the specific topology of artificial neural networks learning to see.