Two fields are arriving at the same problem from opposite directions. In galleries, artists like Claudio Perna, Sandy Rodriguez, and Firelei Báez are making work about countermapping: using cartography as a medium for memory, migration, and resistance against colonial spatial logic. Simultaneously, a 2026 paper on arXiv by Hiroki Fukui titled The Digital Afterlife of Empires found that four major language models systematically process writing systems with radical inequality, essentially reproducing imperial cartography inside the model weights. The map and the model are running the same bias.
Who Gets to Draw the Territory
Rodriguez's work, rooted in Aztec and colonial California cartographies, makes explicit what Fukui's paper quantifies: the default map is always someone's conquest. The Hyperallergic shows note that these artists are reclaiming cartography as a political act, insisting that spatial knowledge is never neutral. Fukui's finding that LLMs converge on the same imperial hierarchies of script and language is the computational equivalent: the training corpus is a map, and it was drawn by empires. A 2026 arXiv paper by McKenna and Tshuma, Insidious by Design, sharpens this, arguing that LLM biases remain inadequately theorized, particularly for the Global South, where the distortion is not random noise but structural design.
The Gallery as Correction
What the countermapping artists and the AI bias researchers share is a methodology: making the invisible infrastructure visible. Rodriguez draws the water rights that were erased from official California maps. Fukui draws the attention weights that were never disclosed by model developers. Both are acts of data visualization as political argument, insisting that the territory is never just the territory. The question neither fully answers is who changes the map next.