CONCEPT
Raw Data Is an Oxymoron
Lisa Gitelman's signature insight that data is never found but always made—shaped by the instruments that collect it, the institutions that commission it, the categories that organize it, and the assumptions embedded in what counts as data at all—a claim that applies with full force to the training corpora of every AI system ever built.
Lisa Gitelman's edited 2013 volume
“Raw Data” Is an Oxymoron demonstrated, through case studies ranging from census data to astronomical observations to internet archives, that the concept of raw data as unmediated reality was always a myth. The word
data derives from the Latin
dare—to give—and its contemporary meaning as “things found in the world” is the endpoint of a centuries-long historical process that naturalizes what was always a construction: data is taken from the world, not given by it, and the taking is always shaped by instruments, institutions, categories, and interests that are invisible precisely because they have been naturalized. AI training corpora are the most consequential instantiation of this principle in the history of knowledge production: a collection assembled from what was digitized, in what languages, by which institutions, under which economic conditions and copyright