Abstract
Much of the epistemic legitimacy attributed to machine learning comes from a
familiar narrative about the relationship between the world, data, models, and de-
cisions. According to this narrative, data objectively represent the world, models
reveal its underlying patterns,and their outputs provide reliable evidence for action.
This thesis argues that this story functions as an operative myth: not becausei ts in-
dividual steps are fictitious, but because it presents epistemic legitimacy as passing
automatically from one stage to another.
The thesis follows this narrative from world to data, from data to models, and from
model outputs back into social practice. It shows how data are produced through
situated practices of classification, operationalization, and quantification; how pre-
dictive performance establishes fit to a particular representation without by itself
establishing validity with respect to a target in the world; and how scores,rankings,
and categories introduce further judgments when they are translated into interven-
tions. Since such interventions can also reshape the conditions under which future
data are made, deployment does not end the epistemic process.
An alternative account in which epistemic legitimacy is earned through warranted
relations rather than inherited across a machine learning pipeline. Representations
are assessed for their adequacy to particular purposes andi nferences. Objectivityi s
understood as situated and referenced rather than perspective-free. And even well-
supported predictions remain distinct from the normative justification of actions.
On this account, machine learning can produce useful knowledge, but its epistemic
force depends on relations between representations,targets,claims,and actions that
must remain visible,contestable, and open to revision.