The word 'inference' derives from Latin inferre, meaning 'to bring in, conclude, deduce'

The English noun inference traces to Medieval Latin inferentia, itself from the present participle inferens / inferentem of the classical Latin verb inferre. That verb is a compound of two roots: in- ("in, into," from the Proto-Indo-European root en) and ferre ("to carry, to bear," from PIE root *bher-, the same root that gives English bear and burden). Together, inferre carried the sense of "bring into" and, by extension, "conclude" or "deduce."

The related verb infer entered English by the 1520s, initially in strictly logical contexts; the more general sense of "draw a conclusion" was attested by the same period, with the intransitive sense emerging in the 1570s. The noun inference appears somewhat later: the sense "action of inferring" is recorded from the 1590s, and the sense "that which is inferred" from the 1610s.

The earliest English evidence cited by the Online Etymology Dictionary dates to the 1590s, with the theologian and philosopher Richard Hooker among the first to use it in print.

Why this etymology matters for AI usage

The root metaphor is physical: ferre, to carry. Inference is literally something brought in — a conclusion transported to us by the carrying-power of prior premises. This spatial metaphor has persisted across every domain that inherits the term: from Aristotelian syllogistics through statistical inference to the contemporary AI engineering sense, where "inference" means the forward pass that carries input tokens through a model to produce output. The word has travelled far from its origins while retaining its core shape.

See also: claim-inference-logical-types, claim-ai-inference-means-running-a-model, claim-statistical-inference-meaning