The genre of “found poetry” encompasses passages of text that were first framed as poetry by someone other than the original writer. But for something to be found, it must first be lost. In an attempt to shed light on the role of computational intelligence in the human creative ecosystem, I argue that many computational generators of poetry function essentially as poetry losers: machines whose central purpose is to arrange units of language, without fully understanding them, in combinations that can later be found to be poetry. This implies a paradox for computational poetry: a poetry machine that too completely understands the poetic effects of its output deprives human readers of the chance to find poetry where the machine did not, fundamentally altering both the reader's poetic experience and the machine's utility. I briefly explore the implications of this view, taking computational poetry as a microcosm of “intelligent” machines in creative contexts generally, and discuss what it means to construct an effective poetry loser.