Crash Blossoms / IF & ONLY IF is a playful new online work made in collaboration using a kind of artificial intelligence called recursive neural nets (RNN) to synthesise past-present-future headlines by combining the news archive at the British Library with user contributions. It was first shown as part of Leeds Digital Festival on 24 September 2020, and is available at torquetorque.net/crashblossoms. The project’s title Crash Blossoms is taken from the name for instances where ‘headlinese’ produces a weird semantic ambiguity, like ‘McDonald's fries the holy grail for potato farmers’. In this essay, we discuss how the relatively transparent, small data approach used in Crash Blossoms can offer a different trajectory, away from singularity and towards multiplicities of human and machine intelligences.