Learning how to do a backflip can be a difficult, risky endeavour. That’s why, after many failed attempts and a stubbed toe, Nikita Diakur, a filmmaker based in Germany, opted for an unconventional method – machine learning. To achieve the feat, Diakur created a digital avatar of himself and, with code and inspiration pulled from a research paper titled ‘DeepMimic’, programmed it to train on YouTube tutorial videos until it could land a backflip successfully. The resulting short is a wildly original and amusing snapshot of both the potential and current limits of AI. And underneath its wry slapstick humour, Backflip is a thoughtful exploration of fear – not just of the unpredictable future of machine learning, but also the physical world we inhabit, where the results of jumping from your feet and landing on your head can’t be shaken off quite so easily.
Struggling to learn how to do a backflip, Nikita takes on an unusual training regimen
Director: Nikita Diakur
Producers: Emmanuel-Alain Raynal, Pierre Baussaron

videoComputing and artificial intelligence
Who, exactly, authored this AI-generated spin on Alfred Hitchcock’s Vertigo?
5 minutes

videoComputing and artificial intelligence
Machine learning is important, but some AIs just want to have fun
57 minutes

videoTechnology and the self
Why we should worry less about ‘sentient’ AIs and more about what we’re teaching them
16 minutes

videoComputing and artificial intelligence
Teaching an AI to beat video games still takes human imagination
5 minutes

videoFilm and visual culture
With human help, AIs are generating a new aesthetics. The results are trippy
9 minutes

videoDance and theatre
A ballerina dances with the geometry of her own movements
3 minutes

videoEngineering
Could this development in biomimetic robotics change terrestrial transport?
3 minutes

videoFuture of technology
Artificial ‘creativity’ is unstoppable. Grappling with its ethics is up to us
23 minutes

videoMood and emotion
Will they or won’t they? Prospective jumpers anguish at the edge of the high dive
16 minutes