Ground-breaking at the time of its advent in 1965, closed-circuit video turned the space between the camera and the monitor into a cognitive battlefield for artists and viewers. Whether in the studio or the gallery, such analogue systems were installed to challenge our perception of the self and others notably with the addition of mirrors and time delays throughout the 1970s. By placing sensitive bodies at the core of architectural feedback loops, many video artists thus put notions of power, control and knowledge at the forefront of their perceptual and psychological investigations (Rosalind Kraus, 1976), overall playing with the desire of people to access the machine’s field of vision so that to confront or reconcile their own image played back in real-time. While the blinding developments of artificial intelligence over the last decade have greatly marginalized human perception by arguably reversing our prosthetic relationship to machines (Mark Hansen, 2014), these notions continue to inform media arts today, notably through the use of generative adversarial networks (GANs). As part of an ongoing research that aims to bridge pioneer video art to current machine-learning art through the prism of the medium’s original agency (Ina Blom, 2016), this paper proposes to compare the anticipatory capacities induced by both feedback and feed-forward systems with further examples taken from recent history. Neural networks, like video before them, are anticipation technologies, which open us onto a new epistemic relationship to the world, and whose processes inscribe within the images a prospective history that blurs the border between the future and the past, all the while ‘showing the impossibility of locating a pure present tense’ (Dan Graham, 1990).