Sackler Centre for Consciousness Science

Time perception without clocks

Our new paper, led by Warrick Roseboom, is out now (open access) in Nature Communications. It explores how AI could perceive the passage of time in the same way that humans do – not as a constant ticking clock but in response to what we see in the world around us.

Dr Roseboom, with colleagues from the University of Sussex and Imperial College London, devised a series of experiments in which an artificial model of a neural system attempted to estimate the duration of short videos of natural scenes filmed from a first-person perspective, such as walking around a busy city, the quiet countryside or sitting in an office or café. 

The study found that the model made the same kinds of mistakes as humans when estimating the duration of videos by overestimating short durations and underestimating longer clips as well as estimating videos taken of busy city scenes to be longer than those of quiet countryside scenes. 

An example of the video used for both the human and model experiments can be seen below.


The research was also able to replicate another human trait - the perception that time passes differently when paying attention to it, like how the last minutes of a shift at work or in a class seem to drag, while time seems to fly when distracted -  with a simple change of the model’s parameters.

More discussion of the paper can be found on Anil Seth's Neurobanter blog

This work was supported by the European Union Future and Emerging Technologies grant (GA:641100) TIMESTORM—Mind and Time: Investigation of the Temporal Traits of Human-Machine Convergence and the Dr Mortimer and Theresa Sackler Foundation, supporting the Sackler Centre for Consciousness Science.