AI predicts ‘full behavioural repertoire’ of pond-dwelling creature
Image credit: Dreamstime
Researchers at Columbia University have developed a machine learning algorithm capable of forecasting the behaviour of a simple animal.
Animal behaviour experts spend years – or even decades – observing species to understand their social behaviour, reproductive rituals and feeding habits. While brainy creatures such as chimpanzees, cetaceans and corvids demonstrate intricate behaviour that can be as hard to predict as human behaviour, simpler organisms can have their behaviour approximated with computer simulations.
“People have used machine learning algorithms to partly analyse how a fruit fly flies, and how a worm crawls, but this is the first systematic description of an animal’s behaviour” said Professor Rafael Yuste, a Columbia University neuroscientist who heads a lab aiming at deciphering the relationship between brain activity and behaviour.
For instance, researchers at Vienna University of Technology have translated the entire neural system of a simple worm into computer code and taught it tricks.
Yuste’s team of engineers at Columbia has now demonstrated that the full range of behaviour of a Hydra – a tiny freshwater organism with tendrils which is mostly sedentary and does not appear to age – can be described with a spam-filtering (“bag of words”) algorithm which analyses the frequency of certain movements.
The researchers used hours of video footage of Hydra to teach the algorithm the “full behavioural repertoire” of the creature. Hydra’s behaviour is controlled by hundreds of neurons running along its body as it does not have a brain or muscles, and its somersaulting movement, feeding behaviour and defence against predators occur in patterns that can be predicted by the algorithm.
“Now that we can measure the entirety of Hydra’s behaviour in real-time, we can see if it can learn, and if so, how its neurons respond,” said Yuste.
The algorithm recognised 10 Hydra behaviours, and analysed how these behaviours occur in respond to stimuli, such as changing the brightness of its environment or feeding it. Bewilderingly, Hydra seemed to follow the same set of behaviours under the full range of stimuli used by the researchers.
Yuste and his team hope to create a fuller model which demonstrates how its neurons create behaviour. This understanding could pave the way, Yuste believes, for sophisticated, biologically-inspired controls for technologies, such as for maintaining stability while navigating ships and planes.