Paralysed by Surprise: Testing the Active Inference Theory with VR
Name
Taavi Gilden
Abstract
In recent years, there have been new theories to change the current understanding of how the brain perceives and processes stimuli. The long-standing belief has been that the brain passively waits for sensory stimulation and the received signals are processed in a hierarchical manner to control behaviour. Instead, a theory called active inference states that the system is always active and rather than waiting for the sensory stimulation, it tries to predict it beforehand. According to this theory, the crucial input to the hierarchical system is based on perceived deviations from the predicted state also known as prediction errors. This means that an action is a way of minimizing the prediction error by trying to fit the actual sensory input to the predicted one instead of it just being a response to an input. According to this theory movement is not triggered by motor commands, but instead by trying to minimize the prediction error in the intended movement. This means that in order to move it is required to suppress the sensory input. In other words, it is necessary to withdraw attention from the actual sensory consequences of the movement to allow for movement to happen. If such withdrawal cannot happen the movement should not take place. The goal of this project was to create software to test this peculiar prediction of the theory. In particular, we tried to interrupt the withdrawal of attention during movement in an experiment using the latest virtual reality technology. In the produced environment people were presented with a task to raise their arm as a reaction to stimuli they received from their surroundings. By manipulating the saliency of the visual environment we tried to affect the reaction time of the movement.
Our preliminary findings, however, do not provide clear enough evidence to support the theory of active inference. It could be explained by the small amount of data collected, which might not be sufficient enough to draw conclusions from. Additionally, the saliency of the shapes might have too little impact on disturbing the process of sensory attenuation. Many different factors might be at play here, but the experiment is up for improvements and allows for the theory to be studied further.
Our preliminary findings, however, do not provide clear enough evidence to support the theory of active inference. It could be explained by the small amount of data collected, which might not be sufficient enough to draw conclusions from. Additionally, the saliency of the shapes might have too little impact on disturbing the process of sensory attenuation. Many different factors might be at play here, but the experiment is up for improvements and allows for the theory to be studied further.
Graduation Thesis language
English
Graduation Thesis type
Master - Software Engineering
Supervisor(s)
Jaan Aru, Madis Vasser
Defence year
2017