Broken AI. A Test Bench for a Non-invasive Experiment in Computational Neuropsychiatry Joining Krankheit-Operator and Artificial Intelligence

We define a parallel representation of a disease model, concerning the brain, and of sensory-information processing as occurring in the mind of a person affected by a neurodegenerative or neuropsychiatric disease. We consider the recently proposed Krankheit-Operator (K-operator), representing the alteration to the brain connectome, and we join it with the alterations induced at specific points of an artificial neural network, to model the sensory-information processing, classification, and storage within the memory, that are altered in some diseases. We draw the formalism and present a toy model of application, where the disease is represented by the altered weights in a simple neural network. The alteration in the weights of the connectome is mirrored in the alteration of the weights of the neural network. This approach may help model healing strategies within the equivalent artificial neural network, leading to hints for therapeutic strategies for real-brain networks.