This talk summary discusses the concept of recovering function from missing information during brain scanning. The video highlights two examples of missing information and explores whether machine learning techniques can fill in those gaps. One example is the absence of neuron biases, and the other is the removal of retinas in visual representation. The research project aims to strategically remove pieces of information from artificial neural networks and investigate the difficulty of recovering the network’s function. The talk also mentions residual networks, which can still produce satisfactory results even when certain layers are removed. The potential for software intelligence is discussed, suggesting collaboration between neuroscientists and machine learning engineers to identify common missing information, test network robustness, and develop methods for retrieval. The initial phase of the project involves specifying different types of missing information, requiring time and computational resources.