To help caseworkers more accurately assess risk for children in their care, we are building a neural network observer that thinks like a human. After a specified number of cases are rated by humans, the network starts thinking and assessing risks like the human brain would, and will make its own suggestions.