The Akinator is a popular neural network turned into a guessing game, where the application attempts to guess a real or fictional character that the user chooses based on a series of "yes" or "no" questions. Each question represents a node of its neural network and the following question represents the most efficient path from the node of the previous nodes. Proper programming must condition the nodes to make better decisions as it traverses for all future iterations. In other words, the programming behind Akinator should become more efficient and one step closer to an optimized traversal every time a user enters an output, assuming they answer the questions to the best of their abilities.
Although, the conditioning to make the neural network learn correctly is complex and the actual algorithm is confidential, we were still able to draw several conclusions through extensive testing. In brief, we concluded that the Akinator must somewhat resemble a binary search tree by partitioning its possibilities into equal numbered sub-groups, "Yes" answers typically partition the set better than "No" answers, there is a recency bias, and the Akinator takes advantage of the user to gain more "knowledge". These conclusions are further discussed within the dropdown menu selections of this page tab.