In the Harry Potter series, the Sorting Hat places students in one of four houses in the wizarding school of Hogwarts. An IBM employee working on the solutions architecture behind the Watson natural language processing system built one of his own for his children.
In an interview with Tech Insider, solutions architect Ryan Anderson explained how he used Watson to interpret which Hogwarts house someone could be in based on how they describe themselves. He developed it for his daughters, Lucy, 8, and Julia, 6, both of whom are “mad keen on ‘Harry Potter’”.
Watson’s Natural Language Classifier and a speech to text feature are attached to the hat to interpret spoken language. In order to determine which words were associated with which house, Anderson and Lucy created a list of traits associated with each house in the books. This served to “set a ground truth,” teaching the system an initial set of words that would generate a particular outcome.
Since Watson is a machine learning system, its sensitivity to particular words and its ability to creatively interpret them will grow as the system does. It’s connected to the internet, and can scan for related words. This can then be tweaked if the system starts assigning houses to words that don’t quite fit the personality traits assigned to that house in the books.
There are also some hardware responses: Slytherin, the house that produced most of the evil characters in the books, elicits furrowed eyebrows, while the “hero” house Gryffindor gets a green glow. Anderson hopes to have even more personality built in by the time Halloween comes around.
IBM Watson’s machine learning capabilities are also being used in edge computing and elsewhere in the Internet of Things.
