A piece of software called Xapagy has learned to tell familiar stories like Little Red Riding Hood not through rote memorization but by maintaining a web of interconnected events it can use to deploy a narrative. “When [the software] comes across words in new stories, it looks for familiar connections in its memory. If it finds any, it uses them to predict what will happen next and then tells the story. The trick is that each word can have many different associations in Xapagy’s memory, depending on the stories it has read.”
What’s the Big Idea?
While most AI systems use the rigid rules of logic to formulate expressions, which can prevent them from tackling unfamiliar tasks, a system based on forming narrative frameworks could result in a much more flexible way of machine learning. AI researcher Andrew Nuxoll at the University of Portland in Oregon says Xapagy is “entirely unique”. The work is a step toward human-like AI, said Nuxoll: “I am confident that when we do create such a brain it will have a capacity to remember, learn and re-learn from its past.”