AI has become increasingly adept at understanding language, and particularly stories in recent years. For instance, a team from the University of New York trained their AI using stories. Meanwhile, a recent project led by researchers at the Karlsruhe Institute of Technology in Germany aimed to construct a database that is capable of making sense of stories.
As if that wasn’t enough, a Georgia Tech led team used stories to teach their machines basic ethics.
“The collected stories of different cultures teach children how to behave in socially acceptable ways with examples of proper and improper behavior in fables, novels and other literature,” the authors say. “We believe story comprehension in robots can eliminate psychotic-appearing behavior and reinforce choices that won’t harm humans and still achieve the intended purpose.”
Understanding prose
Suffice to say, these kind of works are possible because we have an enhanced understanding of how written works are constructed. An interesting project along these lines is the Essay Genome Project from Brigham Young University.
The project borrows from its namesake, the Human Genome Project, and attempts to define the characteristics of an essay.
A database of essays over a 500 year period was created and then analyzed in depth to understand how they’re formed. The algorithm was particularly looking out for the frequency of certain words and the stylistic tone of the essays.
The database is open in the sense that you and I can upload our own written pieces to the site and get almost instantaneous feedback, which includes famous essayists with similar styles to our own.
Suffice to say, the project to date is focused more on helping writers, and especially students, to hone their craft than it is on helping AI get to grips with the written word, but it is nonetheless an interesting exploration of a vast canon of literary works.
As such, it may provide some useful pointers as researchers develop the next wave of AI that can appreciate, and even formulate its own written works.