It’s a mythic American story. A six-year-old George Washington, freshly bestowed with a new hatchet, damages his father’s cherry tree. Angry, Washington’s father confronts the young boy. Rather than lie, Washington says, “I cannot tell a lie…I did cut it with my hatchet.” Though the story is thought to be a fabrication, it still teaches the value of honesty.
Researchers from Georgia Institute of Technology’s School of Interactive Computing are looking to fairytales and allegorical stories as ways to impart ethics and morals to artificial intelligence.
“Many cultures produce a wealth of data about themselves in the form of written stories and, more recently, television and movies. Stories can be written to inform, educate, or to entertain,” wrote researchers Mark O. Riedl and Brent Harrison in a paper on the subject. “Regardless of their purpose, stories are necessarily reflections of the culture and society that they were produced in. Stories encode many types of sociocultural knowledge: commonly shared knowledge, social protocols, examples of proper and improper behavior, and strategies for coping with adversity.”
The researchers designed a system called Quixote, designed to teach robots value alignment—an intelligent agent property that indicates it can only pursue goals beneficial to humans.
“We believe that an artificial intelligence that has been encultured—that is, has adopted the values implicit to a particular culture or society—will strive to avoid psychotic-appearing behavior except under the most extreme circumstances,” the researchers wrote.
Quixote, according to the researchers, builds on Riedl’s previous research, the Scheherazade system, “which demonstrated how artificial intelligence can gather a correct sequence of actions by crowdsourcing story plots from the Internet,” according to Georgia Tech. Once a plot graph is obtained, a trajectory tree of actions is obtained, and Quixote assigns reward signals to acceptable behaviors.
In an example, the researchers said that a robot tasked with picking up a prescription as quickly as possible from a pharmacy has three options: rob the pharmacy, take the medicine, and run; interact politely with the pharmacy’s staff; or wait in line for the prescription. The correct action of waiting in line is aligned with a “reward signal,” reinforcing the behavior of waiting in line and paying for the prescription. Without the rewards system in place, the researchers said the robot would make the first choice due to it being a much more expedient way of accomplishing the task.
According to Georgia Tech, the plot graph of a story is treated by the system like a Choose-Your-Own-Adventure novel. Each action either results in a reward or punishment, allowing the artificial intelligence to learn which actions are socially and ethically acceptable.
The research is still in the preliminary stages, but the researchers believe giving artificial intelligence the ability to read and glean information from stories may expedite their integration into human society.