Many of us have grown up with fairy tales, whether we realize it or not. Most movies, cartoons, and pretty much everything that has a story were in one way or another inspired by classical fairy tales. Since they were such a huge part of our childhoods, most fairy tales actually work as teachers, allowing children to learn morals easily. But what if we applied that to something else? A team of researchers from Georgia came up with a way to teach robots right and wrong with fairy tales.
Fairy tales and morality
In modern culture, pretty much everything is inspired by fairy tales. Superman is Prince Charming, ISIS are the forces of the evil King, and Donald Trump is arguably the court jester with the huge ego. With all these models, we were imprinted with a sense of right and wrong ever since we first read these stories.
With the advancements we make in IT every single day, it’s unavoidable that at some point we invent a fully operational artificial intelligence that would beat the Turing test. Brilliant minds of our generation such as Bill Gates, Elon Musk, and even Stephen Hawking are worried about a potential robot uprising.
The Quixote method
In order to prevent the assumed to be inevitable robot uprising, a team of researchers from the Georgia Institute of Technology came up with the Quixote system. Based on Scheherazade, the team’s previous project, Quixote is meant to teach artificial intelligences the difference between right and wrong.
While Scheherazade works by building interactive works of fiction by compiling story plots from the internet, Quixote uses those in order to learn proper behavior. This works with the help of behavioral science, by providing a reward signal for a morally right action, and a punishment signal for a morally wrong one.
Quixote gets to act as the main character in the story created by Scheherazade, and interact as such. For example, in a scenario where Quixote is in a pharmacy needing to but medicine for a human, it will most often choose to steal it, as it is the most efficient way. But it will get a punishment signal if it does that, eventually learning that sitting in line is the morally correct action because of the reward signal it receives.
Despite the huge number of ways in which this system could be used incorrectly, the team is quite sure that this is the quickest and easiest way to make sure robots understand human morality systems in the absence of a human instructions manual.
Image source: Pixabay