Credit: Chris Kim/USC
Key Points:
- A team of scientists aims to teach machines how to reason by using analogies, which was once thought possible only by humans.
- Machines cannot yet recognize relationships between objects and circumstances that on the surface appear dissimilar.
- Researchers may have found a solution using Aesop’s fables, many of which feature different plot structures, but similar morals and themes.
Scientists at the USC Information Sciences Institute (ISI) are breaking new ground in the field of artificial intelligence (AI). They are teaching machines to use analogical reasoning by programming them to recognize patterns found in Aesop’s fables. The research team presented their paper last week at the International Joint Conference on Artificial Intelligence.
Analogical reasoning is a type of logic predicated on the discovery of a shared similarities between two objects or experiences. What is understood about one circumstance may be used to deduce new knowledge about the other. The core concept underpinning analogical thinking is that when there are significant parallels between circumstances, additional similarities are expected.
A toddler, for instance, may pick up a basketball for the first time in their life, and perceive that its spherical shape corresponds to the shape of a beach ball they have previously rolled around in the sand. Analogical reasoning would allow the toddler to conclude that the basketball would also roll across the court due to the objects’ similar shapes.
“People working in AI have been trying to get the same level of reasoning that humans have into AI systems,” explains Jay Pujara, “and it is a really hard challenge trying to mimic the analogical reasoning humans take for granted.” This is due in part to the fact that analogical reasoning allows humans to form meaningful connections between objects or circumstances, even when they first appear dissimilar, a feat AI cannot yet achieve.
Surprisingly, scientists are overcoming this modern hurdle with the help of wisdom from the 6th century B.C. Pujara says the team chose Aesop’s fables because many of the stories appear unrelated since they follow different plotlines. The same fables, however, often share similar morals and messages. The researchers are applying natural language processing (NLP) techniques—which make words understandable to computers-—to help AI recognize those relational structures.
Interestingly, the research team found that analogical reasoning is a more personal, idiosyncratic process than it previously thought. Past personal experience, for example, influences how a person makes connections when presented with new objects or circumstances. This makes analogical reasoning an intensely individual experience, which researchers say will motivate them to better understand how humans engage in the process.