Although they are a key component of human cognition, analogies are surprisingly challenging to master.
Starting a new job or visiting a foreign nation are just two examples of situations that people frequently find themselves in that appear to be utterly uncharted and unpredictable. It can be challenging to shake the ingrained feeling of "I have no idea what I'm doing" as we encounter these situations.
However, even though we may not be aware of it consciously, our brains are capable of handling novel circumstances. Humans naturally possess the capacity to draw analogies between unfamiliar situations and those we have already encountered. It's essential to enable us to adapt and even flourish in unfamiliar surroundings.
In many facets of life, from engineering and health to education and advertising, the ability to reason logically is essential. Although our capacity to generate robust analogies is far from flawless, doing so helps us to avoid mistakes and replicate prior triumphs. Drawing connections between past occurrences and the stages of new initiatives and procedures allows us to learn from the past.
Humans have a built-in capacity to draw parallels between unfamiliar situations and those we have already encountered.
Follow me on substack.com.
Recently, scientists have started to investigate how AI might be taught to adhere to the same logic while also avoiding the typical flaws in human reasoning.
However, the complexity of the parallels that these algorithms can create is currently constrained. A group of academics at the University of Southern California (USC) have looked to some of our oldest documented proverbs—the fables—to better understand how far AI may go in mimicking human reasoning.
The difficulty: Analogical thinking can take on many different guises. It might be literal, like equating a lover to a summer day, or figurative, like comparing a fire engine and a tomato based on their colors.
Alternately, analogies can point to parallels in the ways that various items are related to one another, such as how the moon revolves around the Earth and how the Earth revolves around the sun.
They can also involve drawing comparisons between specific events and the factors that led to them, such as the Wall Street Crash of 1929 and the Financial Crisis of 2008.
Reading fables: Before AI can handle such a wide range of parallels, it must improve at making these kinds of connections with human-level precision.
A group of researchers studied Aesop's fables, a collection of short stories with moral lessons that originated in ancient Greece, to better understand the limitations of AI.
The lessons taught in Aesop's fables are frequently extremely similar, despite the fact that their protagonists and locations may be completely dissimilar. Some caution against the perils of greed, naivete, or laziness, while others promote qualities like friendship, charity, and respect.
Finding consensus on which fables should be linked together proved to be more challenging for the researchers than they had anticipated.
Jay Pujara and colleagues at USC started their article, which was made available as a preprint, by analyzing how a person's mind processes the analogies that are offered in the stories. Overall, they found similarities between several pairs of fables with similar messages and highlighted a number of crucial elements that readers might notice as they draw moral conclusions and judgments about the stories.
This framework allows for the pairing of Aesop's fables based on similarities in the connections between the characters as well as in the physical and psychological characteristics of the characters.
The group also took into account how the fables' events related to one another and how their outcomes were comparable. Finally, they demonstrated how morally similar stories may be related to one another. For instance, the moral lesson "know thyself" relates to the lesson "know your worth."
Building on this study, Pujara's team then suggested a series of activities that AI could carry out in order to draw these similarities and analogies.
However, their strategy had one flaw: as they combined their dimensions of comparison, the researchers discovered that it was harder than they had anticipated to come to a consensus on which stories should be matched together.
Their conflict demonstrated that, despite its immense strength, analogical reasoning is remarkably arbitrary—two different individuals could approach the moral of a story in two quite different ways. The group contends that these variations in interpretation ultimately result from the particular information, abilities, and life experiences that each person has thus far amassed.
They now intend to investigate these intricate intricacies of the human mind in greater detail in further studies in order to eventually understand how to incorporate them with AI.
The work provides encouraging initial steps toward an AI that can match our incredible capacity for making comparisons between the present and the past and new, uncharted scenarios. It may even be able to detect connections that are so intricate that they elude even the most brilliant human minds.