Human beings tell stories. And they tell them all the time. About 40 percent of all everyday conversation involves some form of storytelling—sharing the moments of our day, gossiping about the exploits of others, commiserating with friends over past tragedies, and sharing laughter over past joys. Can artificial intelligence (AI) share stories with us? And can these stories be meaningful?
Research from The Family Narratives Lab has shown that shared family storytelling begins virtually at birth, with parents and grandparents whispering stories into infants’ ears, and continues throughout the preschool years, as parents help their young ones structure their experience into coherent story forms. By the end of the preschool years, children are able to tell stories that follow a discernable chronology, but perhaps more important, even these early stories include emotional aspects: intentions, motivations, and emotions of the characters and why people behave as they do, as well as evaluative aspects of characters and events—if what happened was good or bad, if people acted out of love or anger. We tell stories that convey human meaning about people in a complex world.
Through adolescence, these stories become more complicated and more individual—both stories that we tell about ourselves and others. Stories begin to include more and more information about why people (including ourselves) behave as they do; stories include more complex perspectives and a more nuanced understanding of human motivations.
Fictional stories may play a special role—we learn about faraway places in time and space, and we learn about different peoples and different perspectives. People who read more fiction show higher levels of empathy, and they feel more deeply about a wider variety of characters’ predicaments. Empathy certainly involves a cognitive component, understanding why people feel the way they do. But perhaps more potent is emotional empathy, actually feeling alongside the characters. Clare Grall and her colleagues found that it was during the most emotional aspects of the story that tellers and listeners showed synchronized brain activity—we truly resonate with others’ emotions. Empathic storytelling and listening connect our brains and connect us to the human experience.
Studies of AI and Storytelling
So can AI tell stories that we resonate with, learn from, and care about? It is still early days for AI, but so far, the findings suggest that, although AI can tell stories, they somehow miss the human mark. My colleagues and I have conducted several studies comparing personal stories produced by humans and AI-generated stories, both on telling stories and in responding to stories.
Asked to tell stories about experiences that are self-defining, meaningful turning points in one’s life, AI can generate such stories very easily. But when human participants are asked to judge whether a series of stories was generated by humans or by AI, they had no trouble distinguishing—they were about 70 percent correct, so certainly not perfect, but pretty good. When asked to explain their judgments, humans agree that AI stories are “too perfect”—perfectly grammatical, big vocabulary, a little too insightful. And the emotions do not “feel real.” We humans are a bit more imperfect, a bit more sloppy in our emotional responses. And we can tell the difference!
We then asked AI to evaluate a set of stories produced by humans about difficult and challenging personal experiences. Often, humans provide a redemptive resolution to such stories, finding the silver lining, learning a life lesson. But not always, and, even when they do, humans often express a slight hesitation, a slight nod to things being more complex, that nothing is that easy. But AI interprets all the human stories as highly redemptive. The silver lining shines with no blemishes. And when humans are then asked to judge these evaluations without knowing they were AI-generated, once again, humans found them “off.” Too simplistic, too redemptive, not really understanding the nuance of human emotion. In a word, insensitive.
Why Humans Tell Stories
So what does this tell us about AI’s ability to tell stories? Certainly, AI is excellent at gathering large amounts of information to abstract typical story structure and content and to produce a reasonable response to a story prompt. But AI may not yet be able to understand why people tell stories in the first place, why people share their experiences with each other, and why this matters.
Humans share stories to share a part of themselves, to learn about other people, and to forge meaningful human connections. Stories are told for validation, vindication, entertainment, and to try to make sense out of what are often senseless happenings. Stories are told for social and emotional reasons, and this is what resonates with human listeners. AI can fulfill the cognitive aspects of storytelling and even listening, but not the emotional aspects.
Will AI ever be able to tell stories that resonate with human experience? Stories that feel authentic, that resonate as real, and that connect with us as stories of human foibles—human tragedies and triumphs? Stories of others resonate with us when our emotional experiences connect in some way. If AI cannot have emotional experience, can AI ever produce authentic stories?
In Shannon Vallor’s words, AI is a reflection of human experience, and “a reflection of a face is not a face. A reflection of a flower is not a flower; it needs no water. A reflection of a flame is not a flame; it does not burn. A reflection of a mind is not a mind, and it does not think.” I would add, a reflection of a story is not a story; it does not express the human condition, the emotional fragility and resilience of us all.