Stories as Context Sensitive Models

Human knowledge. Vivified in narrative.

After more than a decade of studying psychology, I often find myself reading a great book or watching a movie and thinking, “This author understands people way better than I do.”

Interpret that statement as you wish–as praise for great authors, an indictment of psychology as a field, or even a personal critique of my own failures as a psychologist. All of these interpretations likely hold some truth. But there is another idea that just won’t quite leave me alone. An idea that irritates me, and which I nevertheless find quite plausible. The idea is this; that as a medium, stories are just better than psychological theory.

This isn’t to say that stories are always superior to scientific research. We couldn’t have landed on the moon through storytelling alone, and I’m sure stories won’t unlock all the mysteries of the mind. But there is something special about what stories can do — something that scientific models seem to fall short of. The question I seek to answer is the following: What can stories offer that research cannot? And more specifically, what do they offer that psychological theories cannot?

There might not be a single answer, but we can start by asking a question posed by economist Tyler Cowen in 2005: “Is a novel a model?”

Stories as models

In science, researchers build models. Whether it’s a simple linear equation or a complex simulation, these models rely on specific assumptions about how the world works. Once a model is built, researchers use data, real or fabricated, to help simulate ‘what if’ scenarios. This serves two purposes.

• Calibration: To compare the model’s predictions to reality and adjust it for accuracy.

• Estimation: To predict what might happen in future situations.

Cowen argues that stories serve these same two goals. They offer a form of “mental simulation” (a phrase also used by psychologists) that helps us explore possible realities. However, instead of relying on mathematical equations, stories are simulations of the mental models of the storyteller. The author uses their understanding of the world to craft a narrative that invites both calibration and estimation.

By entering into the dreams of others, the movie Inception explores a similar idea

While Cowen limits his analysis to novels, the concept extends much further. A story could be anything from gossip shared between friends to a blockbuster movie. What matters is that the story presents a sequence of events that conveys an underlying mental model of the world.

Before we talk about why these narrative simulations are so different and powerful compared to the models of researchers, we need to first better understand the concepts of calibration and estimation.

Calibration

In scientific research, calibration refers to the process of adjusting a model to better fit new data. For example, a researcher might notice a bias in their model and tweak its parameters to align with reality. Stories work in a similar way — serving as tools to recalibrate our mental models based on narrative input. Perhaps this is why common reactions to gossip include phrases like “Are you serious?”, “No way!”, and “I don’t believe it.” Such reactions signal a mismatch between our expectations and the information we’re presented with, prompting us to adjust our understanding.

Similarly, a compelling novel or film can challenge our assumptions about morality, human nature, or even the plausibility of utopian ideals. In this way, stories act as pseudo-experiments that put our mental models to the test. Our normally abstract understanding of the world and human nature is forced to make concrete predictions of what will happen next within the constraints of a story that has placed people in bizarre circumstances which we may have never seen before — what we might call an experimental condition. These expectations are then tested against the plausibility of the story we are being told. If plausible we may update our understanding of the world, whereas when implausible, we may reject the story and its underlying models altogether. In this way, surprises in stories, when they are found to be plausible, can serve a similar role to falsification in science — a challenge to the theories and models which led us to predict incorrectly what would happen.

However, not all stories surprise. Some resonate deeply with beliefs we already hold, putting into words and images something we’ve long believed but struggled to articulate. This too is a form of calibration — an increased confidence and understanding of the implicit theories we already hold. Such stories breathe fire into our deepest beliefs and values, and allow us to see them come to life in narrative.

Speaker for the Dead was the first story to resonate with me at such a deep level. As a teenager, I read it repeatedly.

While I have focused on the story receiver, there are also benefits for the storyteller. Writing stories forces storytellers to create characters, plots, and worlds that feel real, which in turn requires them to become keen observers of humans; what they believe, what they find plausible, what they find beautiful. This has potential drawbacks. What is plausible is not always the same as what is true. But the criterion of plausibility constrains writers, and thus stories act as both individual and collective sensemaking. ‘Individual’ in that authors must make sense of others, and ‘collective’ in that popular stories represent a peek into what a culture finds to be plausible portrayals of how humans would behave in the bizarre circumstances in which stories place them in.

Calibration, at its core, is about building better models of the world. And in humans, this is done not through thousands of data points fed to an algorithm as it is in Machine Learning, but through the plausibility of stories.

Estimation

While calibration helps us adjust our models to better fit reality, estimation allows us to project those models into possible futures by asking the question, “what if…?”

When we use actual data to estimate outcomes, we often call it a prediction. But estimation can also involve hypothetical (or fictional, if you prefer) data, which allows us to explore scenarios that haven’t yet occurred, may never occur, or even which may have already passed. This makes the term “estimation” more accurate for our purposes here.

Estimation permeates our lives and daily decision-making. We mentally rehearse conversations, imagining different outcomes depending on what we say or how others might react. We try to predict how an important meeting will go, what challenges we might face, and how to navigate them. Sometimes, long after a heated argument, we find ourselves replaying the scene in the shower, crafting the perfect comeback for a conversation that will never happen again.

These acts of estimation allow us to explore possibilities, helping us anticipate challenges and potential outcomes. We don’t need hard data to run these simulations — they’re based on experience, imagination, and stories we’ve heard from others. Such “What if…” narratives are a sandbox for testing human behavior, giving us an opportunity to explore the space of infinite possibilities to better prepare for the uncertainties of the future.

Literature provides a fertile ground for such estimations, as seen in works like Fyodor Dostoyevsky’s novel The Idiot. When he set out to write the novel, Dostoyevsky didn’t know how it would end. He wrote the novel as an experiment to discover the fate of “a completely beautiful human being” set against the backdrop of a morally bankrupt society. Would this beautiful soul change those around him? Or would they reject him, or even kill him? Dostoyevsky didn’t know the answer. He was just a Christian trying to understand what it would mean to truly live the principles he professed to believe. In this way, the novel became a tool for estimating the consequences of living a truly good and moral life in a corrupt world.

The titular character, while intelligent, is called “the idiot” because his innocence, sincerity, and compassion are misunderstood by others, who see these qualities as naiveté in a world driven by selfish motives and cynicism

Dostoyevsky set out with the explicit purpose of exploring what would happen, which may be the exception rather than the rule. However, even storytellers with tightly planned outlines know that characters sometimes deviate from the arc the author carefully crafted. Storytelling is not merely the act of writing out our beliefs; the process of estimation is partly a process of exploration. Oftentimes storytellers surprise even themselves with how a story unfolds, and often seem unable to see how a story will develop and end even though they put all the characters and pieces into place. In science, this sort of unpredictability is known as computational irreducibility — which is the idea that in complex systems, predicting outcomes can be impossible, and instead you must just let things play out. Just as scientists must run simulations to see how complex systems evolve, storytellers must let their narratives unfold to discover what outcomes result from that which they have set in motion.

Stories — whether personal or literary — are powerful tools for estimation. They help us to take our beliefs and assumptions, and put them into a sandbox where they might fail or flourish. But because the story must follow plausible steps to reach its conclusion, even our own stories may surprise us with how they develop and eventually conclude. This is why stories are not mere representations of our beliefs, but also serve as estimations in the tradition of all scientific models.

Context is that which is outside our model

I opened this essay by asking, “What can stories offer that research cannot?” So far I have offered little by way of answer to that question. Mathematical models, like those used by researchers, are just as capable of calibration and estimation as stories. In fact, they are far superior at it — offering an objective and mathematical precision that the human brain, noisy and biased as it is, cannot hope to ever match. In controlled environments, models can predict outcomes with near-perfect accuracy. For example, a model can calculate the trajectory of a baseball with millimeter levels of precision.

However, this precision comes at a cost. Outside the lab, real-world unpredictability — a gust of wind, or a bit of moisture on the ball — throws the model off course. If researchers cannot isolate, label, and count a variable, the model must ignore it, or treat it as an unpredictable error. Models are only able to be so precise by stripping away context to focus on only a few measurable variables. But this means that this precision comes at the cost of losing most of what is happening in the world.

Context is all that which doesn’t easily fit into our models but which is still relevant, while that which is inside our models we call content. Because context is external to our models, it causes researchers much grief. Many of the most difficult scientific problems are caused simply because we don’t know how to turn context into the content of our models, and thereby incorporate these features of the world into our predictions and theories. But this is how it must be. Models are reductionistic, they simplify and abstract. In fact, the very purpose of scientific models is to turn most of the world into context — to dismiss most of the world as ‘out of scope’ — so that we can focus on the content of the phenomenon we are studying.

But humans cannot do what scientific models do and dismiss most of the world as ‘out of scope.’ Things as seemingly random as “things that happen on Tuesdays” or “all white things” could suddenly be relevant to an individual. Things which were irrelevant a second before may become central to our goals, and as such our way of engaging with the world must be fundamentally open to considering any and all context as potentially relevant — that is to say as content.

Jurassic Park, both the book and movie, are in many ways about the impossibility of anticipating all context

Scientific models, on the other hand, are closed to considering anything outside of the explicit terms of which it consists. This limitation of mathematical models is known as the open-endedness problem. A mathematical model can only account for a finite number of features of the situation, whereas humans have no such limit. We can take anything into account–a casual comment, the weather, a personal memory, or an unexpected event. We seem to swim in the waters of the context which researchers work so hard to eliminate, and which models can’t capture because of their finite nature. We are not algorithms, and so are not constrained by their limitations–we are not bound by what can be isolated, labeled, and counted.

This closedness of models is a strength and a weakness. It allows us to reduce the world in to models that are human intelligible. In fields like physics, where much of the phenomenon is context insensitive, these models have revolutionized the world and gifted us with science. But since humans are influenced by so many constantly changing variables, our ability to control for context in psychology is limited — doomed from the start. Instead of discovering fundamental laws of human behavior, we typically can only identify “tendencies”, like the tendency to be risk averse, or the tendency to be extroverted. But these are averages, and as such cannot offer any insight into the variation that is actually observed. To explain variation we would need a model of these phenomena, but we don’t know how to model something that is as context-sensitive as Risk Aversion or Extroversion. Indeed, it is likely that no scientific model can capture the full context sensitivity that humans handle for even mundane problems.

But there is one type of model that does allow such open ended-ness; stories.

Stories are models with context

Stories don’t start off with a set of relevant variables, but instead develop organically in a way that can incorporate any context as it becomes relevant during the course of events. Only stories, with their open-ended nature, are equipped to explore the depths of human psychology and so are uniquely adept compared to psychological theory.

Stories thrive on context and are not confined by the need for precision, or the demand to isolate variables, and so are not forced to talk about tendencies or averages. Instead, they embrace the full complexity of human nature, showing both the tendencies and variance that humans are capable of. Rather than reducing human behavior to isolated variables in controlled settings, stories weave the messy context into a coherent whole — a tapestry of interrelated factors rather than an algorithm of discrete variables. Stories allow us to see humans in their fullest expression — rich in history, emotion, and reasoning, and immersed in a world of context that scientific models cannot even hope to comprehend.

“The world will never be the same once you’ve seen it through the eyes of Forrest Gump”

In this sense, the difference between a writer and a psychologist is the same as the difference between any practitioner and an academic. While I as a psychologist might understand explicit models of decision-making, that does not mean I would be a great CEO because most of what the CEO knows is mere context to me — that is to say, out of the scope of my decision models. Similarly, while I might understand explicit models of human behavior and decision-making, great storytellers have a hands-on understanding of human nature that embraces the messiness of life, and in so doing can develop accurate portrayals of humans that psychology will always struggle to do.

Recently, psychologist Paul Bloom remarked in a podcast: “I don’t think psychologists are any savvier about human nature in that sort of day-to-day, on-the-spot way than anybody else,” and I think this is the reason. Psychologists, like all researchers, seek to ignore most of the world so that we can focus on small individual parts of it. And in so doing, we often do not know how those individual parts interact with the context around it.

While psychology gives us useful insights and general tendencies, stories often feel more “true” to human experience because they embrace what psychologists must dismiss as context. They give us a way to understand behavior in a way that models and lab experiments cannot, because they mirror how we actually live — in the complexity and unpredictability of the real-world.

You can’t model the modeler, but you can tell stories about ‘em

Of all the tools in the human toolbox, stories are perhaps the most extraordinary. We are the only creature capable of true storytelling, and this may be what distinguishes us from other animals. Through stories, we can communicate our understanding of events, people, and the deepest aspects of human nature, perhaps more faithfully than any scientific model.

Of course, storytellers can also introduce elements that are unrealistic, such as overly pristine dialogue devoid of all the “ums” and “uhs” of ordinary speech, or even elements that are fantastical or futuristic. But even these unrealisms are used strategically to better throw into relief the central mental models which are being ‘tested’ by the narrative.

In Big Fish, an estranged son realizes there was more truth to his father’s fables than he realized

Scientific models have their place. They offer precision and clarity, allowing us to strip away the complexity of the world so that we can focus on some small part of it. But they also have inherent limitations. No model can fully capture the unpredictable and richly textured nature of human life, because we, as humans, are modelers ourselves. A model cannot model the modeler, because models must be finite and closed, and modelers must be continuously open to infinite possibilities.

Stories have other advantages besides what I have talked about here: they are memorable, emotional, and experiential. They engage us at a deeper level, moving us and teaching us things that data and abstractions often cannot, allowing us to learn from a sample of one as opposed to millions of data points. They have been used to change behavior, teach tacit knowledge, and coordinate. As Johnson 2022 says, “We impose narrative structure on information to explain the past, imagine the future, appraise that future, take sustained action, and coordinate actions socially.”

And of course, perhaps the largest strength of models is this; they are not limited to psychology. Stories can simulate the abstract, as well as the specific. Many sci-fi stories are fundamentally philosophical treatises. The book Les Miserables struck me as fundamentally a book about the sociology of the underclass of France. And my colleagues at ShadowBox have even used stories to help expert petrochemical workers convey their mental models of the complex engineered systems they work with to novices! Psychology is merely where I have started this foray into narrative, but it is not where it should end.

Of all the strengths of stories, for me, the fundamental one is this; they are open-ended models. Unlike scientific models, which are constrained by what can be isolated and measured, stories are limited only by the storyteller’s and the audience’s sense of what feels real. And because they embrace the full complexity of human experience, stories allow us to explore the richness of human nature in ways that no mathematical model ever could.