5 Jan 2019

Notes on Narrative, Intelligence, and AI 

Escher, Relativity.

Notes for the 2019 MIT IAP Seminar. See the syllabus here.

Rob also prepared notes which provide some important background context on our motivations for this topic.

i. Notes on the definition and history of narrative

As background for our discussion of the current landscape of narrative and the future of narrative, I think it’s useful to first examine some of the functions of narrative and consider what types of content most lend themselves to the form. The reading for our discussion called to mind three distinct (but interrelated) functions of narrative: it is (a) a communication method, (b) a technique for exploring and testing new ideas, and (c) a technique for explaining observations about the world. I’ll briefly discuss each of these functions, how they influence each other, then share some ideas about what types of content narrative is best suited for and why, given these functions.

Narrative Communication

When I hear the word narrative, I think of stories, and when I think of stories, I think of storytelling. Narrator implies audience. It implies communication of a message.

And when the message is a story, the message is typically (but not always) some sort of cultural learning.

But there are other types of messages which are not stories, other types of communication which are not narrative. A mathematical formula. An analytical essay.

In this infamous rant, David Mamet notes the key element which differentiates the story (dramatic content) from other types of information:




Drama involves the intentional withholding of information. If your goal is clear communication of a message, when you withhold information, you run the risk of misinterpretation, of the audience completely missing the point–an outcome which might have been avoided if you had been totally explicit. So, why would you ever withhold key details?

Because, Mamet argues, “the audience will not tune in to watch information.” It is precisely the withholding of information that attracts and holds the attention of the audience.

This makes intuitive sense. It can be difficult to stay awake during a lecture that is a mere litany of facts, yet we can stay awake–on the edge of our seats, sometimes–for hours listening to a great story.

Perhaps this is because the story makes us a more active participant than the litany of facts. We imagine ourselves in the world of the story, with the information that the characters have, wondering how the characters will resolve the problems they encounter. It can both be satisfying when the characters take the same actions we imagine ourselves taking in their situation, and informative when they do something different that surprises us.

Perhaps because we are more actively engaged in the problem solving of the story, we are more likely to retain the things we learn from stories than from attempts at rote memorization of facts.

Narrative Exploration

Many animals learn about the world purely empirically–they explore the world, collect observations about cause and effect, and model future behavior based on past observations. Social animals can leverage collective knowledge, sharing learnings their observations with others, saving them the cost (and perhaps pain) of each individually collecting the same observation.

When humans tell stories, they’re doing more than sharing observations. They’re mimicking one of the advanced properties of human intelligence: simulation.

Simulation is a key way that individual humans learn about the world without subjecting themselves to danger. If I toss a cantaloupe off a cliff and watch it smash as it hits the ground, I can imagine what would happen if I throw my own body off of the same cliff. I can learn in a simulation of the world, rather running an experiment in the physical world and collecting an observation.

Simulation is such such a powerful cognitive ability that Joscha Bach goes so far as to enumerate it in his list of functions necessary for consciousness (From Computation to Consciousness). While humans are the only animals that possess the ability for simulation, modern machine learning software (eg, for driverless cars) has benefited from training on simulated (vs observed) data (see also World Models for a more technical explanation of how and why training in a simulated / virtual vs a ‘real’ environment works).

Simulation is particularly useful for exploring a space where cost of error is high. Achieving 90% confidence in a belief is preferable to achieving 100% confidence by validating the belief when validation carries a high cost (for example, when there is a high probability of error that incurs us some cost, or when there is even a low probability of a gameover outcome).

Narrative Explanation

Narrative exploration uses simulation as an alternative to collecting observations entails asking the question: what would happen in the future if… Narrative explanation, on the other hand, is a technique for understanding observations that we have already collected, looking for a pattern or cause that would give us a more predictive world model. Narrative explanation asks: what might have happened in the past to explain this observation…

In The Black Swan, Nassim Taleb argues that our desire for explanation is so strong that by default we seek (and are apt to see) a narrative explanation for every observation. When we discount randomness in favor of overfitting to narrative causation, we are prone to error:

We like stories, we like to summarize, and we like to simplify, i.e., to reduce the dimension of matters…. The [narrative] fallacy is associated with our vulnerability to overinterpretation and our predilection for compact stories over raw truths. It severely distorts our mental representation of the world; it is particularly acute when it comes to the rare event.

Beyond our “predilection for compact stories over raw truths” – our bias towards any causal explanation over a random one – we also tend to have a bias towards explanations that confirm our existing models of the world. This bias towards the familiar not only is problematic when we construct our own stories about the world, but also makes us vulnerable when narrative is the vehicle for communicating an idea–a classic technique demagogues use when constructing false stories that further their own ends is to ensure the stories they tell confirm the existing beliefs and fears of their followers.

Many great novels and films explored these pitfalls of narrative through the narrative form itself–for example, by using an unreliable narrator to tell a story. Stories that use this meta-move call into question the trustworthiness of narrative as a mode of thinking and communicating. This capacity for the narrative form for self-examination mirrors a similar capacity for self awareness that exists in consciousness.

Narrative Content

Stories with an unreliable narrator, where the purpose is to examine the narrative form itself, comprise only a small portion of the many stories that have ever been told. Given some of the functions of narrative I’ve enumerated–communication, exploration, explanation–it is next worth asking what other topics lend themselves to the narrative form.

In The Poetics, Aristotle notes that Character is one of the fundamental elements of a story. All but the most experimental narratives involve a character, usually multiple characters, and the central figure of a hero is so prevalent that scholars like Joseph Campbell have gone so far as to attempt to describe the basic pattern all stories have in common as the ‘monomyth’ (The Hero with a Thousand Faces).

Stories of Character explore questions of morality and social interaction, domains where a violation of trust can cause significant and permanent damage to a relationship. Given this high cost of error, running simulations that explore different styles of social interaction via the narrative form makes a ton of sense.

Narrative is also a great way to explain social interaction–the behavior of someone we encounter in the world may seem to us inscrutable, crazy, obscene, immoral, barbarian. But human behavior is largely a function of context, and something that seems crazy given our own context may make more sense if we’re able to adopt the context of someone else. Narrative can be a tool for communicating the backstory of a particular context motivating the actions of “the other,” a tool for building understanding and empathy.

In The Philosophical Baby, Alison Gopnik suggests that we learn moral behavior through empathy:

Empathy requires that you recognize the similarity between the feelings of others and your own feelings, but it also requires that you take on those feelings as your own. When you imitate an emotional expression or an action or an intention you make that feeling or action or intention your own—you act as if you were experiencing that mental state rather than just observing it.


If witnessing another person’s pain literally feels painful to a baby, he might act to try to alleviate that pain as he would act to alleviate his own misery. If witnessing joy makes him joyful, he might try to bring about that joy in others.

For Gopnik, empathy is the simulation of the feelings of others. The act of empathy, you might say, is a form of narrative, in which we simulate someone else’s worldview in order to understand their behavior. And, morals are the set of rules we develop–based upon this understanding–to govern our interactions with others.

Once we have developed morals to govern our social interactions, it can be more efficient to dispense with the effort of empathy and go on autopilot following the set of moral rules we have developed. While this may be more efficient, the ruleset we developed may be incomplete or grow outdated. Gopnik suggests a return to empathy as a way out of this failure mode:

Rules allow us to perform complex, coordinated behaviors—they let us help other people in new and powerful ways. But intimate, emotional empathy is a force that can change even the most entrenched rules. If we discover that a rule leads to harm rather than good we can reject that rule. This is especially true if we experience that harm in the rich, intimate way that comes from interacting face-to-face with a real person in real life.

Often a return to the intimate empathy of infancy—that immediate sense of how other people feel—can be the most powerful way to change what people do. For example, we dehumanize people in the “out-group”—people who are not like us. This impulse is deep-seated and very difficult to overturn completely. One of the best ways to change it is to actually become intimate with the out-group—to recognize that those people are actually like me. People who come to know someone well who is openly gay are much more likely to support gay rights. Individual stories are powerful agents of moral change—often more powerful than rational arguments.

If empathy is indeed a force for clearing out entrenched or outdated morals, narrative could be the tool for inspiring/distributing that empathy at scale, for humanizing the “out-group” suffering from the injustice of the moment.

This, I think, is one of the most attractive features of the narrative form–the promise that it might be a more effective tool for enacting moral change at scale than rational argument.

Concluding Thoughts, part i.

At it’s best, narrative can be a tool for exploring modes of social interaction, for developing empathy ourselves, for inspiring empathy in others, and for driving moral change. At its worst, narrative thinking can lead us to confirm our own biases and storytelling can be used as a tool for manipulating listeners, by earning their trust and confirming their biases, yielding support for arbitrary ends.

In a sense, we can think of narrative as a technology, which can be used to do good, or to do harm, to improve our understanding of the world and of each other, or to perpetuate ignorance and false beliefs.

ii. Notes on the current landscape of narrative

Someone asked me when signing up for this class why there isn’t more about video games in the syllabus–this is a great question to kickoff thinking about the current landscape of narrative.

I don’t play a ton of video games these days, but I have a lot of really fond memories of playing video games in my childhood. Some of my favorite games were very narrative in format: Final Fantasy VI and VII, for example. Also, Metal Gear Solid. Other favorites had minimal narrative and the fun was more in exploring a fictional world and solving puzzles: Super Mario World, The Legend of Zelda, and Super Metroid come to mind. Finally, there were many games were the pure mechanics of gameplay itself comprised the fun: Tetris, Mariokart, GoldenEye, eg.

This last category of game feels more like chess or a sport, where the fun is in building some sort of skill and competing, either against software or against other people. I don’t think this sort of game classifies as narrative. It is these skill based games, many of which fall under the sub-genre ‘casual games,’ that are most popular today.

Rather than challenging our existing world model and leading us through some sort of thought experiment, these games feel like they tap into our hard wired circuitry of hand eye coordination, response to visual stimulus, abilities for tactical and strategic thinking, and the dopamine response to pseudo-random reward. Fortnite, perhaps the most popular game today, combines all the fun of addictive gameplay with social networking features like stats and bragging rights and throws in the opportunity for consumerist self expression via the purchase of digital bling (How Fortnite Became the Most Popular Video Game on Earth).

The popularity of casual games–despite the infinite possibility for immersive narrative presented by the medium of video games–calls to mind Susan Sontag’s observations about the tourist’s compulsion to photograph:

As photographs give people an imaginary possession of a past that is unreal, they also help people to take possession of space in which they are insecure. Thus, photography develops in tandem with one of the most characteristic of modern activities: tourism. For the first time in history, large numbers of people regularly travel out of their habitual environments for short periods of time. It seems positively unnatural to travel for pleasure without taking a camera along. Photographs will offer indisputable evidence that the trip was made, that the program was carried out, that fun was had. Photographs document sequences of consumption carried on outside the view of family, friends, neighbors. But dependence on the camera, as the device that makes real what one is experiencing, doesn’t fade when people travel more. Taking photographs fills the same need for the cosmopolitans accumulating photograph-trophies of their boat trip up the Albert Nile or their fourteen days in China as it does for lower-middle-class vacationers taking snapshots of the Eiffel Tower or Niagara Falls.

A way of certifying experience, taking photographs is also a way of refusing it—by limiting experience to a search for the photogenic, by converting experience into an image, a souvenir. Travel becomes a strategy for accumulating photographs. The very activity of taking pictures is soothing, and assuages general feelings of disorientation that are likely to be exacerbated by travel. Most tourists feel compelled to put the camera between themselves and whatever is remarkable that they encounter. Unsure of other responses, they take a picture. This gives shape to experience: stop, take a photograph, and move on. The method especially appeals to people handicapped by a ruthless work ethic—Germans, Japanese, and Americans. Using a camera appeases the anxiety which the work-driven feel about not working when they are on vacation and supposed to be having fun. They have something to do that is like a friendly imitation of work: they can take pictures.

A tourist (or an Instagram user) who photographs, a gamer who chooses the familiar casual game over the avante garde narrative, wards off the foreign – “The very activity of taking pictures [and playing casual games, I would argue] is soothing, and assuages general feelings of disorientation that are likely to be exacerbated by travel. Most tourists feel compelled to put the camera between themselves and whatever is remarkable that they encounter.” Sontag locates this compulsion in the tourist’s habituation to robotic work – an open possibility space is frightening, the confines of a limited set of moves make a game feel safe.

Critics of social media take a similarly echo Sontag’s critique: the obsession with the image refuses experience by favoring a highly edited ‘hyperreality’ over reality. Rather than portraying their actual lives, people project the ideal version of themselves on Instagram.

While there is validity to the concern that social media can generate psychological stress for participants (eg, for teens struggling to acquire a sense of identity and self esteem), there’s also a more generous interpretation of the hyperreality of Instagram: social media is no more fictional than the social reality and identity we each construct in the physical world – both are creative acts, assertions by actors authoring the stories they’re telling the audience they encounter in life.

Perhaps nowhere is it more clear than in the realm of social media that as storytellers we crave (need?) audience. The narrator’s (and story’s) desire for an audience leads to another set of important questions to address in our discussion of the current landscape of narrative. What medium(s) still have an audience? With so many new forms of media (video games, social networking, video and text and photos in every form factor imaginable available instantly, from anywhere in the world via a mobile phone), are some formats of the narrative dead? Is there any appetite for longform narrative anymore?

Something I’ll often ask when I meet someone in college or high school is: do you and your friends still watch movies? I still love the feature film, but know that my own habits have changed quite a bit over the past few decades–eg, I read far fewer novels than I did when I was younger (and I read more novels than almost all of my friends). When I adopt the role of storyteller and consider which medium to use to tell a story, I often think of this scene from La Notte:

Mr. Gherardini : So as I was saying, my friend, it’s absurd to speak of wealth now. No one’s wealthy anymore. But should anyone still think of becoming rich, my only advice would be, “Don’t worry about the money.” I’ve always looked upon my businesses as works of art. Whatever profit I earned was of practically no concern to me. The important thing is to create something solid; something to be remembered. What sustains a writer - you for instance - is certainly not the idea of profit but a sense of necessity. You write because you have to…

Signora Gherardini : Still, one has to live.

Mr. Gherardini : I never worried about that. Life is what we make of it through our own efforts. What would you do if you couldn’t write?

Lidia : A few years ago he’d have killed himself. Now I don’t know. Tell us.

Giovanni : I don’t consider myself that important. There are other solutions. A writer of today constantly wonders if writing isn’t some sort of irrepressible but outdated instinct. This lonely craft of painstakingly joining one word to another that absolutely can’t be mechanized.

Mr. Gherardini : Are you sure of that?

Giovanni : No. But you industrialists have the advantage of constructing your “stories” using real people, real houses, real cities. The rhythm of life today is in your hands. Perhaps even the future.

Mr. Gherardini : Are you one of the many worrying about the future? I’m building my own future, though the present keeps me plenty busy. Besides, the future will probably never come. Who knows what the future holds? Perhaps our privileges will be swept away. That would actually be a good thing. When I was young - long ago, now, sadly - I imagined a world like this, and I set to work creating such a future. Bah!

Sometimes it feels that in the age of the image, the casual game, and the infinite stream of pull to refresh media, that long form narrative is outdated, because no one has the attention span for it anymore. We live in an era of unprecedented information abundance, and a 700 page novel or even a 2 hour feature film present a significant cost and time investment compared to scrolling through photos for 30 seconds at a time.

Some might celebrate the age of bite sized media–the emergent beauty of it when viewed as a whole (the New Aesthetic)–and embrace the decline of narrative. If I put on my ‘child of the early days of the internet’ glasses, and look back to what got me excited about watching a bunch of the trends of the New Aesthetic as they happened, I would wax poetic about decentralization. I would argue that narrative has authoritarian elements–a narrative, after all, has an ‘author.’ The narrator speaks, the audience is relegated to listening. We should be advocating for a more participatory form of art, watching it emerge from the commons of the internet and new media.

And perhaps we should be celebrating the modern preference for pull to refresh social media and video games as the victory of this new form of participatory media.

As much as I love the idea of this sort of distributed creativity and emergent art–it was one of the things that initially got me very excited about the internet–I think it’s generous to label much of popular media today as “participatory.” Playing Candy Crush and scrolling photo feeds feel about as participatory as pulling the handles of the slot machines (with which our apps share many design principles).

Rather than helping us learn an important but expensive to acquire lesson about the world through the safety of a simulation (as I argued the best narratives do), these addictive media rarely teach us anything new. They are less respectful of audience than a narrator trying to share an important lesson; they merely hack the information processing pathways of the audience for their own ends.

The audience is not the only victim of this attention hacking–creators can be just as much slaves to the attention market as audience. Humans engineer musical hooks using a formula for catchiness, write nonsense lyrics and produce insane music videos optimized not for a human audience but for the Youtube autoplay algorithm, or produce superhero after superhero sequel in franchise after franchise–the attention algorithm is commanding the creators, as well.

When I see all of this pandering for attention by creators and thoughtless consumption by audience, I think we can stop debating when the singularity will happen. The algorithms have already taken over, in the form of markets. The singularity happened with the invention of money, the API for people, and it has just become more pervasive with the rise of the dopamine hacking attention economy.

Although I bemoan the dominance of superhero movies, I don’t want to sound like a hipster who thinks that nothing popular can also be cool. There are some great movies which are both thought provoking and popular–The Matrix comes immediately to mind, as does The Dark Knight, an excellent, thought provoking superhero movie.

So, to the question is long form narrative media dead, the answer is not yet, though there is increasing competition for attention from other forms of media that require less time and cognitive effort from the audience.

iii. Notes on the future of narrative

Competition for audience attention in mainstream media points to a paradox of the narrative form. Some of the best narrative is the kind which demonstrates the power of narrative to the audience, effectively granting them this new mode of thinking and of communication, transforming them from audience into narrators themselves. There’s a self defeating aspect to this cycle, however. If everyone hears the narrative that teaches them to be a storyteller, and if everyone prefers the role of narrator to that of audience, who will be left to listen when everyone is speaking? Like-counts on social media are the brutal reminder that it is demoralizing to speak when no one is listening.

Charlie Kaufman’s film Anomalisa beautifully illustrates this point [warning: spoilers in this para]. Every person Michael Stone encounters speaks in the same empty customer service chirp and lacks their own voice and identity. They’re completely obsequious, leaving Michael’s world empty of diversity and depressing. When he finally encounters someone with a voice of her own–Lisa–it’s a miracle. Michael falls in love, but quickly his self absorption and desire to fit his experience to his existing world model robs Lisa of her voice, leaving her another empty husk. This is the lonely world in which everyone is their own narrator and no one is listening to one another.

Perhaps it’s representative of a version of a world where everyone is plugged into their own virtual reality in isolation.

Such a world would have many advantages. There would be no conflict, no war, no ridicule, no hurt feelings. Each person’s VR would be completely tuned to her specific preferences. There would be no need to compromise on which movie to see, which game to play, because each person would choose exactly what she wants.

In a way, the infinite choice of the internet gives us a glimpse of this world. Already, today, for example, we could customize our entire media diet exactly to our own preferences, without a thought to the preferences of others. But we often compromise, aligning our choices with those of friends and family, because there is something valuable in shared experience.

Do we value shared experience just because we crave audience? I don’t think so – I want to listen just as much I want to tell my own story. Shared experience is a kind of parallel processing. Listening to others tell stories of their experience is a highly efficient way to explore alternative worlds. Different actors are running their own simulations, with similar but slightly different environments, and choosing different actions. And they’re processing these in parallel to my own experience. Narrative is a way for each of us running a different simulation to sync up, report findings, and update our models. The parallelization may be more efficient than me exploring–with my single thread of consciousness–all of the possibilities. And, people I encounter may discover things I never would have found on my own, because they come from a different context and have a different set of priors than I do.

It is this desire to learn from others that governs the type of stories I seek out. I want narratives that challenge me. I love when a story changes my mind about something or helps me understand a behavior that previously seemed crazy. This goes back to my comment on empathy from part (i) – narrative is one of the best ways for making sense of human behavior.

This is why I don’t love the argument reducing all narrative with any sort of commercial interest into appropriation (If you walk in someone else’s shoes, then you’ve taken their shoes: empathy machines as appropriation machines). I understand that every author will bring their own personal and historical bias into any story they tell, but I worry arguments like this throw the baby out with the bathwater, border on a dangerous form of state or mob censorship – part of the power of fiction and drama is imagining a different reality, part of which might be imagining a protagonist with an identity different than your own. Limiting all authorship to first hand experience of one’s physical identity would rob narrative of it’s primary power.

This debate about how much the identity and context of the author contributes to the understanding of the story (if at all) or, on the other hand, whether there is “only the text” leads me to a final thought experiment: suppose you saw an incredible film, one that changed your mind on a matter of ethics. Eg, suppose you went into a film believing the death penalty was just and emerged having changed your mind on the topic. You learn later the film was authored completely by an AI algorithm.

Does authorship by a non-human agent change the meaning of the film for you? If so, is it valid to change your opinion based on this? It seems this detail about authorship should not matter, that you should hold on to the learnings you took away from the film.

Yet something nags about this thought experiment… an algorithm cannot die, so how can it speak with an authentic / valid perspective on matters of human life and death? Would the algorithm be appropriating some element of human identity? Even if you sympathize with the perspective that some sort of appropriation has occurred, would you prefer the world where this film is authored by an AI and changes the mind of citizens and legislators, who change laws, resulting in fewer human lives lost? Or, would you choose the world where the film is censored, and the death penalty is still practiced?

Rather than offer my own answers to this final series of questions, I’ll bring things full circle, back to Aristotle who said in The Poetics that drama was the imitation of the ‘real world’ through art. Reducing narrative to imitation seems to entirely miss the point, in my opinion. The real power of narrative and fiction is what it can teach us about our ability to author our subjective experience of this world. For the simulation of the ‘real’ world is just another simulation, just as malleable as the invented simulations we might conjure up in our imagination.

Perhaps no hero illustrates the power of the placebo–the ultimate hack of consciousness which occurs when we apply the creative capacity for narrative fiction to our perception of the world–quite as well as Don Quixote, who transforms his mundane, impoverished experience into an incredible adventure. Yet Quixote also reminds us of one key difference between the simulation that is our model of this world vs all other simulations. In virtual worlds, we benefit from rewinds, do-overs, and infinite lives. In the simulation of this world that is our consciousness, death is final and game ending.

Read Next

The Emperor Has No Clothes, There is No Santa Claus, and Nothing is Rocket Science