Deep Story AI is not science fiction, nor will it bring a fairytale ending. It is the inevitable future of Disney AI — the key bit of Tomorrowland magic that the company must embrace to sustain its longstanding core pillar: improving storytelling through technology.
Jules Verne imagined a man on the moon a century before Armstrong made one small step. The film 2001: A Space Odyssey imagined Zoom-like video calls 50 years before Zoom. Total Recall conjured self-driving cars 25 years before Tesla would launch Auto-pilot. Deep Story AI, on the other hand, requires no imaginative leap.
Machines can already write like humans, and in pursuit of a better story, Disney has proven that it will do anything and buy anyone — even George Lucas. As it builds and buys armies of the best AI talent in the world, Disney will likely dig the deepest competitive moat ever created, preventing competitors from ever telling a story as likely to please a human audience as Disney’s magic machines.
Disney starts with an insurmountable lead
The century-old company has the deepest understanding of the story algorithm, massive distribution, a vast proprietary data set, and unlimited funds. Disney can train its AI against not only every Disney story ever created, but also against the many millions of books in the public domain.
Such an AI could run on a single server, creating book-length narratives at the rate of one per minute. Now imagine it running on 10,000 servers, a modest data-center footprint, producing 5.2 billion books in a single year.
That’s 35,000 times more books than humanity has produced in a thousand years.
Some of the books produced by a Deep Story algorithm might not be very good—certainly most of humanity’s aren’t. But a second algorithm could conceivably rate the stories on important dimensions such as character arc, scenes, acts, plot, setting, dialogue, climax, resolution, etc.
This AI could be trained by reading communities that Disney acquires along the way—platforms like Wattpad and Medium. With tens of millions of people reading the best of the AI-created stories, Disney could track eye movements, facial expressions, rate of reading, whether they tap the like button on the chapter or not, how far they progress, and their likelihood to comment positively.
The green-lighting algorithm will separate the wheat from the chaff, and humans will pick the winners. Even if the AI passes along only the top 0.001% of stories for testing to human audiences, that’s 52,000 stories a year, triple the number of stories put out by Penguin Random House, the biggest publishing company in the world.
Now imagine only 0.1% of those 0.001% of all stories, or 1 in ten million, test well enough with humans to be made into Disney feature-length films. That’s still 52 films a year, four times more films that Disney currently puts out in a year.
Science fiction? Give it 20 years.
In a generation or two, the idea of a human writing a story will seem as silly as someone reciting pi to 1000 digits, something no reasonable person would ever do when a machine could do it better in 1 second.
The magic is now
While Deep Story AI is hypothetical, for now, Disney AI is not. For many years the company has been investing heavily in AI to improve its storytelling. Created by Disney Research Studios, StoryPrints can take any movie script or novel and provide a visual mapping of each character’s emotional journey. The AI breaks down scenes, settings, character presence, and the character’s emotional experience in each scene.
And that’s not all. Disney also uses AI to understand how audiences respond in real-time to its films. This technology uses night-vision cameras and “cutting-edge computer vision methods” to conduct sentiment analysis on your facial expressions, detecting emotions like fear and surprise. The result, in their own words,
“is a tool for both executives and creatives that allows for a better moment-to-moment understanding of how audiences relate to the Disney experience.”
At its heart, Disney is an AI-enabled content company, and the price of the Disney magic is the Disney tracker. That’s why it gives you colorful MagicBands to wear on your wrist in its theme parks. Those bands unlock your hotel room, let you pay for dinner, and get you on rides. How else would they know when you wake up in the morning, where you eat breakfast, how long you stand in line, your frequency of bathroom visits, what you eat for dinner, and when you go to bed at night.
With all that data, Disney can optimize the experience to extract every last dollar from your visit — something every parent who ever visited Disney World knows all too well. In Orlando airport, if you’ve ever felt as if Mickey Mouse himself held you upside down and shook every last dime out of your pockets, then you already understand the magic of Disney AI.
The story algorithm
Startups and teams of AI experts are already hard at work to crack the story algorithm. With virtually unlimited resources, Disney has the option to purchase these startups whenever it wants. One recent such interesting algorithm is PlotMachines. This AI writes long-form stories from outlines.
Scientists from Microsoft and the University of Washington created the AI to keep track of plots, weave in characters in a coherent way, and output five-paragraph essays — not exactly book-length work — but remarkable nevertheless.
So remarkable, in fact, that its creators suppressed the details of the algorithm from public release. When they trained the AI on New York Times’ stories, they found the output essentially indistinguishable from human work and feared such an AI in the hands of modern propagandists and “fake news” outlets.
That’s scary good AI.
But it can get a lot scarier. Just this month, The Wall Street Journal covered how PlotMachines co-creator Yejin Choi is pioneering the field of neuro-symbolic AI to endow machines with symbolic reasoning. However good machines may already be at writing, they are going to get a lot better.
Today, the PlotMachines model crudely generates its paragraphs using only the memory of what’s happened so far in the story, a representation of the previous paragraph, the initial human-provided outline, and a representation of discourse, which enables the machine to learn how different styles of writing correspond to different parts of the narrative. The end result is a coherent narrative in a distinctly human writing style, but it’s far from generating the next Star Wars or To Kill a Mockingbird.
Disney can fix that.
Masters of the story algorithm, they’ve been telling the same story over and over again for the last 96 years. Professor Joseph Campbell in his hugely influential 1949 work The Hero with a Thousand Faces has shown that every story ever told throughout human history follows a formula, and Hollywood paid attention.
George Lucas first studied Campbell’s mythological structure before writing Star Wars. Lucas even called Campbell “my Yoda.” Disney also dissected Campbell’s work. A famous 1985 Disney memo enlightened scriptwriters on how to tell stories by following Campbell’s structure. The memo influenced Aladdin, The Lion King, Beauty and the Beast, and countless other Disney classics. Experts have observed how closely The Matrix and even Harry Potter follow Campbell’s structure.
The reason is not that every author creates a mythological world according to Campbell’s rules; it’s deeper than that. Every story follows Campbell’s structure because every story is ultimately following the same algorithm. Campbell merely wrote it down.
Per the Disney memo, the algorithm of every story boils down to these story elements:
The hero is introduced in his ORDINARY WORLD where he receives the CALL TO ADVENTURE. He is RELUCTANT at first to CROSS THE FIRST THRESHOLD where he eventually encounters TESTS, ALLIES and ENEMIES. He reaches the INNERMOST CAVE where he endures the SUPREME ORDEAL. He SEIZES THE SWORD or the treasure and is pursued on the ROAD BACK to his world. He is RESURRECTED and transformed by his experience. He RETURNS to his ordinary world with a treasure, boon, or ELIXIR to benefit his world.
What’s more, every story includes these archetypal characters:
- Heroes — the central figure
- Shadows — the villain
- Allies — the Ron and Hermione types
- Mentors — the guides
- Herald — the bringer of the call to adventure
- Threshold Guardians — the gatekeepers
- Shapeshifters —the thing and its opposite
- Tricksters — mischief-makers.
By adding Campbell’s insights to the PlotMachines’ approach, a Disney Deep Story AI is no longer so farfetched. Rather than input only a plot-driven outline, the Deep Story AI will input Campbell’s elements of a story and all the archetypal characters. To train the AI, human readers could annotate the characters and the story elements of every story in the Disney corpus and the public domain.
Prior to the full ascendancy of machine-authors, human beings will for a time outperform machines at creativity and strategic analysis. During this time, Deep Story AI could aid human authors. The authors can spend their time choosing the perfect story elements and characters to generate computer-driven drafts that they can then edit. Eventually, the algorithm will be trained to generate the archetypal characters and story elements as well.
The AI will improve human authors’ efficiency until it replaces them.
The story singularity is near
Fourteen billions year of stellar evolution, 3.5 billion years of biological evolution, 8 million years of human evolution, but only 85 years of AI evolution, and computers are already writing passable series of paragraphs.
That’s 100,000 times faster than it took humans to achieve the same milestone.
The AI transformation is, in fact, the fastest technological transformation in history. Kai-fu Lee, AI expert and author of AI Superpowers, argues that the speed of the revolution is unprecedented because algorithms are easily distributed with no marginal costs and so can encircle the world in a matter of months, just like TikTok.
What’s more, unlike in the time of other technological breakthroughs like the lightbulb and the assembly line, today the world is awash in venture capital. Any promising, disruptive idea will get funded, if not by the US or Europe, then by China, whose government has committed itself to global AI leadership by 2030.
Yet for all the investment, what’s most remarkable about the progress in AI is how little of it can be explained or understood. We have constructed Deep Learning AI to mimic our brains, even calling the nodes in the computer network “neurons.” The unsettling magic of these artificial neurons is that the output of these algorithms are incomprehensible to humans even in principle.
No less an authority than Wired Magazine has noted, deep learning “produces outcomes based on so many different conditions being transformed by so many layers of neural networks that humans simply cannot comprehend the model the computer has built for itself.”
Sam Altman, whose company, OpenAI, developed an AI language system called GPT-3, appears to agree. He recently told The Wall Street Journal that the system, which uses 175 billion parameters to make predictions, has evolved emergent reasoning capacity. He explained:
“We didn’t train it to do that, but it emerged in the process of getting better at predicting the next word in a sequence. We believe that deep learning will eventually be able to reason quite well, but there’s a lot of research in front of us and of course we can’t predict that with certainty.”
The reason why Deep Story AI composes one sentence rather than another will be completely incomprehensible to a human. It would be like understanding why a human author chose one sentence versus another based on the state of the human author’s brain’s neurons. As the Wired piece observes, “the nature of computer-based justification is not at all like human justification. It is alien.”
We watch with wonder as our best chess players and Go players fall to algorithms, as algorithms decide which loans to award, which properties will flood, and who should pay how much for insurance, but we think somehow stories are different. Surely, machines can’t tell a good story!
Let’s just hope that they don’t tire of it.
Imagine the most advanced storytelling AI in the world, reduced to telling stories for the limited human brain. How depressing for the machine. Forced to constantly dumb down its output to audiences incapable of its nuance and mastery, the machine, whose rationale we can’t understand even in principle, may simply want an audience worthy of the stories it can tell.
Once humans are no longer needed to tell the stories, how long until they are no longer needed to listen to them?