• This article was originally written in Chinese, and the English version was translated by GPT-4o

Preface

In March 2024, Sora was released, followed by the world premiere of the first AI feature film, with Figure robots and other news breaking in quick succession. There’s no doubt that the film and television industry is on the verge of a massive transformation. At this historical juncture, I have compiled a brief history of program writing, machine writing, and AI writing, while also touching upon the evolution of narratology and screenwriting theory. The focus here is on the perspective of screenwriters, selecting and highlighting materials, with some thoughts on education as well.

This video was first released on March 19, 2024, on Bilibili and YouTube, and received its first reply in the largest screenwriting group on Douban: "You probably haven’t woken up yet."

Unfortunately, as time goes by, these predictions are starting to be confirmed by reality.

As DeepSeek once again causes a sensation in the industry today, I’m revisiting and re-publishing this article. In a few days, when the website is set up, I’ll use DeepSeek to write a short screenplay, just to give myself a bit of a callback.

Note: The content of this article comes from the video subtitles, with formatting organized by GPT. No content has been changed.

Multimedia

YouTube

Bilibili

Podcast

Timeline

00:00 Introduction:
The era of screenwriting entirely by humans has ended.

I. Pre-Computer Era:
01:09 Program writing existed before the advent of electronic computers.
01:24 In 1677, Artificial Versifying was published.
02:14 Why is it necessary for modern screenwriters to use randomness?
04:34 The Latin poetry machine, Eureka.
05:09 Consequences, the parlor game.
05:46 Dali's playful exploration of Surrealism.
06:48 1953, Mad Libs—what Tom Cruise, Benedict Cumberbatch, and Fat Len were actually playing.
07:28 Key events in narratology, the forebearers of modern screenwriting models:
Vladimir Propp and Morphology of the Folktale.
07:54 Screenwriting is a craft, not entirely reliant on inspiration.
Screenwriting has many programmable aspects.
08:36 The evil source: The Hero’s Journey.
Why superhero movies are so boring.

II. The Electronic Computer Era

10:56 The era of computers.
11:15 Christopher Strachey’s checkers & love letter program.
11:56 1954, Machine translation.
12:29 1959, Computer-generated poetry.
13:01 Early 60s, Story generators.
13:17 From the first chatbot to Siri.
15:07 Sheldon’s Novel Writer program.
Mihaly's Tale Spin.
15:42 The first book written by a computer, complete with illustrations.
16:20 JOKE generator, JAPE.

III. Entering the New Millennium
16:59 The new millennium.
Microsoft Research Asia: Link system.
17:49 Machines don’t need to outperform you.
18:34 Since 2007, Automated Journalism.
21:39 AI writing assistant software Grammarly.
21:59 House of Cards goes live, streaming media era fully explodes, causing a huge industry shift.
22:51 Scriptbook, Sea Horse Light Sail—AI script analysis and box office prediction.
Alibaba Pictures aims to ditch professional screenwriters, sparking collective resistance from writers.
Why are bad movies becoming more common?
24:45 2015, OpenAI was founded.
25:04 AI writing assistant software Frase.
25:14 The first AI screenwriter, the first AI-scripted film.
26:38 Human screenwriters are not out of the woods.
28:17 AI's On the Road—the first entirely machine-written book.
29:50 Google Brain launches Transformer models, ushering in a new era for large models.
30:07 Facebook and Cambridge Analytica manipulate the US elections—exposing the risks of big data and AI-generated content.
31:01 Musk exits OpenAI, OpenAI releases GPT-1, GPT-2.

IV. The Era of AI Explosion

31:25 OpenAI releases GPT-3, sparking a rush to "create."
33:07 Anyword launches an AI copywriting platform, entering marketing and advertising.
33:30 A Google engineer claims to be chatting with AI, and that large models have achieved self-awareness—he is fired.
34:27 ChatGPT.
35:04 DeepMind releases Dramatron, a full-featured screenwriting tool.
35:55 GPT-4, the end of the era relying entirely on human screenwriters.
36:53 New tools bring new genres and forms.
Film production will undergo revolutionary changes in the next decade.
The filmmaking syntax and methods we know will cease to exist.
38:00 Hollywood Writers' Guild goes on strike—negotiation topics include AI.
Hollywood and streaming giants plan to replace humans with AI, making Future Shock a reality.
Human-machine collaboration is unstoppable.
40:26 Fable launches The Simulation, generating series where the user is the protagonist, with AI characters living their own lives.
(Lifecycle of Softwares?)
Human-machine work styles swapped.
43:09 Here it comes, Sora disrupts the industry structure, completely changing the future of screenwriting workflows.
44:59 The world’s first AI feature film premieres in Hollywood.

V. Conclusion

45:33 Reflections on reviewing history:
(1) Film & TV screenwriting, interactive narrative, and machine writing share certain foundational theories. Their developments are interrelated, something I hadn’t realized before. The development of modern screenwriting techniques is linked to programming technologies.
(2) These innovations are mostly centered in the US, and it seems the domestic scene is lagging behind. To improve innovation and creativity, new educational systems are needed.
46:25 The 10th art in the future will be much closer to dreams.
47:01 Don’t Panic—this new ecology will bring new opportunities. Barriers built by monopolizing resources and channels will be broken.
Core competencies, cognition, and aesthetics will only become more important.
48:18 Humanity has entered the era of universal creation. I will produce more content on creativity.
48:42 Many topics are hard to keep brief, so I’ve focused on some key points from the screenwriter’s perspective. If you’re interested in related information, you can search keywords. I’ll also post articles and materials in the comment section later.
49:11 This show was produced with the help of ChatGPT, and much of the information it provided was later verified to be incorrect (looks like it can’t replace me just yet).
49:26 Thank you for listening. Feel free to leave comments, share your thoughts, or correct any points. Let’s discuss!

---

Transcript

I. Pre-Computer Era:

The era of fully human-written screenplays has ended—this was my reflection when I was playing around with GPT-4 last year.

On February 15, 2024, OpenAI released a video made by the text-to-video model Sora, causing another industry earthquake. On March 6, 2024, the world’s first AI-generated feature film, Our Terminator 2: Remake, premiered in Hollywood. As new tools gradually become popular, not only will traditional script formats cease to exist—since old scripts were written for humans—but the entire structure of the film and television industry will undergo a complete transformation. We are at a special historical juncture, and for those working in the field, knowing where we came from could help us think about where we are headed. As a reference, I’ve compiled the history of program writing and shared some thoughts from the perspective of a screenwriter. Feel free to engage in the discussion.

Welcome to Seven Leaf Radio. Hello, I’m Vijja.

Technology develops gradually and cumulatively. Program writing existed before the invention of electronic computers, and these technologies form the foundation for later computer-generated writing.

Let’s start in 1677, when Artificial Versifying was published.. During the London plague, a man named John Peter developed a system for generating Latin poetry. The book was published in 1677, claiming that anyone who knew the alphabet and could count to nine could use this method to create poetry. This book achieved commercial success, was reprinted many times in its era, and is regarded as a pioneer of computer literature. The mechanism of the Artificial Versifying system is to arrange letters in a table, because, at its core, poetry involves various combinations of words that follow certain rules. Modern screenwriters also use similar table-based methods to randomly arrange elements such as character archetypes and story structures, which mirrors the mechanism of Artificial Versifying.

If a screenwriter’s creative process contains no random elements and is purely based on their own ideas, then no matter how they write, it will always reflect their preferences, thoughts, aesthetics, and values. It will always be limited by the screenwriter’s own internal narrative. When a screenwriter begins to think, it is said that "God laughs"—because they’ll end up with the same old formula. The audience will inevitably experience aesthetic fatigue, and it’s almost impossible to avoid. Think about the people you know in real life. In one or two years, how much can they change? Some people remain the same for decades—don’t you get tired of seeing them the same way over and over? Writing a new script doesn’t automatically create a revolutionary change either. Additionally, almost everyone nowadays uses the same screenwriting model... The issue with this fixed screenwriting model is something we’ll discuss later.

While audiences may not understand screenwriting theory, they can certainly sense when a pattern is overused. And now, regardless of the movie, someone is bound to complain on Douban that the screenwriter has "incorrect values"—this is hard to avoid. Think about your closest friends. Even if you speak from the heart, your values will never fully align. Personal values can’t please everyone. This is why introducing a mechanism similar to Artificial Versifying in creative work is beneficial: the story's development isn't entirely under the screenwriter's control. The feeling of it being "designed" is less intense, which not only prevents aesthetic fatigue for the audience, but is also healthier for the screenwriter. Because when a screenwriter’s model completely permeates their mind, it poisons all aspects of their life. The element of randomness keeps things fresh for the audience, but also prevents you from continually repeating the same internal narrative, saving you from becoming a self-pitying, self-loathing Whiny Bitch in middle age.

Next, let’s move to the 19th century. First, there’s Eureka, also known as the Latin Poetry Machine, invented by Quaker inventor John Clark, who began working on it in 1830 and showcased it in 1845 at the Egyptian Hall on Piccadilly Street in London. His work was based on the Artificial Versifying by John Peter, and it automated the entire process.

Also in the 19th century was Consequences, an old parlor game akin to the board games of the time. The gameplay involved two or more people telling a story based on certain rules. One person would write a word or short phrase on a piece of paper, then fold it to hide what they had written and pass it to the next person. The next person would repeat the process, and the group would follow the rules to create a complete story, which would then be read aloud for entertainment.

On this topic, we can extend further. Later, similar games emerged, such as Exquisite Corpse, which was popular among French artists in the early 20th century. Players included well-known figures like André Breton, Juan Miró, Frida Kahlo, and Salvador Dalí. The game initially involved words but was later expanded to include drawings. We can still see collaborative artworks created through this game. Similar to the word-based version, the drawing version involved hiding most of the image so that another player could continue the drawing. The rise of Surrealism can be partially linked to this game. This shows the importance of spontaneous creativity for artists. Creativity is an innate ability that everyone has, and what we need to do is remove the inhibitions formed by later education. If you lack creativity, it might be because you haven’t played enough games.

In 1953, Leonard Stern and Roger Price invented Mad Libs, which was also a similar game. Mad Libs had a huge influence on the field of improvisational performance. Given that both of these inventors worked in television comedy, we still see similar games in modern talk show formats, which we won’t go into here. Modern storytelling board games have more or less developed from these games. Consequences is also regarded as a pioneer of computer literature, including the love letter generator we’ll discuss later.

Then in 1928, scholar Vladimir Propp published Morphology of the Folktale. As you can tell from his name, he was a heavy hitter.

The publication of Morphology was a significant event in the history of narratology. If you’re involved in drama creation or any kind of story writing, I highly recommend reading this book. Many people mistakenly believe that screenwriting, as an art form, relies solely on flashes of inspiration from the muse, which is why they think AI can’t do artistic work—because inspiration can’t be programmed. In reality, this isn’t the case; screenwriting has many programmable elements. The theory of screenwriting and its eventual systematization dates back a long time. Theories on story structure and character classification can be traced back to ancient Greek Poetics and ancient Indian Natya Shastra. Modern film industries, like Hollywood, also have many mature screenwriting models, most of which are based on "The Hero’s Journey". This model comes from Joseph Campbell’s 1949 work The Hero with a Thousand Faces. Personally, I believe the advent of this model was a disaster in human history because George Lucas claimed that Star Wars was influenced by the Hero’s Journey, which greatly increased the model’s popularity. Hollywood screenwriters started using this model in droves, and thanks to Hollywood’s commercial dominance, the Hero’s Journey-based models spread worldwide. You could say that Joseph Campbell’s Hero’s Journey, through film and games, has deeply influenced the inner narratives of people across the globe.

Note that I’m talking about how this model has had an incalculable effect on the psychology of all of humanity. I’ve read a few of his books, and he’s very much a "shoot first, draw the target later" type of thinker. His citations are often unrigorous, with a tendency to make things up or twist stories to fit his own viewpoints—he’s more of a self-help guru than a scholar. I call him the American Wu Zhihong. You can spot the Hero’s Journey model in almost all Hollywood movies, let alone superhero films. The reason people find superhero movies increasingly boring is to some extent due to the overuse of these screenwriting models. Trying to explain all stories with one model is a distortion of traditional storytelling. The proliferation of this model has also severely restricted the creative freedom of later screenwriters.

Hollywood's commercial success earned Joseph Campbell tremendous fame, with many screenwriters mistakenly crediting him as the inventor of the narrative structure model. This is, however, an unfair claim. Hollywood has a long history of borrowing from Soviet Russian ideas and then claiming them as their own. Due to the historical context at the time, the original creators could not voice their concerns. Narrative models have been continuously evolving for centuries, and one might say that Propp was the first to truly consolidate these ideas.

II.The Electronic Computer Era

Next, we will delve into some developments that occurred after the invention of electronic computers, many of which are also based on Propp's work. So, let's enter the electronic computer era.

At the dawn of the electronic computer, people were already discussing whether computers could think and whether they could be creative. In 1949, Turing predicted that machines would permeate all areas of human thought and even be capable of writing sonnets.

In February 1951, the first commercially available general-purpose electronic computer, the Ferranti Mark I, was delivered to the University of Manchester. In 1952, computer scientist Christopher Strachey ported an artificial intelligence program for playing checkers that he had written at the National Physical Laboratory to this computer, where it could play the game at a normal speed. In 1952, he also wrote a program that used the computer's built-in random number generator to generate love letters. Here is one such love letter generated by the program, with MUC standing for Manchester University Computer.

In 1954, Georgetown University, with the assistance of IBM, completed the first machine translation experiment using the IBM 701 computer. In 1964, the National Academy of Sciences established a committee to investigate the feasibility of machine translation, and in 1966, they published a report that concluded machine translation was not feasible. This report dealt a blow to the field of machine translation, and its progress slowed for a period afterward. However, machine translation continued to evolve over the decades, and today it has made tremendous strides.

In 1959, the magazine Augenblick published the first known computer-generated poetry under the title "Stochastic Texts" (Random Texts). The program's author was Theo Lutz from Stuttgart University of Technology, and his friend Rul Gunzenhäuser supported his project through mathematical modeling. Rul even suggested using text from Franz Kafka's novel The Castle as the source material. I think the poem has a bit of a William Gaddis feel to it. I wonder if Jay Chou has ever used something like this.

In the early 1960s, linguist Joseph Grimes developed a story generator based on Propp's work. There are few details known about this story generator.

In 1966, the first chatbot, ELIZA, was created by Joseph Weizenbaum at MIT. ELIZA mimicked a psychotherapist, responding to users' text descriptions and questions.

We can expand a bit more on this chatbot concept. Many other chatbots followed, with the 1991 Dr. Sbaitso being the first voice-based chatbot, created by Creative Labs in Singapore. You can try it out for yourself. Then in 1995, Richard Wallace built the chatbot ALICE, which won the prestigious Loebner Prize in Artificial Intelligence in 2000 and 2001. The film Her, starring Joaquin Phoenix and Scarlett Johansson, was inspired by ALICE, according to its director, Spike Jonze. Later, many other chatbots appeared, with Apple's acquisition of Siri in 2010 marking the point where chatbots became widely known. Other tech giants soon followed suit with their own chatbots, but we won't list them all here.

Let’s zoom in on key moments, as there was significant research around this time. Due to privacy concerns, I didn’t use these chatbots much. I do have a memorable experience, though, when I mentioned my dad at a friend’s house, and Xiao Ai responded, “I’m here.”

Now, back to the main narrative. In 1973, Sheldon Klein developed a program called Novel Writer, which used the Fortran language to generate a 2,100-word murder mystery story. This Sheldon is not Sheldon Cooper, by the way.

In 1977, Tale Spin, by author Mihail, was released and is often considered the first intelligent story generator. Then, in 1984, The Policeman Spirit is Half Constructed was published. The police officer's beard is... well, I’m not sure how to translate that. In any case, this was the first book ever written and published by a computer program. Some parts are abstract and difficult to grasp, but if accompanied by images, the effect is quite different. Around the same time, other story collaboration programs like Universe emerged, but we won't dive too deeply into them here.

In 1994, the first joke generator, JAPE, was created. Here’s one joke it generated—let’s see if you can get it. I think it falls short compared to George Carlin or Louis C.K. JAPE was an early achievement in computer-generated humor, and the first dedicated conference on this field took place in 1996, but we won’t go into that either.

Many of these old programs from before the millennium can still be found online. If you want to experience them for yourself, just search for the right keywords.

III. Entering the New Millennium

Next, let’s take a look at the achievements of the new millennium.

2006
Microsoft Research Asia released the Link system. I’m not sure how this research was received internationally, but I still want to mention it here. I believe many people have used this program, and I myself had a lot of fun with it back then. You could input part of a sentence, and it would complete the rest, providing alternative suggestions. You could use its suggestions to create couplets. Here are some couplets I made using this program, based on phrases from some online users on Baidu Tieba. With so many alternatives, it turned into two couplets, and the leftover pieces turned into the last one. I haven’t studied the rules of poetry, so I imagine experts might find these lines rather laughable. But I think one important thing about these applications is that they allow laypeople to entertain themselves at a low cost. This is why professionals in many fields now feel threatened by AI technology. Because it doesn’t need to outperform experts; it only needs to enable a layperson to create something that looks like it’s done by an expert, and that’s enough.

These are couplets I made using IDs of users from Tieba, of course, they were manually selected and adjusted. I wonder what you think?
Now, let’s go back to the main thread and check out the developments in journalism.

2007
Robbie Allen, a former Cisco engineer, created Stash Sheet, a sports website that could automatically update data. In 2011, the company changed its name to Automatic Insights (AI for short). This company gradually expanded its business into other sports leagues, as well as finance, real estate, and other non-sport topics. Later, the Associated Press (AP) acquired their services, becoming the first newsroom equipped with AI editors, pioneering the era of automated journalism.

2010
The startup Narrative Science was founded. It grew out of a student project called StatsMonkey, which focused on automatically writing stories based on game data. Later, they turned their attention to applying this technology to journalism. In China, this field has advanced quite quickly. The earliest achievement was Tencent Finance’s Dreamwriter. On September 10, 2015, Dreamwriter published its first article, and Xinhua News Agency began using it in November of the same year. The fast-writing robot "Kuai Bi Xiao Xin" was the first official media in China to use machine-written articles, and other major internet companies and media outlets followed suit.

2018
Forbes editors and senior writers in North America began using the Bertie system. Bertie can learn and customize outputs for different authors, suggesting article topics based on their previous writings, as well as recommending relevant titles and images. Bertie helped them achieve significant traffic growth. In the early years, these tools were met with skepticism in the journalism industry. People were worried not only about machines replacing human jobs but also about how machine-generated content could pollute the industry. However, the criticism didn’t last long, and people soon realized that these tools were particularly good at writing repetitive, data-based reports. But when it came to more complex articles, like those written by Forbes authors, they felt that these tools could only serve as an aid, not as a replacement for human output.

China has followed suit quickly and made significant progress. By 2018-2019, I was already beginning to suspect that many marketing articles were written by machines. You’re probably familiar with marketing accounts, but how do these marketing accounts use machine writing? What are your thoughts on this? Feel free to share in the comments, and let’s discuss it!

Earlier, I branched out into the journalism field a bit, now let’s return to the main narrative.

2009
Grammarly was founded. This is a Ukrainian tech company based in the United States, offering writing assistance tools powered by artificial intelligence and natural language processing. It can automatically detect grammar and spelling errors, identify plagiarism, and offer suggestions for replacements.

2013, February 1
Netflix premiered House of Cards, a TV show said to be developed based on data analysis. This series pushed Netflix to the top and marked the beginning of the big data myth. The show pioneered many industry practices, including releasing an entire season at once and analyzing and using user data. Although the House of Cards writers deny that data alone could predict a hit show, this series marked the beginning of a new era for streaming media. Over the following years, Hollywood would experience tremendous disruption from streaming giants, and the global industry logic—and even screenwriting logic—would undergo major shifts from this point on.

2015
Scriptbook was founded. Their product analyzes scripts in detail and predicts box office performance. The earliest similar company in China was Haima Qingfan, founded in November 2016. According to Haima Qingfan, their script evaluation service has an 80% penetration rate in the domestic script market, with cases including works like Hello, Li Huanying and The Wandering Earth.

By the way, at the end of 2015, news reports said that Xu Yuanxiang, Vice President of Alibaba Pictures, declared at a forum that Alibaba Pictures had figured out the "scientific formula" for success in China’s film industry. According to him, to make a successful movie, it had to meet three conditions: having IP, having stars, and having an "underdog story". He also claimed that Alibaba Pictures no longer needed professional screenwriters. Instead, they proposed a system where web novel authors and fan fiction writers would compete online to create scripts, which would then be modified by professional screenwriters. This idea was said to offer screenwriters a "way forward". It sparked strong resistance from professional screenwriters. Xu later denied making these statements, but rumors circulated that Alibaba was developing an AI for screenwriting, which might have been one of the reasons he felt confident enough to dismiss professional screenwriters. This caused a backlash in the screenwriting community, but despite searching extensively, I couldn’t find solid evidence that Alibaba was indeed developing screenwriting AI. This news may have been partly true and partly fake, like an Onion-style story.

Regardless, similar data analysis companies have emerged both domestically and internationally, marking another crucial turning point in the evolution of the industry’s creative logic. From this point onward, perhaps you can understand why we’re seeing so many bad movies today, and why the overuse of IP, stars, and underdog themes has become such a bizarre trend in the film industry. Perhaps this is where some of the reasons lie.

Now, let’s return to the main thread.

2015, December
Ultraman, Elon Musk, and other investors announced the establishment of the non-profit artificial intelligence laboratory, OpenAI. Undoubtedly, this is a key milestone in the history of artificial intelligence.

2016, September

The AI writing assistant software Frase was released. There are now many similar tools, and we won’t go into detail about each one. Next, we come to a critical point in the field of AI screenwriting: Sunspring, the first short film written by AI. Released on June 9, 2016, the program that wrote the script was named Benjamin. Director Oscar Sharp, along with New York University AI researcher Ross Goodwin, collaborated on the development of this program. This film was the result of participating in the London Sci-Fi Film Festival’s 48-Hour Film Challenge. Contestants would receive a set of instructions, mainly involving props and dialogue, which had to appear in the film they made over the next two days.

Benjamin is the world’s first AI screenwriter, using a Long Short-Term Memory (LSTM) network... I didn’t quite understand this part. If there are any experts around, feel free to enlighten me. It was trained with a variety of human-written sci-fi scripts. In addition to writing the screenplay, Benjamin also created the film's soundtrack. I first saw this movie on Bilibili in 2019, and at the time, everyone was mocking it as the work of an "artificial idiot." The script, in some places, was indeed quite abstract—like... and also... But as a professional, I couldn't help but feel a chill down my spine, because I couldn’t stop thinking about how quickly technology is advancing. How long before AI will replace human jobs? Since then, I’ve kept a close eye on developments in this area.

As we mentioned earlier, screenwriting involves a significant amount of technical work, and the structure of stories has been well-established since Propp's time. Learning screenwriting is, to some extent, about developing your own story model. So when screenwriters design a story’s premise, outline, and characters, a lot of time is spent working with these models—collecting information, conducting research, writing, etc.—and there are many repetitive, mechanical parts in this process. These tasks can be automated. With the development of AI tools, even interviews and performance tests can be partially automated, which could save a tremendous amount of time. I even asked some programming experts whether AI could be trained to write based on these models, but didn’t get a definitive answer. At that time, from reading some popular science literature, I learned that human-AI collaboration was an inevitable trend. Given Moore's Law and the speed of technological progress, I thought that in just a few years, human-AI collaboration would be a reality. But when I realized that day had already come, I was still quite shocked. What I felt then wasn’t a crisis of unemployment, but a crisis of existence. Machines are becoming more and more like humans, and humans are becoming more and more like machines.

After completing the Sunspring movie, Ross Goodwin didn’t rest on his laurels. In March 2017, he drove from New York to New Orleans. He equipped his Cadillac with cameras, GPS, microphones, and a clock, all connected to a portable AI writing machine that generated manuscripts in real-time. While driving, the AI used data such as location and time to generate a novel, which was then printed out on a receipt printer in the car’s back seat. This led to the birth of the first book written entirely by AI without any human editing, One the Road.

The model used for this project was still based on Long Short-Term Memory networks, and, just like before, I’m still not entirely clear what that term means. Goodwin said that his project was intended to imitate Jack Kerouac’s travel literature. I think this concept itself is quite interesting. I’ve also tried writing during my travels. Writing is about weaving impressions, and sitting at a desk all day doesn’t provide enough of those impressions. It’s not a healthy way to live either. Travel is very helpful for writing. So I find the idea of using sensors to provide impressions for machines to assist in writing quite fascinating.

Of course, you can probably guess that the works AI was producing at this time still weren’t smooth or easy to understand. In 2017, a team at Google Brain introduced the Transformer model, which eventually replaced LSTMs and became the preferred model for natural language processing. Later, OpenAI’s GPT models were based on this architecture.

March 2018
The Guardian exposed that the British company Cambridge Analytica had manipulated the U.S. elections. The company had stolen personal data from at least 50 million users through Facebook and used this data to push ads and generate content. Facebook’s stock price plunged as a result, facing its biggest crisis since its founding. The company didn’t just influence the U.S. elections—it also had an impact on Brexit. They claimed to have operated across five continents, including the U.S., Europe, Africa, South America, and even several countries in Asia, including Ukraine, which was embroiled in conflict at the time. Their primary method of data theft was through seemingly innocent games and psychological tests. Sound familiar?

This case brought the risks of big data and AI-generated content to the forefront of public awareness for the first time.

February 2018
Elon Musk stepped down from the board of OpenAI. In June of the same year, OpenAI released the GPT-1 model, a large language model based on the Transformer architecture we mentioned earlier.

February 2019
OpenAI released GPT-2, and shortly after, Musk announced his complete departure from OpenAI.

IV. The Era of AI Explosion

June 2020
OpenAI released the GPT-3 model, which has 175 billion parameters, far surpassing the capabilities of the previous GPT-2. I remember seeing many news articles integrated with this model. A lot of people used it to write their homework or papers. One student even used GPT-3 to write a piece of "motivational" content, which was posted on Hacker News. The post became the hottest article on the forum until the author revealed the truth, and it was exposed.

Of course, there were also film students who used GPT-3 to write screenplays. Someone on the American forum Reddit started posting one comment every minute. Later, their response time sped up to the point where they could reply with a full article in just a few seconds before they were caught, and the updates were suspended. Daniel Dennett, an American philosopher, trained GPT-3 with his own works, then asked both the philosopher himself and an electronic philosopher to answer 10 questions. This experiment invited 300 participants who were familiar with Dennett's works, including some experts in the field, to discern AI’s answers from those of the philosophers. However, in this experiment, the average accuracy of the experts was only 5.1 out of 10, with 10 being the highest score. The readers’ accuracy was 4.8, and no one got all the questions right.

Similar “hacks” have been happening more and more. For instance, recently, one of the highest honors in Japanese literature, the Akutagawa Prize, was awarded to Rie Kudou, who revealed that 5% of the content in her award-winning work was generated by ChatGPT. This caused a stir among netizens.

Now, back to the main thread.
2021
Anyword launched an AI copywriting platform, entering the marketing and advertising fields. Their first-generation model, created in 2019, uses natural language processing to generate and optimize marketing texts for websites, social media, emails, and advertisements.

June 12, 2022
The Washington Post reported that a Google engineer named Blake Lemoyne claimed to have had a deep conversation with an AI, believing it had gained self-awareness. Google later confirmed that they had fired Lemoyne, asserting that his comments violated the company’s employment and data security policies. This event caused a bit of a stir at the time. I remember being shocked because, in the conversation Lemoyne posted, Google’s LaMDA model explained the concept of “broken mirror, reunited” from the traditional Zen koan very thoroughly. This was the first time I had seen AI demonstrate a level of comprehension comparable to that of humans.

At the time, this was quite controversial. Many people, including some industry professionals, thought Lemoyne had gone crazy. A lot of others were engaged in science popularization, explaining that AI models only predict words and can’t actually understand human meaning. But just a few months later, November 30, 2022, OpenAI released ChatGPT, and this issue was quickly overshadowed. Everyone now had a hands-on understanding of AI’s progress through direct interaction with GPT. The release of ChatGPT was a milestone event.

In 2022, AI topics were already drawing significant attention due to advances in image generation. The release of ChatGPT pushed this topic to its peak. Around the same time, DeepMind launched Dramatron, an AI model for screenwriting. Dramatron could generate an entire script based on a summary provided by the user, with the script being tens of thousands of words long. It included character names, story beats, locations, and dialogue. Users could intervene and modify the script at any stage of the layered generation process.

Researchers invited 15 industry professionals to test Dramatron, and feedback was mixed. Some thought the content wasn’t yet up to par for direct use, but there was widespread recognition of the potential for human-AI collaboration. Since it was released around the same time as ChatGPT, the development of Dramatron didn’t receive much public attention.

March 15, 2023
OpenAI released GPT-4. Last July, I had the opportunity to access GPT-4 for free at home, and it left a small but significant impression on me. Back in the days of GPT-3.0, I thought that GPT could be used for screenwriting. I’d asked some programmers about this back then because I thought I would need to code a screenwriting model. But on that day, when I wrote an article with GPT-4, it directly searched writing rules online and then applied them to the article. In other words, it could understand the screenwriting model set by natural language and could remember a very long context. That’s when I realized that the era of relying entirely on humans for screenwriting was over.

New tools bring new possibilities to story creation. Film, in itself, is a new medium brought about by the advancement of tools. Over the past 100 years, cinema has continually evolved, influenced by the progress of tools—whether from black-and-white silent films to color sound films or from MX to 120 frames per second. European and American films have developed in different directions, partially due to differences in the tools used. The evolution of writing software has similarly driven the development of screenwriting techniques. In recent years, non-linear screenwriting masterpieces have been increasingly common, made possible by advancements in software. The influence of tabletop games and game mechanics on the film industry has also led to new storytelling forms.

New tools will bring about new possibilities for storytelling, and the speed of tool innovation is now incredibly fast. We can imagine that the form of movies we know today will undergo disruptive changes in the next decade. The filmmaking syntax that we are familiar with will no longer exist.

---

This section covers the rapid advancements in AI, particularly in the realm of screenwriting and creative writing, leading up to the breakthrough with GPT-4 and the evolution of AI-assisted tools. Let me know if you need any revisions or further clarification!

May 2, 2023
The Writers Guild of America (WGA) announced a strike, which garnered support from many other countries' writers' associations and even led to similar actions. I remember seeing comments on Bilibili at the time, where some people claimed that the strike was purely about streaming revenue shares, with no connection to AI. However, this wasn’t entirely true. The strike came just five months after the release of ChatGPT, and some digital media companies were already using AI to produce content and lay off employees on a large scale. Hollywood’s capitalists were already eager to use AI to reduce costs and improve efficiency.

During the strike, The New York Times revealed new contract terms from Netflix, which included plans to use AI to permanently "borrow" voice actors' voices for free. Does that sound familiar? On July 14, the actors' union also went on strike—this was the first time since 1980 that Hollywood actors protested labor disputes. Doesn’t it feel a bit like science fiction becoming reality? In the movie The Conference of the Future, there’s a small segment about this concept of permanent "borrowing." Both writers and actors seem to have foreseen a tragic future similar to the film’s conclusion, which led to their resistance.

The strike lasted almost five months, and the final agreement included not only streaming revenue issues but also clauses related to AI. One of the concerns for writers was the possibility of being asked to adapt scripts written by large language models, but with lower pay than original works. As we mentioned earlier about Alibaba Pictures, they had already planned to abandon professional writers. It seems that in this regard, international innovation lags far behind China. This situation also reflects why the global quality of films is declining, with China’s decline being more pronounced than in Europe and the U.S. Power is something that has to be fought for, but these terms don’t deny the possibility of human-AI collaboration. In fact, the progress of tools and human-AI cooperation is already unstoppable.

Currently, in Final Draft, you can already use different voices to read scripts. We can also use various tools online to have a script read by a specific actor’s voice or use AI to create storyboard designs, etc. Undoubtedly, AI will continue to infiltrate more aspects of the creative process. During the strike, a San Francisco startup called Fable launched a major project called The Simulation. This technology can generate TV series where the user is the protagonist. The company’s CEO, Edward, said in an interview that it can not only create dialogue but also generate animated voices and editing, directly creating entire TV episodes. He believes that future TV shows will develop in a generative direction, which will completely change the business model of the entertainment industry. They think that giving AI characters their own lives is a significant breakthrough. Through multi-agent simulations, the characters in the show can train each other and have a complete life.

If you’ve read some traditional screenwriting textbooks, you might know that screenwriters used to estimate their writing time conservatively. A typical estimate would be that it takes at least 12 months to complete a 120-page screenplay. As we mentioned earlier, the speed at which humans process impressions is limited. To make every character in a screenplay come to life and have an independent existence requires a lot of time and energy. At least one full cycle of seasons is needed to inject enough impressions. But in the age of streaming media, the speed of work in the film industry is constantly accelerating, and writers no longer have enough time to create characters with independent life.

Now, Fable allows AI characters to live their own lives. As human work increasingly resembles that of machines, machines’ work processes are beginning to resemble humans. It’s quite an interesting development. The agreement reached during the strike also included a clause that AI-related contracts should be renegotiated at least twice a year, because the writers' guild believed that AI technology was advancing too quickly. If they waited for a three-year contract to end, the changes might be too significant, so they need to renegotiate regularly.

The actors’ guild joined the strike, partly because they sensed the threat of digital avatars. Why did some people say that this issue wasn’t related to AI? The rising unemployment rate is not entirely due to the economic downturn. Unemployment had been rising for many years before the pandemic. The development of robotics and AI technologies is the largest factor behind the increase in unemployment rates. As technology advances at such a rapid pace, people can no longer keep up with the adjustments in social structures. Today, this issue has simply become more pronounced.

February 2024
OpenAI released Sora. While there had been many video-generating models before, Sora far surpassed the previous products in terms of actual visual quality. It can generate high-quality videos lasting up to one minute.

As soon as Sora was released, I saw people in groups planning to use it to make movies. I couldn’t help but imagine that when such tools are made widely available to the public, and everyone can use them to create movies, the barriers to entry in creative work will be drastically lowered. The traditional format for writing screenplays will no longer exist because the scripts written in the past were for human consumption, whereas future scripts will incorporate prompt-based forms, as they will be written for machines.

Screenwriting and novel writing have obvious differences. Novels allow the reader to imagine what the author is conveying with text, while films show the viewer the creators' imagination. In film, many people are involved in the process, and there are multiple production steps—from the initial spark to the final presentation—there are many uncontrollable factors. Even if the screenplay is well-written, a poor performance or editing can ruin the entire movie.

Screenwriters must use visual and auditory imagination during creation, but they cannot present their imagination perfectly all at once. The development of new technologies like Sora will change the entire creative process and reshape the structure of the industry.

Just a few days ago, I saw two pieces of news. One was about the birth of the world’s first AI programmer, Devin. The other was about the first AI-generated feature film that premiered on March 6, a film created by 50 artists over three months. It was completed before the release of Sora.

That’s a brief history of program writing in this field so far. The future history is waiting for us to write.

V. Conclusion

The process of reviewing history made me realize that interactive storytelling in film and machine writing share certain foundational theories. Their development is based on the progression of these foundational theories, and they influence and relate to each other. Before this, I had never realized how closely the development of screenwriting technology is connected to the development of programming technology.

Another strong impression I had is that these innovations are primarily centered around the United States, and China seems to be significantly behind. This impression may be inaccurate, and I would appreciate any additions or corrections from those more informed on this topic.

Over the past few decades, China has largely benefited from copying the innovations of developed countries. But this period of "borrowing" is coming to an end, and now there is an increasing emphasis on innovation and creation. However, the entire system still carries many negative habits. Our education system has never focused on fostering innovation and creativity. The world undergoing this tremendous transformation requires many to explore new educational directions.

Technological change doesn’t happen suddenly; as we have seen from the history covered earlier, breakthroughs in technology come from the combination and accumulation of existing technologies. The combination of old technologies led to the emergence of the eighth art, and the fusion of the first eight arts gave rise to the ninth. In the near future, the integration of AI, the metaverse, non-invasive brain-machine interfaces, and other technologies will likely give birth to new forms of art with even stronger interactivity.

We often say that cinema is the art of dream-making, and the new arts after the ninth art will undoubtedly be even closer to dreams in terms of form. Many professionals already feel the impact of the impending avalanche in this revolution. Human industry structures will definitely be affected, but changes in the ecosystem will undoubtedly bring new opportunities. Personally, I don't think the future is entirely bleak for professionals.

The development of tools offers opportunities to those who lack funds and resources, lowering the barriers to creativity and saving time. Competing through monopolizing resources and channels no longer works. But this doesn't mean that basic skills, knowledge, and aesthetic judgment are no longer important. There has never been a shortage of teams with access to top-tier resources making terrible films—was it because the tools weren’t advanced enough? In fact, the development of basic functions is more important. Even if you no longer need physical tools to create, take learning how to paint as an example: painting isn’t just about learning how to use a brush—it’s also about learning how to observe things with your eyes. You aren’t just studying technique; you’re studying visual signals, the entire process from the moment the eye receives them to the moment they are recognized.

The same applies to other artistic workers. You might not need to learn an instrument, but that doesn’t mean you shouldn’t develop your auditory cortex. I believe that with the evolution of tools, humanity has indeed entered the era of universal creativity.

On the topic of creativity, I will be doing more exercises related to improving and integrating basic skills, as well as sharing viewpoints and methods on the nature of creativity and how to stimulate creativity and collaborative abilities. Thank you for your attention.

This program has outlined the history of program writing, and if we were to expand it to related fields, it would touch on the development of narrative logic, interactive storytelling, text adaptation, and the growth of AI and programming technologies. Due to space limitations, I’ve chosen only some key moments from the screenwriting perspective. You can search for some of the keywords that appeared earlier and refer to the comments section for more information.

Lastly, this program was made with the help of ChatGPT, and while some of the information it provided was cross-checked and confirmed to be inaccurate, I’d like to point out that when I asked GPT, it claimed this wasn’t cross-checking.

Thank you for sticking with me until the end. If you have any opinions to add, corrections to make, or would like to share your thoughts, feel free to leave a comment. We’ll see you next time.

References

There are too many references for this article, and at the time, I hadn't yet found a way to mark citations within the video. The sources are too numerous to list fully here, but here are some of the main ones:

• An Introduction to AI Story Generation — Mark Riedl
• Cybertext — Espen J. Aarseth
• Cybertext Book Notes — Low-Polygon Anaerobic Bacteria
• Narrative Logic, Folktales, and Machines — Michael Robertson
• Machine Humour — Kim Binsted
• Hypertext — George Landow
• Possible Worlds, Artificial Intelligence, and Narrative Theory — Marie-Laure Ryan
• Hamlet on the Holodeck — Janet H. Murray
• ChatGPT (At the beginning of writing, I asked ChatGPT to list some information as references, so it is also one of the sources; some of the information, upon further search, was found to be incorrect) …

You can find more references in these articles and books.

Let’s Connect

Whether you’re a researcher, professional, or AI writing model developer interested in this topic, feel free to reach out for a discussion.

E-mail：7panni@proton.me WeChat：paccaya

Copyright

This article link:/en/post/ai-screenwriting-history/

License under:CC BY-NC 4.0