Can AI crack the formula for writing a Hollywood hit?

Bloomberg Opinion — Ten years ago, I co-wrote and sold a comedy script to 20th Century Fox.

It was called “The Lose” and the plot was “The Fugitive combined with Harold and Kumar in Southeast Asia”.

Fox ended up shelving the project, but I always cherished the experience.

Fast forward to May 2022. I wrote about “The Lose” in my newsletter and got a response from Yves Bergquist, the CEO of an artificial intelligence (AI) company called Corto AI.

Bergquist’s company developed a tool that analyzes scripts and provides insights into how the content will resonate with different audiences.

It was too late to save my script, which was a lot of fun, but I needed to know why it failed (besides the fact that most chosen scripts never get done).

I sent Bergquist my very funny script and Corto put it through a multi-step process:

Intake and analysis: Corto analyzes and labels script variables such as narrative types, emotional tones, character arcs, themes, and much more. Having defined the “narrative DNA” of my script, Corto compares it to a database that, according to Bergquist, has more than 700,000 television and film titles.

Generate a list of comparisons: Short identifies the best matches based on “narrative DNA.” (He was glad to find the classic Southeast Asian comedy “The Hangover II” among them.)

Social media analysis: Short picks the closest 10 movies that have grossed at least $50 million and pulls social media engagement (via Instagram, Twitter, Reddit, TikTok) around these titles.

Extract audience segments: Short examines the commercial potential of my project based on attractiveness comparisons across different demographics (age, gender) and which communities to target to help the project go viral. Marvel fans, for example, are good at attracting different communities to their projects; could “The Lose” have somehow been marketed to these crowds?

The analysis showed that my film scored poorly in two categories: uniqueness (was the “narrative DNA” similar to the competition?) and interest (did the script have a large set of characters with a wide range of archetypes? ).

The bottom line: Only one star -Corto recommended Chris Pratt- could make my formula movie successful. Ouch – another blow to my failed career as an aspiring screenwriter.

According to Bergquist, several major movie studios already use the process I just described through Corto’s web platform. The Hollywood connection comes from Bergquist’s role as director of AI at the University of Southern California’s Entertainment Technology Center (ETC@USC).

Founded in 1993 with the help of USC alumnus George Lucas, ETC@USC receives funding from Paramount Pictures, Sony Pictures Entertainment, 20th Century Fox, Walt Disney Studios, Universal Pictures, and Warner Bros. Entertainment, among others.

According to The Hollywood Reporter, before joining ETC@USC in 2016, Bergquist had his own somewhat checkered background in storytelling. In the early 2000s she went by the name of Alexis Debat and frequently appeared on ABC News as an expert on the war on terror. In 2007 he was “exposed for falsifying her resume” and “accused of falsifying a series of interviews with famous people”. It took Bergquist years to rebuild his professional career, including stops at Ray Kurzweil’s Singularity University and the Ranker survey analysis firm. As ETC@USC CEO Ken Williams told the hollywood reporter: “I believe that people deserve a second chance if they have earned it, and I believe that [Bergquist] he has earned it.”

Bergquist is using his second chance to bring an AI solution to an industry that has long relied on instinct.

Legendary screenwriter William Goldman wrote of the industry that “no one knows anything…no one person in the entire field of film knows for sure what’s going to work. Every occasion is a guess and, if you’re lucky, an educated guess.”

Some assumptions have worked very well. Decades before he unleashed his infamous Oscar slap, Will Smith found a pattern in blockbusters: “Nine of the 10 greatest movies of all time have special effects; eight out of 10 have creatures in them; seven out of 10 have a love story. So if you want a hit, you should throw them in the mix. I just study the patterns and try to place myself where the lightning strikes.” This is how we have “Independence Day” and “Men in Black”.

Another safe guess: Since 2000, the highest-grossing movie in 19 of the subsequent 21 years has been a sequel or franchise.

The most systematic method of finding new Hollywood content is the Black List, an annual survey of film development executives who vote on which unproduced scripts they like best. Launched by Franklin Leonard in 2005, the Black List has included more than 1,000 scripts. Of those, 440 have been produced, including four of the last 10 Best Picture winners.

Using technology to find great Hollywood content has a much more complicated history. Ryan Kavanaugh launched Relativity Media in 2004 with an algorithm that he claimed could predict blockbusters. The company went bankrupt in 2016.

Recent efforts using AI have been small-time. Lexus made a 60-second ad based on an IBM Watson script (unlike my very, very, very funny movie script, the ad is very uninspiring), while Fox turned to Google Cloud to create a trailer for a movie.

For its part, OpenAI’s GPT-3 language model is producing scripts that, so far, stand out for their nonsensical dialogue. It is true that the model should improve over time.

Kim Benabib, co-creator of HBO’s “The Brink,” is skeptical of AI in Hollywood. “Silicon Valley has long dreamed of removing the creative human element from the entertainment business because it’s expensive and organized,” he tells me. “What makes a success is intangible. It’s a combination of voice, serendipity and magic that can’t be broken down into math.”

As you might expect from someone who writes creatively for a living, I largely agree with Benabib’s opinion.

According to Bergquist, Corto is taking a new approach because it combines a deep understanding of a film’s content and potential audience engagement.

“We tell development executives how different, unique and fresh their new script is. Next, we give them a deep insight into the audience that is likely to be activated by that story and those characters,” Bergquist told me via email. “And we tell them what attributes of their project (cast, visual effects, music, etc.) will be important to that audience.”

There are AI competitors in the space, but with different approaches. Cinelytic taps into a deep talent database to help with casting calls. Movio informs production decisions based on transaction data from 10 million viewers. On the other hand, companies that primarily measure social media sentiment, such as Talkwalker and Meltwater, do not have a “narrative DNA” database for more than 700,000 movie titles.

The recent TikTok success of “Minions: The Rise of Gru” proves that social can be a box office dealbreaker. And that’s one area where Corto thinks he has an advantage.

As for “The Lose”, I hope Chris Pratt is reading this article (and if you are, feel free to email me).

This note does not necessarily reflect the opinion of the editorial board or of Bloomberg LP and its owners.


Leave a Comment