Chat GPT 3.5, Content Creation without Creativity

To declare that artificial-intelligence generated media has begun to threaten everyday forms of news and entertainment consumption is the understatement of the decade. While the rapidly evolving course of the media form has begun to flip entire industries on its head, with the introduction of free, online forms of artificial-intelligence engines. While the most obvious example of this can be seen with the currently running ChatGPT 3.5, which can be accessible with the simplest step of having a google account, or competitors like OpenAI playground, or JasperGPT are only increasing the speed at which artificial-intelligence tools are catching up with the duties of manned-positions, whether they be data analysts or writers, this threatens modern conceptions of plausible careers and labor. While contemporary knowledge at the rate at which artificial-intelligence is somewhat insecure, its first targets undoubtedly begin with content creation, with larger corporations where cubicles are commonplace, often likely to be the first targets (Miller & Cox; 2023). It may be impossible to tell whether a new modern conception of late 18th-century luddites will return, but what is undeniable, is the effect that artificial-intelligence has begun to form on the media world at-large. Though this essay will be dedicated solely to snippets of script written by the established ChatGPT, the use of artificial-intelligence has been found in all forms of media and entertainment today en large. Whether it be the congress hearing set-up by internet music personality Rick Beato, warning of the threats to the foundations of childhood creativity with AI written samples, or the most recent strike against the majors studios including Warner Brothers, Disney, etc., which named a clause to supposedly ban AI partially or wholly written scripts, media, popular entertainment, and what people consider artforms in particular has taken on artificial-intelligence as a direct threat. Fundamentally, the critique that exists in any AI piece of media, entertainment, or art, is clearly absent from a human element which adds a layer of personal experience, depth, and even the writer’s subconscious through indirect themes, motifs, and ideas. This is where a piece of entertainment, or art, may have the ability to seem to breathe new life upon each viewing which can range from a painting to a pictograph. These abilities are thus inverted upon using an artificial writer rather than a human one, which will thus be examined. However, what will also thus be analyzed, is how each script is artificially written, what elements does the intelligence tie together to complete a script, and how does this combination of intelligence differ from that found in the script of a human’s. This analysis will take three scripts written by ChatGPT 3.5 and analyze not only what elements this AI draws to compose each scripts, but how each script is written, the three given titles ascended in specificity, ‘a romantic comedy’, ‘a Roman era epic’, and a script by the late, great Paul Thomas Anderson. 

Upon a first examination of the most broad definition to begin a script from with ChatGPT 3.5, ‘romantic comedy’, the foundational elements seem to be rather familiar and almost purposefully non-threatening in nature. The setting of a New York City pastry shop as the place of work for this main character serves as the perfect example. The setting of New York, an unfamiliar location to, in-practicality, no studio executive, serves as the most obvious baseline example as the location of New York City has been the setting for not only classic romantic comedies, take When Harry met Sally (1989), but also is the setting for two Netflix movies hitting netflix and selected theaters right now Your Place or Mine (2023) and Love Again (2023). However, upon further examination into the script the influences which ChatGPT, the clear influence of mass volumes of multidimensional data, including images, words, etc., takes the possibilities of what could be in the script and dilutes them to their most basic and recognizable level. According to research performed by professors Luchen and Zhongwei from Guangdong Business and Technology University and the Baptist University of Hong Kong, it is the ability to amalgamate such a mass amount of information immediately which directly resorts to the current forms of script writing templates that pluck repetitive ideas, dialogue, and images associated with the input script definition (Luchen & Zhongwei; 2023). This can be seen being reflected clearly in the choices that ChatGPT made in the creation of the ‘romantic comedy’ script. While the choice of a coffee shop, pastry store worker as the female protagonist is entirely formulaic its rigid, highly conformed designs of protagonists shares more in common with its opposites which, again, is a near stereotype, action flicks where the muscles make the movie including the likes of the former governor of California. Fundamentally, this reveals that the vastness of information which the artificial-intelligence collects ensures that it will only collect the most repeated, commonplace, and stereotypical story elements. The most obvious example of this can be the introduction of Alex with Emma’s line referring to the rain. 

ALEX:

(slightly flustered)

Sorry I’m late, folks! Got caught up in a breaking story.

EMMA:

(teasing)

More like you got caught in the rain, Alex. Again.

The choice of rain  in this scene by ChatGPT is undoubtedly nothing more than an algorithmic conclusion made through an analysis of hundreds upon hundreds of romance comedies in seconds. However, as previously mentioned by the combined Guangdong, Hong Kong analysis, such a massive amount of information will ensure that only repetitive ideas associated with the given definition will be utilized. Therefore ChatGPT with a reference to rain, while not taken arguably its most cliché with a mid-downpour makeout session, still must insist on creating script with the most typical and repetitive settings, beats, and dialogues due to its mathematical nature by design. Hence the character of Alex must not only be caught up in a ‘breaking story’, but he must also be ‘caught up in the rain’. Rule of numbers, or volume, of data collected with the most stereotypical groups 

In an attempt to examine how an AI generated script would react to a more directly a new definition for a ChatGPT script was created and its title was likewise more narrowed down with ‘epic, Roman era’. The results ChatGPT presented, however, seemed to portray a similarly self-referential nature to the admittedly more defined genre chosen but again are steeped in stereotypes which result from the artificial-intelligences formulaic nature. The characters presented seem to indicate ChatGPT’s reliance on repetition rather than an amalgamation of conscious and subconscious decisions made by a writer, however what’s worse is their traits are more in-line with those of 50s and 60s Hollywood stereotypes than with their real-life counterparts. The late Roman emperor Marcus Aurelius, Cassius, and a soldier, Maximus, are the only characters which appear in the script, however the writing follows not the conventions of their ancient counterparts but simple script writing conventions with the epic genre. Unsurprisingly the script features lines which seem to only focus on the main ideas found in the work of Roman era epic pictures, rather than the determination of the characters the script portrays. While the artificial-intelligence script chooses the characters of Marcus Aurelius and Cassius which are clearly linked to the renowned emperor of Rome from 161 to 180 AD, and the Shakespeare character of the same name, Cassius, from his late 16th-century history, Julius Caesar. However, upon a basic examination of the generated script, the not only are the given words highly formulaic, but whose chosen lines, settings, and verbal indicators reveal that the dependence on recurring patterns inherently limits what dialogue, settings, and narrative can be chosen by the artificial intelligence. Ann Hornaday sums-up the ideas surrounding the ability for an artificial-intelligence to write its own script by clarifying that it is ‘not possible, but utterly probable’, that the future of ChatGPT will include the ability to create complete, film-length scripts (Hornaday; 2023). Still, she combines this critique with the rule of repetition found both in the previous ‘romantic comedy’ template as well as this current ‘epic, Roman era’ template, to levy a critique against a method found present, today in Hollywood and ChatGPT alike, the recycling of ideas and content. 

The script written for the corresponding ‘Roman epic’ template, upon deeper examination, clearly uses an obscene amount of recycling in ideas, dialogue, and character traits, from other films of the chosen genre. While the introduction of Marcus Aurelius, may at first hold some clear connections to films of its genre, obvious examples being a direct reference to roman citizens, but his ideals are so recycled that they morph into nothing more than empty platitudes. 

MARCUS AURELIUS:

(voice booming)

Citizens of Rome! We stand on the precipice of greatness, bound by the ideals of honor, duty, and glory of our empire!

The crowd erupts into cheers, their voices echoing off the marble columns and statues that adorn the Forum.

What the writing choices of the artificial-intelligence ChatGPT 3.5 reveal is a reluctance rather to give a precise definition as to the Roman-era subject at hand. This inability to focus on a single idea, moment in history, or even a fictionalized one, to soak and gather themes from adjacent material, and rather to take the purely formulaic approach of whatever is most repetitive and commonplace in the defined genre given, deflates the writing. Hence, when Marcus Aurelius refers to the Roman Empire, the artificial-intelligence sternly keeps its ways from the often small-spoken yet decisively brilliant statesman that led Marcus Aurelius to write Meditations, a form of self-poetry expected from a highly resigned individual, chooses to give speeches to the Roman masses. Fundamentally, the inaccuracy of the writing of the real-life character of Marcus Aurelius, in the formulaic creative mind that is ChaGPT 3.5 at this point, has been remedied by the reference to honor, duty, and the Roman empire, provided by the character’s dialogue. 

Finally, the most specified template given for ChatGPT 3.5, writing a Paul Thomas Anderson script, may have resulted in the most limited script with, likewise, the most recycled elements. Unsurprisingly, according to the previous limitations attached to ChatGPT by work of both the research by the Guangdong and Hong Kong combined analysis and the work of Washington Post writers alike, artificial-intelligence scriptwriter, particularly the freely available ChatGPT 3.5, will inherently rely on the most repetitive elements found in each multidimensional scan of online data relating to the given script template. However, the elements found here reflect not only the rule by volume element which led to the previous examples of inaccurate descriptions of Roman emperors made-up for with buzzwords like ‘honor’ and ‘duty’, as well as the choices of a downpour cliché for the entrance of the Alex character. However, with the choice of Paul Thomas Anderson as the input director for ChatGPT to create a script from, the inherent nature of ChatGPT as an artificial-intelligence ensures that only the most repetitive results related to the name, PTA, will be mirrored in the generated script. However, by using a director whose still actively producing pictures, the most repetitive and recent results related to this director all tie to his newest film, Licorice Pizza (2021). While the overwhelming majority of movies produced by the highly respected auteur take place in California, whether it be the turn of the 19th to the 20th century in the immortal There Will Be Blood or his follow-up which takes place in the middle-of that same decade, the clear majority of movies made written and produced by this director take place in seventies southern California, namely Boogie Nights (1997), Inherent Vice (2014), and the already mentioned Licorice Pizza (2021). Chat GPT’s entire design revolves around repetition, thus the chosen location for PTA of the San Fernando valley seems not only unsurprising, but in-practicality, feels dry in comparison to what the writer-director just created. 

Nevertheless, the most obvious shortcomings of this most specific template given to an artificial-intelligence, the PTA script, can be seen in a comparison of the ChatGPT results to Licorice Pizza (2021). Unsurprisingly, the artificial-intelligence of 3.5 chose the same location of his previous picture, the San Fernando Valley, however the artificial-intelligence’s attempts at writing the playfully awkward nature of the characters PTA seems to so often bring to life, take any official attempts by Gary Valentine, portrayed by the son of long-time collaborator and friend Philip Seymor Hoffman to court Alana (Alana Haim), have been entirely removed. ChatGPT, through its inherent reliance on formulas categorizing repetition as input to the script template, the spontaneity, humanity, and thus the complexity found in the writing of ChatGPT is anything but. A clear distinction between the two can be seen comparing Jack’s introduction to Lily in the ChatGPT script with Gary Valentine’s introduction to Alana in PTA’s script. While Cooper Hoffman’s first introduction to Alana could be at its very worst could led to what could be categorized as sexual interactions with a minor, due to the age difference between Gary and Alana given purposefully in the film, ChatGPT’s response includes anything but such a sensitive and frankly far more controversial whose unnatural romance can seem to only be presented through the lens of a writer-director who knows exactly what he wants. The Lily and Jack scene, however, presents anything but. 

JACK:

(voiceover)

They say love is like a song – unpredictable, yet impossible to resist.

Jack approaches Lily, his heart racing with anticipation

JACK:

(awkwardly)

Hi, I’m Jack. Mind if I join you?

Lily’s eyes sparkle with amusement as she gestures to the empty seat beside her.

LILY:

Be my guest.

As Jack takes a seat beside her, a spark ignites between them, setting the stage for a love story that will echo through the valley for years to come. 

While the dressings of Licorice Pizza may be all here, the crucial first-interaction between Gary and Alana, or in this case Jack and Lily, lacks any authenticity or complexity found in the human created script. Frankly, the first lines that Gary speaks to Alana could at their very worst be categorized as cat-calling, insisting on the beauty of one of the assistants preparing school photos, Alana, however due to the context, delivery by PTA, and charm of Cooper Hoffman, these moments feel warm and loving rather the cold, uncomfortable, and sexualized. However, when comparing any word of dialogue between the two, clear distinctions in the human-nature of the work can be made. Take the wonderful moment when Alana insists that Gary Valentine ‘stop breathing’, or upon first being asked out by Valentine, insisting he’s twelve when he clarifies that he’s fifteen. This latter quote to the film, occurs upon the couple’s first interaction with one-another, while the first introduction to Jack and Lily, whose romance will supposedly echo throughout the valley for years to come, doesn’t include any charm, charisma, or chemistry at all, the things which would supposedly catalyze a romance to blossom. A final article which analyzes the capabilities of ChatGPT can be found in the writings of Stuart Heritage for The Guardian whose experiments revolved around giving templates of scripts that had certain purposes in mind, namely one for generating massive capital, and one for winning an oscar. However, Heritage after researching into what ChatGPT can produce through a script, concludes that it is not the ability to write a script that ChatGPT has, but the ability to generate plot ideas, still Heritage admits that any lines produced by the artificial-intelligence are simply too dull for a billion-dollar script (Heritage; 2023). Fundamentally, this dullness can be seen across all three script prompts. Whether it be the accuracy of past historic figures to conform to genre expectations, the recycling of narrative and setting elements without the ability to conceive unique, character-driven dialogue, or the inclusion of story elements rather than character development which seem to relate to the given prompt, a rainstorm but no romantic dialogue in the ‘romantic comedy’ script translates undeniably to a lack of creativity and pension for reality in interactions made-up through dialogue in the generated script. 

The level of reliance upon artificially generated scripts, series plotlines, music, and frankly all forms of media, have undeniably prompted complaints from the industry’s professionals today. While the most obvious example was seen with the most recent writer’s strike which attempted to eliminate any possibility of an AI generated script from being released, the results seen in the movie theater with the third big-budget studio Willy Wonka effort in fifty years. This is a clear example of recycling which just as well could be found in the script to a children’s comedy written by a ChatGPT, with the transformation from a chocolate factory to a fudge factory and the name Willy changed to Billy. Fundamentally, this is the issue that Chat GPT presents, repetition for the sake of efficiency, a lack of interest, and frankly, just laziness. The nature of ChatGPT can be seen at its most obvious when creating Stuart Heritage’s movie that will make billions of dollars. In a reference to arguably the most successful film series of all time, Star Wars, the title given reflects the first of the series, A New Hope, with the artificial-intelligence’s adaption of this to ‘The Last Hope’, the ultimate irony here is such a name for the original Star Wars, wasn’t conceived until years after the movie was released, the name ultimately having no effect on the picture. 

https://chat.openai.com/share/00f42816-5072-4b0c-a82f-0aa5f511b179

Hornaday, Ann. O’Sullivan, Michael. February 17th, 2023 “AI is a time-honored movie trope. Now it can write its own script.” The Washington Post

Luchen, F., & Zhongwei, L. (2023). ChatGPT begins: A reflection on the involvement of AI in the creation of film and television scripts. Frontiers in Art Research, 5(17).

Heritage, Stuart. March 6th, 2023. “Can an AI program really write a good movie? Here’s a test.” The Guardian.

Cho, T. (2023). A Study on Dramaturgy for AI Screenplays: Writing Alternative Narratives Using GPT (Doctoral dissertation, State University of New York at Buffalo).


Comments

Leave a Reply

Your email address will not be published. Required fields are marked *