Film production and the industry is changing constantly since its very beginnings in the 19th century but the digital transformation is a fundamental change like nothing before. Most of all there are unparalleled hopes and fears; on one side this is the Yukon, the great gold rush, on the other side it’s overwhelming if we do not commit ourselves to a constant learning process we don’t survive.
Digital transformation can mean a lot of things. There is no simple definition. It’s both a challenge for artists as production companies or any distributor. Endurance Entertainment GmbH approach was from the beginning to be an open-minded learner. Being in Berlin and in a constant exchange with digital companies some insights formed over the last years. There are no golden rules here, but things appear to become more clear over the time and we might find a sort of a “survival guideline”:
I. First of all, any discussion or any attempt to adapt means a lot of organized thinking. Since the early years, things got easily mixed up often in public panels. A good example is video-distribution. Even it’s very obvious content produced for platforms like Youtube is a completely different thing like Netflix. It’s true that change is constantly happening and it takes some effort to be informed about new business models or new platforms.
II. Second, don’t be intimidated because you are not a digital native and ask questions. A lot of film producers didn’t have the time or see any sense in investing their precious time to understand what a multichannel-network is or how they should develop an online marketing campaign. Today, there are many possibilities for digital marketing but we need to know if it serves the intended purpose. Even if you don’t want to do your social media work yourself it’s always good to understand how it works. If we understand something it enables us to ask the right questions.
III. The 4K Age – Anything is possible: Yes and no. Digital transformation in the first place is a massive change in technology. Not so long ago at the time of “Super-8” there was an eternal border between the “amateur” and “the professional”. That changed (even if some people will protest here), but its good to differentiate. Basically, there are a huge variety of standards here. And its true that smartphones and DSLRs can do amazing things and yes we can do the edit with freeware. The question is what do we want to achieve and/or what is the client’s expectation. This means we need to do our homework and have the necessary info about our equipment before we make choices, i.e. DSLRs might do great pictures but could pose problems in VFX-postpro.
IV. Try harder. If we like it or not but digital transformation requires more technical understanding for many professions in the industry. The producer today needs technical knowledge and you need to be a fast learner. This is often a frustration for people who want to use things and not always to learn “how to”? The experience is that many technologies need time to understand or some attention but learning-by-doing works more often better as expected when we overcome the first difficulties.
V. Long-Term Commitment: sometimes it appears that efforts with video/film in the digital world do have no clear business models or being completely unpredictable. It’s true there are often very short time spans or anything appears fast-paced but it’s also true that successful format development often needs a lot of time.
The conclusion is that it’s always better to be active rather than waiting for what is happening.
This was about scratching the surface. In the blog section of Endurance Entertainment GmbH more will come up about Film and the digital age hopefully helping professionals as interested people.
The rapid evolution and integration of KI tools in the film industry requires a substantial evaluation process about potential risks, impact on creative processes, and benefits.
In nearly any stage of feature film development, production, and distribution, KI tools are present in workflows, such as screenplay writing, development, decision-making processes by producers and studios, and further handling of movies.
There are different tools in the writing process, which help writers in research, developing ideas, characters, and story arcs, and provide instant feedback. It can be assumed that many screenplay development processes are meanwhile shaped by the involvement of KI-Tools.
AI-driven programs like Greenlight Coverage orAI Script Coverage Pro offer instant feedback on screenplays and give both overall ratings and detailed ratings on certain aspects, such as characters. They also provide different types of loglines and synopses of the project they analyzed.
Subsequently, not only writers will use these tools but also producers, development executives, acquisitions in studios, film subsidies, and other involved parties. Time pressure and opportunities to save money might lure executives into using shortcuts via KI-generated coverage services.
It doesn’t end here. Executive summaries, correction of text, and research requests inside studios and production entities might also be generated by a KI-agent like CHAT CPT.
As an immediate result, there is a potential impact on the decision-making process: To save time and money, executives might choose the KI-agent for an opinion and coverage of less important projects. which makes it even harder for new or unknown writers to break through. Unconventional projects or special interest projects might also be stopped before they reach the point where consideration by humans takes place.
The advantage gained by those using automated (KI) tools might also pose competitors at risk, since decision-making processes happen within unparalleled timeframes.
Normally, neither companies nor any subcontractor should use KI tools that use the whole available information like a screenplay WITHOUT explicit consent of the writer or the owner of the material: Once fed to a system, you never know which ways the data travels, who has access to sensitive data, and if the provider has sufficient security measures in place to prevent data breaches.
However, we must assume that stakeholders make use of available tools, especially if time-sensitive tasks are involved. It is an open question whether markets like EFM, AFM, or Marché du Film have already fully realized the impact. It’s up to the players in the market to commit to fair agreements, which both protect their own interests, their projects, and the creatives.
The highest risk
If we must assume that we have many parts in the workflow of writing, developing, assessing, producing, and distributing a movie where KI is involved, it is a much bigger risk involved here:
A negative feedback loopthat leads to poor results and the decline of the industry. It could also be a development that leads to politically and socially problematic trends.
This is not about KI vs creativity; this is about how KI tends to shape content due to its very nature.
It is essential to understand what we are dealing with here and gain an idea of what this actually entails. We are working here with so-called Generative AI:
“Generative artificial intelligence, or GenAI, uses sophisticated algorithms to organize large, complex data sets into meaningful clusters of information to create new content, including text, images, and audio, in response to a query or prompt. GenAI typically does two things: First, it encodes a collection of existing information into a form (vector space) that maps data points based on the strength of their correlations (dependencies). Second, when prompted, it then generates (decodes) new content by finding the correct context within the existing dependencies in the vector space.” George Lawton, TechTarget
Reading definitions can’t replace the experience of months working on the training of these KI models, which I had done over several months for an additional income.
The smooth and, for some people, exciting experience of using KI apps is often the result of an army of data annotators and labelers scanning the results of AI over months, and the work of huge development teams.
In the first place, mismatches to the prompts were flagged, and other, often technical, failures. In the end, it’s often an evaluation of what’s right and wrong, for example, does a speaker sound like a native speaker, or does a KI-created picture look realistic? Linguists check if language is accurate, translations are wrong, or if there are typical failures like misinterpretation of metaphors.
In all these cases, there are references. Language has rules; a picture looks suspicious if things are appearing with the wrong lighting.
The question is, what is the reference of a KI-model that helps a screenwriter to write or provide coverage about a screenplay? What are the factors involved in programming and designing software that, in the end of the day, provide ratings for screenplays?
And most of all: What exactly is the database that was used to create the software?
Who wrote the rules on how to use the database?
Certainly, there is a reference, and this reference is known from endless Screenplay Writing Books and tutorials. It has to be admitted that, especially, structural elements have become stable conventions. Often, there are prominent “User cases” where the screenplay of a successful film has become a reference. And there is dramaturgy, one of the oldest sciences, which is basically the rhetoric, the arrangement of elements to make it compelling. Simplified.
We know from decades of experience that forcing screenplays into mechanical rules, you get mediocre results at best, and you repeat what has been done before, and thrillers, for example, become predictable.
Success is often the result of a mix of clever use of narrative clichés and innovative elements (Charles Eidsvik – quotation from memory).
We all know the smart use of tropes and references, which is a huge part of the appeal of movies.
However, any convincing screenplay has a strong core idea. All the tropes and new ideas are orchestrated in a way that the core idea finds the best expression. The core idea is the rule and presents the intention of the writer. This intention can be a painful personal experience in a drama or an appeal to stand up against injustice. This requires will and an agenda.
When you rate a screenplay, you need to have an approach where you grasp the core idea and whether the writer has found a convincing way to express it. This includes often unorthodox ways and even going against the rules. The rhetoric of convincing screenplays demands a lot of subtext and metaphors. It is an experience that forms in the mind of the reader and later in the mind of the viewer in the movie theatre. This requires consciousness and a will to understand.
As human-like as a KI program might appear, it will never have the ability to create the “experience” a reader has when reading a screenplay, because it is something fundamentally different. The KI-models create their result on the basis it was fed with and based on an algorithm, but not based on the complex thoughts of a living being. Its decision-making comes from statistical preferences and not from the framework of consciousness.
Results might sound convincing, but in the long run, they will limit the innovative and creative potential of screenplays because they filter information, not making decisions based on human awareness and intelligence. The result will become dull, less intelligent, predictable, mediocre, conformistic, and too similar to what we have seen before. This is due to the reductionistic tendency of all KI-Models. A KI-software rating screenplays can only be conformistic; it cannot go against the rules. It might also not be sensitive to the rules that apply to different genres.
The limitations of using KI models for evaluation might be limited if they occur at only one stage in the development and decision-making process. However, as stated in the introduction, if KI is involved in many stations in the workflow, the tendency of KI to streamline, be reductionary, and overlook whole dimensions of a project due to its very nature leads to a negative feedback loop. Let’s say you had the writer using KI, which pushed him already a bit into a direction that works but is very conventional, then you have summary and coverage by KI, and this coverage gets further reduced by an executive summary, and a lot of important information could get lost along the way: easy to read, but not a precise statement anymore.
However, there is even more to it: recent studies found an alarming trend in KI models that they chose themselves AI-generated Content over human-made content and have their own bias, which could lead to unintended tendencies. This is the real dangerous negative feedback loop. Even more: AI might not only homogenize screenplays and movies but the decision-maker and creators themselves: https://www.newyorker.com/culture/infinite-scroll/ai-is-homogenizing-our-thoughts
There are even more factors that need to be considered:
The system of norms and values in the narrative of a screenplay is important in how we, as the audience, react and engage with a movie. Norms and values determine which emotions a narrative triggers. Think of “Gladiator”: the injustice keeps us engaged, and we suffer with the hero.
And here is the question: can KI understand the norms and value system in a screenplay, and how does the bias of the algorithm shape the judgment of the narrative?
In the worst case, a political actor could influence the KI-model in a way that its ratings conform with a certain political intention. But even without any bad actor, if the software is inclined to rate the norm&value system in a narration, there can be a bias which might not be helpful if the intention is a successful and convincing movie.
But can any KI-model rate a norm & values system in a text or a screenplay?
In any good movie, the norm and value system is not revealed in a placative way or in simple statements in the dialogue. If that happens, we have propaganda. Intelligent screenplays use images, actions, not simple words in the dialogue. Rest assured, a lot of these elements will fly under the radar of any KI coverage software.
Context
Another important aspect for any producer, buyer, developer, or screenwriter is context. A screenplay is not a standalone. How a movie is perceived by the audience depends on the undercurrents in society,trends, and preferences in certain genres. The most important context is the director of the movie – if a director is attached or if he is the writer, this makes a huge difference in how a producer should read a screenplay. You know where your director is strong, where he has certain habits, or what he can do with a certain scene. Then comes the equally important cast. Sometimes, a screenwriter breaks the rules because it benefits a prominent actor who is strong in a certain situation. Again, you read a screenplay differently when a cast is attached. You ask, can it work with these actors, or does the screenplay need to be rewritten? But the environment is much bigger: Is the screenplay something a certain production company can do successfully (do they have the experience and the resources)? Does the writer have the necessary technical knowledge to judge if, for example, a lot of use of Unreal Engine with costly digital effects is necessary? Does that work visually and technically? There are a lot of complex factors involved.
This is what project assessment (and what Endurance Entertainment is doing) comes into play: not rating a screenplay in isolation, but within the complex network of numerous factors.
Recommendations
That said, it is not judging KI models for the screenplay development and coverage entirely negatively. It can have a place in a certain stage in the development and production process, especially as a quick check of the mechanics and the construction of the building works. It can inform us where we are in the development process.
The main point is that it requires an experienced pilot in the cockpit who makes the final decisions. If KI is involved repeatedly in workflows, it might not trigger conformistic and boring results, which might work “technically” but will not produce hits that the audience loves. It will lead to degeneration and loss of innovation. Therefore, the pilot has to put KI-generated results in the right place and use them as an auxiliary tool.
Last but not least, a lot of knowledge and experience is required by the pilot giving the rating and the final decision due to KI’s tendency to hallucinate: If the model doesn’t know the proper answer, it starts to lie or comes up with misinformation. This is especially relevant for any comparison with other movies. The list of comps requires a lot of research and background knowledge. Here is an inherent risk that KI models provide misleading information.
Everyone using KI-Models needs to have background knowledge, instinctive awareness, and the ability to fact-check everything. As a decision-maker, you have to always read a screenplay yourself or hand it to someone you trust because you need to know where KI misinterprets or falsifies important facts or produces misinformation.
There is even more to this when it comes to the future of whole companies: Companies heavily relying on KI models in every stage of development might lose their advantage and relevance in the industry, because the knowledge base is the common knowledge base everyone has access to – KI models use public and broadly available data. The unique strength of companies lies in their “superior wisdom“, the inspiration and wisdom creative people and team members bring to the table, which can’t be replaced by machines.
Notes:
Definition of negative feedback loop: “A negative feedback loop is a regulatory process where the output of a system, process, or mechanism reduces or dampens its own activity, leading to a decrease in the output or a return to a stable state or equilibrium. “