AI and Behavioral Homogenization

speculations about machine generated world simulators

compiled by GPT4 and Christian von Borries

“standardisation, homogenization, massification”, Klaus Mann, 1942

ANALOGUE:

Behavioral homogenization refers to a process where different cultures or societies start to become more alike in terms of behaviors, customs, and values. This concept is often discussed in the context of cultural globalization, where the spread of certain cultural elements leads to a reduction in cultural diversity.

In the broader context, it’s related to biocultural homogenization, which is the reduction of biological and cultural diversity due to the spread of certain lifestyles and the loss of unique cultural and biological traits. This can have significant implications for both human societies and the environment, as it can lead to the loss of traditional ecological knowledge and practices that are vital for the sustainability of social and ecosystems. It’s a complex capitalist issue that involves the interplay of civilisatoric, economic, and political forces on a global scale.

DIGITAL:

Digital technologies significantly contribute to behavioral homogenization, particularly through their surveillance capabilities.

Especially social media platforms and search engines often promote content based on algorithms that prioritize the duration of individual engagement. This can lead to a narrowing of exposure to cultural expressions and reinforce a homogenized set of behaviors and values.

The extensive data collection capabilities of digital technologies allow for detailed tracking and profiling of individuals. This can influence behavior by encouraging conformity to perceived norms, as people alter their behavior due to awareness of being monitored.

The surveillance capabilities of digital technologies can have profound effects on political systems and individuality.

“There are forms of oppression and domination that become invisible.” (Foucault)

Economic imperatives drive the design of digital technologies, often leading to a focus on standardization and efficiency over diversity. This can result in social pressures that favor a homogenized global culture over local and indigenous practices.

The idea of anarchic structures (including P2P technologies) as a solution to thoses challenges is intriguing. Anarchic structures, in this context, refer to systems that operate without centralized control. While they can promote freedom and equality, they also risk creating a lack of accountability and could potentially exacerbate the very issues they aim to solve.

GENERATIVE:

For the first time, film uses generative technologies to create models of dystopian, anarchic world models that might show us ways out of today’s political dilemmas. They might even refer to our own unconscious.
Machine learning is ultimately used here as an extension and correction tool for human failure of perfecting this world, or a translation of a technical perfectibility, i. e. the will to perfection instead of Nietzsche’s will to power.

Nowadays, a lot in film becomes AI-generated, poems from AI-generated books, but also other texts generated by Chat-GPT/GPT4, voices reading text, then music that might sound like disturbing orchestral soundtracks.

But above all video clips, generated with diffusion transformers.

Video generation involves maintaining coherence across frames. Diffusion transformers excel at capturing temporal dependencies, ensuring that consecutive frames flow smoothly. In video generation, we need to condition the model on specific information (e.g., text descriptions, previous frames). Understanding diffusion transformers is crucial for advancing AI-generated video content. Diffusion transformers are an “attention mechanism” that weighs the relevance of every other input for every piece of input data — in the case of diffusion image noise — and generates an output result, i.e. an estimate of the image noise.

The challenge is to use these generative technologies against their capitalist intentions and show rather dystopian worlds defined by the homogenization of behavior, a characteristic inherent in a technology based on statistical predictions. But here, it could as well be a way of intimidation that in this case needs to be overcome.

What are the effects of AI-generated videos?

1) By showing something like a distorted mirror of reality, they resemble the unconscious. The unconscious refers to human thought, feelings and actions that are determined not only by conscious decisions and processes, but also by impulses, structures or conflicts that are hidden from consciousness and therefore cannot be controlled by it. These videos can show this for the first time.

2) AI produces effects rather than moralistic stories.

The authors would call them post-morals: it’s never a question of HOW, but WHY, because morality depends on historical and psychological influences, which makes it difficult to take a political view of society. Marx observed that morals are always lead by one’s own interest, and Nietzsche, building on this observation, says that they lead to fragmentations of the people.

We can witness a deconstruction of meaning and truth by being anti- moralistic (neither good or bad, nor good vs. evil). Instead, these generated clips show a supposedly subjective but ultimately unempathic staging of world simulations.

3) By simulating an incomplete AMI (Advanced Machine Intelligence), the software ultimately builds models of the world and imitates the uncompounded and unorganized as a representation of the world.
The slight strangeness emanating from the imperfection of videos currently being generated has an inherent utopian force, as it surpasses human imagination.
We are witnessing a simultaneous, multiperspective form of perception of heterogeneous styles beyond dialogue — it seems almost like a self-reflection of form by interpreting the world in unknown ways — that reacts to the simulated world of digital media, and ultimately becomes part of our reality.

4) It questions the function and authenticity of the author, becoming postdramatic and narrativizing.

POST-DRAMATIC:

The postdramatic film moves away from traditional drama-centered approaches. Instead, it emphasizes jump cuts. The goal is to create an effect among spectators rather than adhering strictly to a story. It combines diverse styles and footage itself as “after” or “beyond” dialogue. It challenges the dominance of explanatory speech. Protagonists become themes only for short moments of inconsistency. It might even use overload as a meaningful tool in reference to the world.

NARRATIVIZING:

Narrativizing films refer to works that intentionally play with narrative conventions, often challenging linear storytelling and inviting viewers to engage with the process of constructing or shaping a narrative.

They frequently break the fourth wall (i.e. the supposed independence of the action from the camera) by acknowledging their own status as fictional constructs. They may feature characters who are aware of their roles in a story or directly address the audience.

These films disrupt chronological order, presenting events out of sequence. Flashbacks, flash-forwards, and elliptical editing challenge viewers to piece together the narrative puzzle.

Narrativizing films explore different viewpoints, emphasizing subjectivity. They may present conflicting accounts of the same events.

Narrativizing works incorporate diverse elements — text, images, sounds — into a collage. These fragments evoke emotions and ideas without adhering to a linear plot.

Filmmakers reference other films, literature, or cultural artifacts within their narratives. Resisting neat resolutions, these intertextual connections enrich the viewing experience actively.

MUSICALLY:

Both music and AI involve recognizing patterns and share elements of unexplainability. Musicians receive feedback during rehearsals and performances. They adjust their playing based on their own assessment and audience reactions. AI models also improve through feedback loops. Training involves iterative adjustments based on evaluation metrics. Reinforcement learning (RL) in AI is akin to learning from rewards and penalties. RL agents (like musicians) take actions to maximize cumulative rewards. Language models can adapt their tone and style based on context, similar to how musicians interpret a piece differently based on the context of a performance.

BLACK BOX:

The black box problem in AI refers to the challenge of deciphering the reasoning behind an AI system’s predictions or decisions. Essentially, it’s the opaqueness of understanding how the system arrives at its conclusions. Even if we have access to the data it was trained on, we cannot fully understand how it makes decisions.

Music, too, has its own irrational aspects. Certain music pieces evoke strong emotions, yet we struggle to pinpoint why. It’s often a blend of memories, emotions, context, and of course how the piece was modelled. Both phenomena have in common the inability to rationalize the world.

WORLD SIMULATOR:

A world simulator refers to an AI system that constructs an internal representation of an environment and uses it to simulate future events within that environment. These simulators can model various aspects of the world, such as physics, dynamics, and interactions. Video generation models that operate on spacetime patches of video and image latent codes can serve as promising general-purpose simulators of the physical world, predicting real-world phenomena.

“Only when the model is changed can lessons be learned from history” (Heiner Müller).
Today’s model is a Large Language Model (LLM), trained with the entire online knowledge that is constantly self-optimizing.

By definition, AI creates generic content. Couldn’t the lowest common denominator show something of the truth of this world?
On the other hand, don’t the above mentioned attempts of standardization lead to fascism, which leaves nothing but atmosphere behind?

There are signs that machine learning already becomes feminist, for it may itself have uncovered all the prejudices created by generally white, male programmers.

A feminist approach to technology extends beyond gender disparities. It encompasses broader social inequalities, including race, class, disability. Feminist perspectives on Generative AI (GenAI) involve rethinking how we create and apply this technology.

As test screenings show, the AI powered post-dramatic, narrativizing film elicits very diverse reactions, depending on the context and perhaps also the origin of the spectator.
The authors sees this expansion of genres and, ultimately, “passionless dialectic” (Heiner Müller) as its strength.