Living with the Beast, Part 2

Welcome to Celebrate Creativity - Episode 409 - Living with the Beast, Part Two

We begin this episode with a discussion of generative AI - 

Now Generative AI models are trained on vast datasets of existing content (e.g., millions of images, billions of words of text, hours of music).

During this training, they learn the underlying patterns, structures, and relationships within that data. They essentially build an internal "understanding" of what makes a particular type of content what it is. For instance, a text-based generative AI learns grammar, syntax, common phrases, and even nuanced writing styles. An image-based one learns shapes, colors, textures, and how objects are typically composed.

This learning process often involves complex neural networks, which are inspired by the structure of the human brain.

Thank you for experiencing Celebrate Creativity.

Transcript

Welcome to celebrate creativity - episode 409 - living with the beast part two.

We begin this episode with a discussion of generative AI - 

Now Generative AI models are trained on vast datasets of existing content (e.g., millions of images, billions of words of text, hours of music).

During this training, they learn the underlying patterns, structures, and relationships within that data. They essentially build an internal "understanding" of what makes a particular type of content what it is. For instance, a text-based generative AI learns grammar, syntax, common phrases, and even nuanced writing styles. An image-based one learns shapes, colors, textures, and how objects are typically composed.

This learning process often involves complex neural networks, which are inspired by the structure of the human brain.

Creating new content:

Once trained, when you give a generative AI model a prompt (which can be text, an image, or another form of input), it uses its learned understanding to generate novel output that aligns with the patterns it has identified.

It's not just stitching together pieces of its training data; it's creating something genuinely new based on its internal model of how that data is structured. It's like it has learned the "rules" of a particular type of content and can then apply those rules to create new examples.

Examples of what generative AI models can create:

Text: Articles, stories, poems, emails, code, dialogue for chatbots (e.g., Large Language Models like Gemini, ChatGPT).

Images: Realistic photos from text descriptions, artwork in specific styles, variations of existing images (e.g., DALL-E, Midjourney, Stable Diffusion).

Audio: Music compositions, realistic voiceovers, sound effects.

Video: Short video clips from text prompts, animations.

3D Models: Objects, environments for games or simulations.

In essence, generative AI is about the AI system being able to produce original content rather than just analyzing or classifying existing content. This ability is what makes it so exciting and, at times, controversial in discussions about creativity.

What are some examples of why generative AI is so exciting and what are some examples of why generative AI is so controversial in discussions about creativity?

Generative AI is truly a double-edged sword when it comes to creativity, inspiring both immense excitement and significant controversy. Let's break down why.

Why Generative AI is So Exciting in Discussions About Creativity:
Democratization of Creativity:

Lowering Barriers to Entry: Previously, creating high-quality art, music, or complex written works required years of specialized training, expensive tools, and significant talent. Generative AI makes it possible for virtually anyone with an idea to create impressive outputs, even without traditional skills. This can unleash a wave of new creators and diverse voices.

Accessibility: For individuals with disabilities, or those who lack access to traditional education in creative fields, generative AI offers powerful tools to express themselves.

Unleashing New Ideas and Overcoming Blocks:

Idea Generation on Steroids: AI can rapidly generate countless variations, concepts, and starting points for creative projects. This is incredibly valuable for overcoming "writer's block" or an artist staring at a blank canvas.
Exploration of Uncharted Territories: By blending styles, generating unexpected combinations, or creating novel forms, AI can push the boundaries of what's creatively possible, leading to entirely new aesthetic experiences and artistic movements.

"Generative Synaesthesia": Some researchers describe a phenomenon where human exploration and AI work seamlessly together, allowing artists to explore new creative areas that might be difficult to access otherwise.

To use a phrase, that I introduced last week this is Learning to live with the beast - AI being the beast.  - depending on your attitude.

Efficiency and Productivity:

Automation of Mundane Tasks: AI can handle repetitive, time-consuming tasks like background removal in images, initial drafting of text, or generating variations of a design. This frees up human creators to focus on higher-level conceptualization, refinement, and strategic decisions.

Rapid Prototyping and Iteration: Designers can quickly generate multiple design concepts, artists can visualize different compositions, and writers can explore various plotlines in minutes, drastically accelerating the creative process.

Scaling Production: Companies and individuals can produce more content, faster, and at a lower cost, which can be beneficial for marketing, media, and other industries.

Personalization and Customization:

AI can generate content tailored to specific audiences, individual preferences, or even real-time user responses (e.g., dynamic chatbots, personalized marketing copy). This creates highly engaging and relevant experiences.

New Forms of Collaboration:

AI becomes a "co-creator" or "creative partner," allowing for dynamic cycles of drafting, editing, and reworking content. Human intuition blends with machine output, leading to novel results.

Why Generative AI is So Controversial in Discussions About Creativity:

Copyright Infringement and Data Scrapping:

Unauthorized Training Data: A primary concern is that generative AI models are trained on massive datasets of existing copyrighted works (art, music, text) often without the original creators' permission or compensation. Artists and writers argue this is a form of "mass copying and exploitation."

Derivative Works: When AI generates content in the "style of" a particular artist, or produces outputs highly similar to existing works, it raises legal questions about whether these are infringing derivative works. Lawsuits are actively being filed by creators and organizations (like The New York Times, Getty Images, various music industry groups) against AI companies over this.

Fair Use Debate: AI companies often argue that training on copyrighted material falls under "fair use," but many creators and legal experts dispute this, especially when the AI-generated content directly competes with the original works.

Job Displacement and Economic Impact:

Threat to Creative Professions: Many artists, writers, musicians, graphic designers, and voice actors fear that generative AI will automate their jobs, devalue their skills, and make it harder to earn a living. This has led to strikes (e.g., Hollywood writers and actors), protests, and demands for regulation.

Unfair Competition: AI-generated content can be produced at scale and often much cheaper than human-created content, leading to concerns about unfair competition in the market.

Misinformation, Deepfakes, and Ethical Misuse:
Synthetic Media: Generative AI can create incredibly realistic fake images, videos (deepfakes), and audio that are indistinguishable from real ones. This poses significant risks for spreading misinformation, defamation, fraud, and even political manipulation.

Bias Perpetuation: If the training data contains societal biases (e.g., gender, racial stereotypes), the AI will likely learn and perpetuate those biases in its generated outputs, leading to problematic or offensive content.

Privacy Concerns: The use of personal data (including likenesses) in training datasets raises privacy issues, especially if AI can generate images resembling real individuals without their consent.

Can you give me a specific example of where an AI produced work has competed with a work that is produced by humans - the controversy - and the ultimate result

One of the most widely publicized and controversial examples of AI-produced work competing with human-produced work is:

The Case: "Théâtre D'opéra Spatial" by Jason M. Allen at the Colorado State Fair Fine Arts Competition (2022)

The piece  is a digital artwork created by Jason M. Allen using Midjourney, a generative AI program that creates images from text prompts. The artwork depicts a fantastical, operatic scene with classical figures looking out into a vibrant, sun-drenched sci-fi landscape, rendered in a style resembling an oil painting. Allen further refined the image using tools such as Photoshop and Gigapixel AI.

The Competition: Allen submitted "Théâtre D'opéra Spatial" to the Colorado State Fair's annual art competition in the "Digital Arts / Digitally Manipulated Photography" category. He explicitly disclosed on his submission that the work was created using Midjourney.

The Controversy:

Winning First Place: "Théâtre D'opéra Spatial" won first place in its category, beating out human-created digital art. This immediately ignited a firestorm of debate online and in the art community.

"Cheating" Accusations: Many artists and commentators accused Allen of "cheating" or undermining the spirit of artistic competition. They argued that simply typing prompts into an AI program did not constitute "art" in the traditional sense, and that it devalued the skill, effort, and years of practice that human artists put into their craft.

Lack of Human Authorship: A core point of contention was whether Allen was truly the "author" of the work, or if the AI was the primary creative force. Critics questioned the amount of human creativity involved in generating the image versus the AI's autonomous generation based on its vast training data.

Allen's Stance: Jason Allen, for his part, was unrepentant. He argued that he spent significant time (reportedly over 100 hours) refining prompts, selecting the best outputs, and making post-processing edits. He maintained that AI was simply a new tool, akin to a camera or Photoshop, and that his creative direction was paramount. He famously stated, "Art is dead, dude. It's over. A.I. won. Humans lost."

The Ultimate Result (as of current knowledge):

Competition Win Stands: The Colorado State Fair officials confirmed that Allen's submission adhered to the rules of the competition, and his win was upheld. They noted that he disclosed the use of AI, and even if the judges were initially unaware of what Midjourney was, they stated they would have awarded the prize regardless based on the artistic merit of the piece. The fair has since indicated they might create a separate category for AI art in the future.

US Copyright Office Refusal: This is where the deeper, long-term impact of the controversy comes into play. Jason Allen subsequently attempted to copyright "Théâtre D'opéra Spatial" with the U.S. Copyright Office.

The Ruling: The U.S. Copyright Office rejected his application for copyright protection. Their reasoning was that works generated solely by AI, or where the human contribution is deemed insufficient to constitute "human authorship," are not eligible for copyright. They found that while Allen had some input (prompts, selections, post-processing), the core "traditional elements of authorship" were determined and executed by the Midjourney AI.

Allen's Appeal: Allen has been appealing this decision multiple times, arguing that his "prompt engineering" and subsequent refinements represent sufficient human creativity to warrant copyright. He contends that denying copyright creates a "void of ownership" and hinders artists using these new tools.

Ongoing Legal Battle: As of late 2024 / early 2025, Allen has reportedly filed a lawsuit in federal court against the U.S. Copyright Office to challenge their ruling. He argues that he has lost "several million dollars" due to people "blatantly and repeatedly stealing" his uncopyrighted AI-generated work.

This specific case encapsulates the core debates: What defines art? Who is the author when AI is involved? How should intellectual property laws adapt to generative AI? The outcome of Allen's lawsuit will be a significant landmark in establishing legal precedents for AI-generated works and their place in the creative landscape.

can you give me a specific example of what kind of prompt Allen may have used!   And think of a prompt in this case as the description that Allen gave to the AI to produce the artwork.

Jason Allen hasn't revealed the exact prompt he used for "Théâtre D'opéra Spatial" (he considers it proprietary, an "artistic product" that he might publish later, and argues it's what he can "take ownership of"), we can infer the kind of prompt he likely employed based on the image itself and general Midjourney prompting techniques.

For those of us unfamiliar with the art produced by mid journey in response to various prompts, Allen’s prompts for this piece of art included - and this is a huge oversimplification:

Subject: What is depicted.

Style/Medium: The artistic style (e.g., oil painting, digital art, renaissance), and potentially specific artists to emulate.

Setting/Environment: Where the subject is located.

Composition/Lighting: How the scene is framed, the light source, mood.

Parameters: Specific Midjourney commands for aspect ratio, quality, stylization, etc.

Given "Théâtre D'opéra Spatial" portrays classical figures in a futuristic, Baroque-style opera hall looking out into a sun-drenched sci-fi landscape and a a series of prompts he refined over 624 iterations)

As well as elements that included the words "Grand Baroque opera hall interior, opulent, gilded architecture": which established the primary setting and its lavish, classical style.

"classical figures in flowing, ornate robes": which generated the subjects and their attire, hinting at a historical or fantastical context.

"gazing out a massive circular viewport": which defined the central focal point and the action of the figures.

"vibrant, radiant, futuristic space opera landscape": which [ntroduces the contrasting sci-fi element and the colorful, expansive view outside.

"golden hour lighting, highly detailed, masterpiece, oil painting aesthetic, sci-fi fantasy art, cinematic lighting": These are crucial stylistic and quality enhancers.

"Golden hour lighting" sets the mood and warmth.
"Highly detailed" and "masterpiece" encourage a high level of visual fidelity.

"Oil painting aesthetic" or "Impressionistic painting style" would guide the AI to render it with brushstrokes and textures reminiscent of traditional art.

"Sci-fi fantasy art" blends the two genres.

"Cinematic lighting" suggests dramatic and professional illumination.

--ar 16:9: This is a Midjourney parameter for a widescreen aspect ratio, common for digital art and displays.

--v 4: This would specify the Midjourney model version used at the time (Allen created his in 2022, when version 4 was prevalent).

Allen famously stated he iterated on his prompts hundreds of times with 900 drafts), so it wasn't a single simple prompt. He would have started with something similar to the above, generated many images, then refined keywords, added or removed details, adjusted parameters, and likely focused on specific elements (e.g., "more light on the figures," "more intricate details on the architecture," "less harsh shadows," etc.) until he achieved his desired result. This iterative process of prompt engineering is a significant part of the human creative input in AI art.

So basically it looks like the discussion has gone full circle regarding AI and creativity.

If creativity is defined as: Generating novel and valuable results or ways of looking at things, then yes, AI demonstrably achieves this. It can combine existing elements in new ways, discover hidden patterns, and produce outputs that surprise and delight.

Sources include Gemini AI, the quick guide to prompt engineering by Ian Khan, and Mastering Mid Journey Prompts: The Complete Guide to Creating Stunning AI Art with Precision and Power
By Walden Wilder.

Joint celebrate creativity or episode 410 for an especially humorous episode in which Adele into role-playing and prompts using the fictional example of Edgar Allan Poe managing a company that rents out Black cats.

Thank you for listening to celebrate creativity.