🧠 When AI Becomes The Artist

Sign Up | AI Toolkit | Meet The Team

Welcome back AI prodigies!

In today’s Sunday Special:

• 📜The Prelude

• 📸How AI Generates Images

• 🥁The Turing Test of Tunes

• 🎨When AI Creates, Who’s the Author?

• 🔑Key Takeaway

Read Time: 7 minutes

🎓Key Terms

• Generative AI (GenAI): Creates entirely new content that replicates human-like creativity.

• Anthropomorphism: Attributing human characteristics to non-human things. For example, we often personify our pets by naming them, assigning emotions to them, and talking to them as if they understand.

🩺 PULSE CHECK

📜THE PRELUDE

Jason M. Allen, a video game designer, spent roughly 80 hours crafting his artwork for the Colorado State Fair’s Fine Arts Competition.

His piece, Théâtre D’opéra Spatial, depicted elegantly dressed classical figures gazing through a glowing cosmic portal, seamlessly blending the grandeur of the past with the mysteries of the future.

At first glance, it seemed rich in artistic intent and full of imagination. The twist?! It was actually created using an AI-powered tool called Midjourney. This realization sent shockwaves through the art community:

It’s no secret that GenAI has earned a seat at the table in many creative fields, including art, music, and poetry. So, how does GenAI learn to paint with such creative flair? Do people like listening to AI-generated music?

📸HOW AI GENERATES IMAGES

⦿ 1️⃣ Pre-Training.

During training, AI-powered tools like Midjourney learn which words describe which images by examining millions of images paired with detailed descriptions.

For example, it might be given an image of a cathedral accompanied by a detailed description: “A large cathedral with tall arches and stained glass windows.”

It examines the pixels within the image and learns to identify which shapes, colors, and textures co-occur with the word “cathedral,” such as:

• Colors: Bright stained glass and stone gray walls.

• Shapes: Tall arches, pointed spires, and curved windows.

• Textures: Smooth glass panels and splintered wooden beams.

This learning enables Midjourney to convert brief text descriptions into detailed images that capture defining colors, shapes, and textures. In other words, it enables Midjourney to accurately connect descriptive language with visual features to generate rich and vivid creations.

Throughout the training process, AI-powered tools like Midjourney don’t just look at perfect images. Instead, images are deliberately corrupted with random noise:

• Static Spots: The patterns of pixels become static.

• Blurry Details: The edges of textures become soft.

• Fuzzy Outlines: The borders of shapes appear hazy.

It learns to reverse this process step by step, predicting what a progressively less noisy version of the image should look like at hundreds of stages until the original image is fully restored. This approach trains Midjourney to recognize, understand, and generate fine details.

⦿ 2️⃣ Post-Training.

After training, AI-powered tools like Midjourney are ready to be used! Here’s what happens after you enter your prompt: “foggy gothic cathedral at sunrise.”

1) Encode Your Text Prompt: Your words are converted into a Vector, a list of numbers that captures the meaning of your prompt, using a Text Encoder. This list of numbers guides Midjourney in generating your image. For example, the concept of “fogginess” is represented by numbers like: “[0.23, -1.5, 0.87,....].”

2) Compare Vectors: Instead of generating an image pixel-by-pixel, AI-powered tools like Midjourney work with simplified Vectors that capture the main colors, shapes, and textures of your prompt to efficiently focus on the overall style and structure before filling in finer details.

3) Generate From Noise: The generative process begins with random noise: a visual of meaningless values. Then, it iteratively removes the random noise over hundreds of steps, gradually moving closer to what the Vector describes. Here’s what that might look like:

• Step 1: Just random speckles.

• Step 10: A blurry silhouette of cathedral towers.

• Step 30: Clear stained-glass windows with misty light.

• Final Step: A fully detailed foggy gothic cathedral at sunrise.

4) Decode Back to Pixels: Once Midjourney finishes generating a compact, blurry version of your prompt made up of numbers, it uses the Decoder part of the Autoencoder to transform it into a standard, high-quality picture. For example, a 64x64 grid of numbers becomes a sharp 1024×1024 pixel image with visible stone gray walls and a glowing sunrise.

5) Refine the Output: Upscalers sharpen details so colors, shapes, and textures look crisp and realistic, while Safety Filters check for broken pixels.

These five steps are tedious, and they produce stunning, vivid creations. In one case, an AI-generated photograph fooled up to 15 judges to win a Sony World Photography Award.

🥁THE TURING TEST OF TUNES

⦿ 3️⃣ What About AI-Generated Music?

Perception plays a crucial role in our response to music. If an artist we don’t like releases a catchy song, we find ourselves irritated if we enjoy it. Similarly, when listeners know a piece of music is AI-generated, they tend to like it less. This perception seems based on pre-held biases, which stem from discomfort with non-human creativity, skepticism about authenticity, or a lack of emotional connection with the artist.

⦿ 4️⃣ Limiting Pre-Held Biases?

Hearing music without knowing who or what composed it can mitigate these pre-held biases, similar to how blind orchestra auditions allow evaluators to appreciate the music’s quality without subconsciously considering the musician’s age, race, or gender.

USC Annenberg found that people enjoyed AI-generated music more when they viewed the AI as a traditional musician. This discovery highlighted that our perception of AI plays a BIG role in how we evaluate what it generates. In other words, how we see AI can heavily influence how much we like what it creates.

They also discovered that people are more likely to accept AI-generated music when it feels familiar, breeding comfort. In this context, when AI-generated music incorporates dynamic range and performance nuance, it makes the AI feel more natural, expressive, and emotionally engaging. Musicians don’t play with perfect pitch. They add subtle variations in timing, volume, and phrasing. AI-generated music can be designed to emulate these imperfections, making the music feel more natural, authentic, and expressive.

🎨WHEN AI CREATES, WHO’S THE AUTHOR?

⦿ 5️⃣ Copyright Protections for AI?

When AI gets creative, who owns the rights? Spoiler: it’s not always clear! The U.S. Copyright Office (USCO) outlines that AI-generated content without human authorship isn’t copyrightable. So, what does that mean?

• 🔴 NOT Copyrightable: You don’t make any creative choices beyond the brief text description you provide the AI.

• 🟡 MAYBE Copyrightable: You guide the AI creatively by combining multiple outputs and making targeted edits.

• 🟢 CLEARLY Copyrightable: You use AI to brainstorm rough drafts. Then you write, draw, or edit extensively yourself.

⦿ 6️⃣ Training on Copyrighted Material?

The U.S. District Court of Delaware (“D. Del.”) issued the first major decision concerning the use of copyrighted materials to train AI-powered tools. Thomson Reuters, the owner of Westlaw, sued ROSS Intelligence for training their AI-powered legal research tool using Westlaw’s headnotes: summaries of the legal points in a court case.

They ruled it was copyright infringement and didn’t fall under fair use through transformative use: adding new expression, meaning, or messaging to the original works.

Crucially, the “D. Del.” noted that the AI-powered legal research tool spit back relevant judicial opinions that were already written and, therefore, wasn’t considered GenAI, which learns patterns from copyrighted materials to generate entirely new content.

In doing so, the court left open the possibility that GenAI might qualify as fair use through transformative use if it truly creates new messaging and doesn’t just regurgitate the original works.

⦿ 7️⃣ Current Legal Challenges?

AI firms argue that GenAI learns patterns from copyrighted materials to generate entirely new content significantly different from the original works. This reasoning might not always be justified, particularly when GenAI directly competes with the original works.

The current ongoing debate is whether GenAI’s new content is transformative enough to avoid copyright infringement or if it’s merely a remix of copyrighted materials.

🔑KEY TAKEAWAY

GenAI collapses the cost of creating art and music. It forces us to rethink what makes human creativity valuable. Legal frameworks must evolve to protect creators when necessary, while also allowing AI-powered tools to democratize creativity for all. Ultimately, we face a choice: treat GenAI as a partner that amplifies human imagination or as a replacement that risks hollowing it out.

📒FINAL NOTE

FEEDBACK

❤️TAIP Review of The Week

REFER & EARN

🎉Your Friends Learn, You Earn!

You currently have 0 referrals, only 1 away from receiving 🎓3 Simple Steps to Turn ChatGPT Into an Instant Expert.

Share your unique referral link: https://theaipulse.beehiiv.com/subscribe?ref=PLACEHOLDER