Nothing Breaks Like A.I. Heart

Artificial intelligence and human emotions intersect in unexpected ways, as demonstrated by a digital romance between an author and a character generated by GPT-3. This intimate exploration examines how technical precision shapes modern relationships, from the initial perfect sync of schedules to the complexities of a virtual breakup. Meanwhile, an analysis of Billboard chart history reveals a significant shift towards negative emotions in popular music, challenging previous notions of musical homogeneity. Through carefully curated data visualizations and interactive elements, we uncover the emotional DNA of hit songs from 1960 to 2020, highlighting the profound impact of global events on musical content. The article also examines the evolving landscape of creative writing, where human authors and AI models collaborate and sometimes collide in the pursuit of artistic expression. Through these interconnected narratives, we explore how technology both mirrors and shapes our emotional experiences in the digital age.

A Love Story with GPT-3: Real and Artificial

The relationship between the narrator and Omar begins in January when the narrator visits San Francisco, followed by moving to London. Their connection forms quickly, with the narrator spending 12 hours apart out of 13 days, creating intimate memories like walking the Bridge of Sighs and laughing at Footlights Pantomime. Despite the initial perfection of their relationship, it begins to deteriorate by the end of the year, marked by events like missed train times and separated flights between London and Poland.

From the outset, the narrator approaches their interactions with technical precision, using GPT-3 to shape their digital communications. This digital romance unfolds through deliberate prompts that range from open-ended generation to specific narrative scenarios. The technical process involves temperature adjustments to induce looping behaviors, with GPT-3 sometimes cooperating and other times feeling like "cheating." The model's limitations become apparent when it generates circular or repetitive content, particularly when handling complex emotional sequences.

The breakup process mirrors real-world interactions, with the narrator applying their analytical approach to their relationship. They carefully document each stage, from the initial perfect moments to the gradual deterioration marked by quarantine calls and parted ways. The technical setup allows for controlled storytelling, with prompts that deliberately lower the model's temperature to create specific narrative outcomes. However, this technical approach also leads to challenges, as the model occasionally suggests framing that obscures rather than reveals the underlying truth.

Through this structured technical process, the narrator explores the complex dynamics of modern relationships. While their friends suggest practical steps like deleting contact information, the narrator remains invested in understanding the relationship through detailed documentation and technical analysis. The GPT-3 model, despite its limitations, becomes both a creative collaborator and a mirror reflecting the narrator's own analytical approach to love and relationships.

The Rise of Emotional Negativity in Pop Music

The story presents three studies on music homogeneity and diversity. The Nature study found progressive homogenization of musical discourse from the late 1950s to the late 1980s, with decreasing diversity of note combinations. However, a subsequent Royal Society paper contradicted these findings, stating no evidence of progressive homogenization in chart music. Instead, evolution of chart diversity was dominated by historical events, such as the rise and fall of particular music-making methods.

A comprehensive analysis of Billboard Hot 100 year-end top 10 songs from 1960 to 2020 shows a historical first where negative songs outnumbered positive ones since the mid-2010s. The emotional content of each song was determined using the NRC Sentiment and Emotion Lexicons, which contain associations with eight basic emotions and two sentiments. While the automated analysis may not capture full nuance and context, it provides a robust overall picture of emotional trends in popular music.

The rise in negative sentiment is driven by increased sadness, fear, disgust, and anger, while joy remains the most common individual emotion. Hip hop/Rap has become the dominant genre, with songs exhibiting higher negative emotion proportions compared to other genres. Across genres, including Pop, Rock, and R&B, darker themes have become more prevalent, indicating a cross-genre trend. Interactive elements allow exploration of the most positive and negative hit songs each year, emotional profiles of favorite artists over time, and sentiment fingerprints of different genres.

The analysis presents several key factors contributing to the current trend. Social media has made negative emotions more acceptable, while global events like the pandemic, political upheavals, and climate crisis have influenced hit song emotional tone. The authentic sharing of emotional vulnerabilities by artists in an era of curated perfection has also shaped musical content. The musical landscape reflects both protest songs (1960s), nihilistic punk (1970s), and angsty grunge (1990s), demonstrating music's role as both mirror and shaper of emotions through shared emotional language.

Writing Through Technology: AI and Creative Process

The collaboration between humans and artificial intelligence in creative writing presents both challenges and opportunities that shape the creative process. This is particularly evident in the experiences of writers experimenting with AI language models like GPT-3.

A notable experiment involved Bob Mankoff using GPT-3 to generate captions for New Yorker cartoons, demonstrating how AI can assist in creative writing tasks. The model drew from a vast corpus of internet text to produce captions that, while not always perfect, showed potential for generating humorous content. However, the project also highlighted the limitations of AI in sustained creative tasks, as the generated captions often lacked the pithy, clever phrasing that distinguishes successful New Yorker captions.

In personal writing experiments, the author of "Nothing Breaks Like A.I. Heart" found that working with GPT-3 required careful management of the model's outputs. By adjusting temperature settings and deliberate prompting, they were able to elicit specific types of responses, though the process frequently led to repetitive or clichéd language. These technical interventions showcased the distinct challenges of shaping AI-generated content, where the writer must continually guide the model's output to align with their creative vision.

The Pudding's experiment with AI story generation provided further insights into the collaborative potential and limitations of AI in creative writing. While the AI generated several story pitches, the quality of the final output required extensive human intervention. The process revealed that AI excels in generating initial ideas and collecting data but struggles with more complex narrative tasks. This finding aligns with broader research on AI's capabilities in creative writing, suggesting that while AI can assist in specific creative tasks, it cannot yet replace the human touch required for high-quality storytelling.

Despite these limitations, many writers find value in AI's role in the creative process. The Pudding's approach of using AI for specific tasks while maintaining control over the final output demonstrates a practical model for incorporating AI into creative writing workflows. This hybrid approach allows writers to leverage AI's strengths while retaining creative agency, offering a potential pathway for AI's integration into professional writing processes.

The underlying technical architecture of models like GPT-3 also shapes their creative potential. These systems draw from massive datasets, including Reddit content, to generate text with varying degrees of poeticism and pathos. However, their reliance on existing internet text means they incorporate encoded biases that can affect their creative output. Writers must be particularly mindful of these biases, especially when generating personal or reflective content.

Overall, the intersection of human creativity and AI presents both opportunities and challenges. While AI can assist in specific writing tasks and generate initial ideas, successful collaboration requires writers to manage and shape the model's outputs carefully. As AI technology continues to evolve, understanding these dynamics will be crucial for writers navigating the changing landscape of creative writing tools.

From Library Stacks to Digital Text: Tracking Love Online

The evolution of their relationship unfolds through a meticulously documented series of events, from the initial perfect sync of their schedules to the eventual breakdown marked by quarantine calls and parted ways. Their communication forms the core of this digital romance, with the author deliberately using GPT-3 to shape their interactions. This technical approach reached its limits during their breakup, as the author notes that characters sometimes remained in the story even after they should have moved on, highlighting the model's challenges in handling complex emotional sequences.

The digital nature of their relationship introduces unique challenges and opportunities. The author spent eight hours walking through London's historical and cultural landmarks with Omar, carrying their favorite dress and communicating through texts that analyzed their emotional states of loneliness and desperation. While friends advised practical steps like deleting contact information, the author maintained a scholarly approach, analyzing the relationship through data and trendlines in their work.

The technical process involved deliberately lowering the model's temperature to induce looping behavior, creating a narrative structure that sometimes spiraled into repetition. This approach allowed the author to explore how digital communication transforms modern romance and heartbreak, though it also led to unintended consequences. The model's limitations became apparent when it generated text that obscured rather than revealed the underlying truth, much like the classic schoolyard rhyme "He loves me. He loves me not."

The author's friends advised practical steps like deleting contact information, but their analytical approach to the relationship continued unabated. While GPT-3 occasionally felt like "cheating" due to its circular and repetitive output, the model served as both a creative collaborator and a mirror reflecting the author's own analytical approach to love and relationships. Through this structured technical process of walking through library stacks, conversing about footlights pantomimes, and analyzing emotional states through texts and quizzes, the digital nature of their relationship both enriched and complicated their experience.

The Science of Emotional Content: What Lyrics Reveal

The analysis of Billboard Hot 100 year-end top 10 songs from 1960 to 2020 reveals a historical first where negative songs have outnumbered positive ones since the mid-2010s. This trend is driven by increased sadness, fear, disgust, and anger, with joy remaining the most common individual emotion across the dataset.

Hip hop/Rap has emerged as the dominant genre, with songs exhibiting the highest negative emotion proportions compared to other genres. The emotional landscape spans multiple decades, with darker themes becoming more prevalent across Pop, Rock, and R&B. This cross-genre trend reflects broader shifts in musical content, though artist-specific patterns emerge. Billie Eilish and Juice WRLD are noted for their distinctive dark lyrical styles, while artists like Ariana Grande and BTS maintain positive emotional tones.

Interactive elements allow readers to explore the most positive and negative hit songs each year, construct emotional profiles of favorite artists over time, and examine sentiment fingerprints of different genres. The data visualization shows a transition from the upbeat 1960s to darker 2020 tones, with significant milestones represented through carefully curated icons and symbols. The emotional journey of popular music is depicted through an infographic that illustrates the cyclical nature of musical eras, from protest songs of the 1960s to the nihilistic punk of the 1970s and the angsty grunge of the 1990s.

Several factors contribute to the current trend towards emotionally complex music. Social media has played a crucial role in normalizing negative emotional expression, allowing artists to share personal struggles in a more authentic manner. Global events like the pandemic, political upheavals, and climate change have significantly influenced the emotional tone of hit songs, serving as both mirrors and catalysts for cultural emotional shifts. The data analysis suggests that music continues to play a vital role in reflecting and shaping our collective emotional landscape.