Podcast Summary
Exploring Unique and Personal AI Art: Artist Holly Herndon uses AI to create uncanny, strange, and personal art, pushing against generic and mimicking systems, and manipulating their understanding of her appearance.
Successful AI art lies in its ability to be unique and personal rather than generic and mimicking. Artist and musician Holly Herndon showcases this by using AI to create uncanny, strange, and more personal art, going against the grain of AI systems that often regurgitate familiar content. Her collaboration with an AI voice trained on her voice and others in her 2019 album Proto is a prime example. Herndon also explores the boundaries of AI systems by manipulating their understanding of her appearance, potentially "poisoning" their perception of her. Beyond her artistic endeavors, Herndon and her collaborators are also blazing a trail in fair AI economics and ethics. Her diverse background, rooted in church choirs, Berlin techno, and the Bay Area's technology scene, has shaped her into a trailblazer in the realm of AI art.
Techno music as a coordination technology creating human experience: Techno music transcends artificiality to create a human experience through dancing and collaboration with technology, exploring the intersection of technology and humanity.
Techno music, despite sounding artificial and machine-like, creates a profoundly human experience through its function as a coordination technology. This is evident in the physical and communal act of dancing to techno music. The appeal of creating music in this way lies in the desire to understand and collaborate with the technological systems around us, transforming seemingly inhuman tools into an embodied and expressive medium. The creation of music like Spawn, which combines high-fidelity recordings with lo-fi generated sounds, presents unique challenges but also offers opportunities to explore the intersection of technology and humanity. Through this process, we can better understand our place in the feedback loop of our technologically mediated world.
Innovative voice model Spawn trained on ensemble voices using timbre transfer: Spawn, an innovative voice model, used timbre transfer to learn logic of ensemble voices, adding value to research process. Modern models like Holly Plus surpass Spawn's quality but serve as reminder to engage with tech for informed work. Potential uses include generating questions, creating op-eds, and more.
Spawn was an innovative voice model created during a pivotal period in machine learning research. It was trained on the voices of an ensemble using a technique called timbre transfer, which allowed the computer to learn the logic of another's voice. Spawn's unique timbral quality, a snapshot in time, added value to the research process for the creator, who wanted to have a deep understanding of the technology. Although modern voice models like Holly Plus have surpassed Spawn's unearthly quality, they serve as a reminder of the importance of dealing directly with technology to create informed work. The potential use cases for these models, such as generating questions or creating op-eds, are vast and can be seen as an extension, partner, or even a scalable version of the original creator.
Exploring musical possibilities with AI-generated voice: Holly Herndon uses AI-generated voice as an extension of herself for new musical creations, but acknowledges the absence of emotional connection and meaning in AI interactions.
For artist Holly Herndon, her relationship with her AI-generated voice, Holly Plus, is an extension and augmentation of herself. She sees it as a tool to explore new musical possibilities and even perform complex pieces that she couldn't accomplish on her own. However, she also acknowledges that the meaning and emotional connection that comes with human interaction is something that AI currently cannot replicate. Her experiences with creating AI friends and therapists have left her feeling that the absence of their being and the choice to engage with them robs the interaction of meaning. In the context of her music, she finds that while the AI-generated voice can mimic human-like output, it lacks the depth and emotional resonance that comes with a human performer. This raises questions about the role of meaning in AI-generated art and the limitations of current technology in replicating the human experience.
Seeing AI as a collaborative tool for art: Artist Grimes emphasizes the importance of human relationships and emotions in her AI-generated music album, viewing AI as a collaborative tool for audiences to explore and engage with her art, rather than a replacement.
The use of AI in art should be seen as an extension of human creativity and collaboration rather than a replacement. The artist Grimes, in her discussion, emphasized the importance of human relationships and emotions in her AI-generated music album. She is more interested in the potential of AI as a collaborative tool for audiences to explore and engage with her art, rather than an AI therapist or therapist replacement. Grimes also highlighted the idea of "protocol art," where the work becomes a collaborative experience between the artist and the audience. She believes that AI is an aggregate human intelligence, trained on all of us, and sees it as a part of our evolutionary story. By emphasizing the collectivity of these models, we can celebrate AI as a beautiful and societal output, rather than an individual technological achievement. This perspective also raises economic questions about who gets compensated and how the training data is used economically. Overall, Grimes' work challenges us to reframe AI as a collaborative and societal output, rather than an individual achievement or replacement for human creativity.
Navigating the Future of AI and Data Manners: Exploring the concept of 'spawning' in AI, finding balance between open use and IP protection, and understanding data manners in the AI era.
As we navigate the future of AI and its impact on the economy, it's essential to recognize the continuum between old and new technologies, and find a middle ground that allows for creativity, experimentation, and compensation. The concept of "spawning," a 21st century corollary to sampling, is distinctly different in what it can do and how it comes about. While sampling is a one-to-one reproduction, spawning involves machines learning from media, creating a gray area in terms of intellectual property. The organization Spawning was founded to explore data manners and handle the messy question of AI training. Experiments like Holly Plus, where people used the creator's voice and generated profits, and recording choirs for a data trust, are promising steps towards finding a solution that works for everyone. The goal is to strike a balance between open use and strict IP lockdown, and to understand the implications of handling data manners in the AI era.
Control and agency over personal data in AI age: People should have the ability to opt out of having their data used for AI training and new legislation and tools are emerging to help facilitate this.
Control and agency over personal data are becoming increasingly important in the age of artificial intelligence and big data. The discussion highlights the case of authors whose works were used without permission for AI training, raising questions about ownership and control. The speaker argues that people should have the ability to opt out of having their data used in such ways and that this is a developing trend. The EU AI Act provides precedent for this, and tools are being developed to help individuals easily opt out of having their data included in models. The speaker also suggests that an economy could be built around opting in for fine-tuning models. The conversation underscores the need for individuals to have more control over their data and the potential for legislation and technology to help facilitate this.
Exploring new opportunities for artists and creators: Collective intelligence and open source models offer new ways for artists to make a living and share ideas, but it's important to consider unique economic needs of different artistic practices.
The idea of collective intelligence and open source models can provide new opportunities for artists and creators to make a living, while also allowing for the sharing and accessibility of ideas. However, it's important to recognize that there is no one-size-fits-all solution, as different artistic practices have unique economic needs. The concept of "All Media is Training Data" can be triggering for some, as it raises concerns about the potential replacement of human creators by AI. However, this idea can also be seen as an opportunity to intentionally create training data as artworks and to explore new economic models that allow for the coexistence of humans and AI in the creative process.
The Role of Human Creativity in a World of AI-Generated Content: AI may generate content, but it can't replace human creativity or meaning-making. The relationship between artists and AI is complex and evolving, with risks of sameness and stagnation, but also opportunities for new forms of expression.
While AI may generate an endless amount of content, it won't replace artists or human creativity. Instead, it adds complexity to the cultural conversation and dialogue. The meaning-making process becomes even more important in a world where infinite media exists. However, there is a risk of a backlash against the sameness that AI can produce, leading to a desire for differentiation and human-created content. As AI becomes more personalized, it may cater to specific tastes, but there is also a risk of a stagnation of taste if people are constantly being catered to. Artists can use the relationship with generative systems to make their work stranger and more refreshing, rather than being flattened by AI's lowest common denominator. The backlash against AI in the cultural sphere is already happening, but we are still in the early days, and the future direction of this relationship is uncertain.
Exploring unique and weird AI sounds: Artist uses custom training data to create unusual AI vocal expressions and sounds, embracing imperfections and pushing towards uniqueness
The artist is using custom training data to create unique and weird AI models, rather than relying on large public models that aim for a more average or "sanded down" output. This approach allows for the exploration of unusual vocal expressions and sounds that neither humans nor machines could create alone. The artist's use of their own voice and collaborator's stems in training their AI model, as demonstrated in the song "Godmother," results in a unique and modern-sounding output that embraces the weirdness and imperfections of AI. This contrasts with the trend of making AI seem more normal and polished, which the artist finds less appealing. The artist's focus on the raw, unpolished aspects of AI and its ability to create strange and beautiful sounds sets their work apart. Overall, the artist's approach to AI creation is a push towards uniqueness and weirdness, rather than sameness and averageness.
Exploring AI: Resources for Individuals: Interact with public models, read research papers, engage with philosophy, and collaborate on projects to learn about AI.
Despite the vast resources and expertise required to create advanced AI models, there are ways for individuals to engage with and learn about the field. Holly Herndon, an artist and researcher in the space, suggests starting by interacting with publicly available models and reading academic research papers. She also recommends exploring philosophical books like "Intelligence and Spirit" by Reza Negarastani, science fiction novels such as "Children of Time" by Adrian Tchaikovsky, and collaborative works like "Plurality," which was written in an open, democratic way. Herndon emphasizes that while the learning process can be messy and challenging, the resources and information are available for those who are interested in rolling up their sleeves and getting involved.