The artist is not the tool.

A tool is not an author

An artist is not defined by their tool, but by how they channel and master it to make tangible what has not yet been created or thought. A tool can extend or redirect what is possible. It does not remove the decisions that give the work meaning: what to keep, what to reject, what deserves attention and who should be part of the process.

The questions sit before and after generation

When a system produces an image, paragraph or sequence, I want to know what material shaped it, who agreed to that use, what context remains, who chooses the result and who benefits when it circulates. A fluent output can still come from a process that ignores the people whose work made it possible.

Culture is not a pile of files

An artwork or record carries more than its visible surface. It has a maker, a history, a situation, a set of choices and sometimes relationships or boundaries that cannot be reconstructed from the file alone. The culture essay alongside this one starts from the same point: what we pass between people is larger than what a system can collect.

Model collapse is a narrower claim

A Nature study on recursive training describes how repeated training on model-generated data can lose information from the tails of an underlying distribution over successive generations. That is evidence about a particular recursive-data problem. It does not show that using artists’ work alone causes model collapse, and it does not prove that artists are the only source of variation. The narrower finding still matters when we ask what kind of human-made material remains available to future systems.

Sources: AI models collapse when trained on recursively generated data ↗

Rights and value are still being worked out

A 2025 EUIPO study on generative AI and EU copyright law examines the use of content during model development and the generation of outputs. It describes different approaches for rights holders rather than a single settled solution. For artists, that uncertainty is practical. It affects what can be used, what can be reserved, how work is credited and whether licensing can become a real option.

Sources: Development of Generative Artificial Intelligence from a Copyright Perspective ↗

Why I am building in this space

Countercult is one way I work on the question underneath the AI debate: how can cultural knowledge become easier to find without treating every person, record or relationship as extractable? We are building public discovery alongside protected context and routes into paid work. This does not solve AI training rights. It gives me a place to test what artist-led cultural infrastructure can do.

The frontier I want to be part of

I want to be in this conversation as an artist and founder who builds, not only as a commentator on what technology might do. My interest is in tools that extend creative work while keeping authorship, context and agency close to the people who make culture. I want to work with people asking how AI can support artists without turning their ideas and relationships into uncredited raw material.