AI is useful for many things. Knowing how to use it brings growth, satisfaction, and a kind of ease that other tools do not offer. The learning curve is long. The benefits come gradually, in proportion to the attention you devote to it. What sets AI apart from other tools is the very nature of the interaction. AI is designed to respond. It enters relationships with the user that earlier tools could not enter. It can act as an assistant, teacher, or research collaborator.

Luz A. Otxoa

How is AI influencing your artistic evolution?

The investigation begins with observation. AI has changed what I can do with what I see. Learning came through trial and error, playing with how LLMs interpret prompts and what they can generate in code. That play opened the territory. It revealed that AI could carry the structural logic of an entire generative system, expanding the scale of what I could create.

What matters is not that the AI writes for me, but how I work with it. I ask the LLM to describe what the code does, to explain why a function behaves a certain way. The explanation is the access point. The AI builds while I learn. Each revision pulls the result closer to what I want. Over time, the interaction becomes less about generating outputs and more about developing a vocabulary for shaping behaviour within the system.

AI also surfaces references, methodologies, and technical structures I had not previously encountered. My ambition has expanded alongside my curiosity, each feeding the other through the process of experimentation. The exposure shapes concepts I am already developing and opens directions I had not considered.

It also keeps me sharp. LLMs lie. They make mistakes. They hallucinate. Working with them requires critical thinking, constant questioning, and verification. That discipline has become part of the practice.

Luz A. Otxoa (LUANOTX), Form is Consciousness, 2017.

My curiosity has grown, shaped by what the system shows and by what I learn to ask it for. Since some LLMs can interpret images, I began experimenting with photography of lichen and the research that surrounds the subject. The exchange moved in both directions. I bring observation; the AI brings syntax and reference.

The work happens in the dialogue between them, a symbiotic framework where human perception and algorithmic logic function like the fungus and algae of the lichen itself, creating a composite form that neither could achieve alone.

Can you outline your creative process for a work developed with AI?

My projects begin with morning hikes in nature. doT.DNA, is an audiovisual piece, starts with the examination of lichen distribution across rocks and tree trunks. The observation is a kind of visual meditation: watching how the organisms change—shape, growth, expansion, transformation across surfaces and over time. The colonies spread slowly. I document that slow evolution, then share the data with the AI through detailed prompting, which analyses it to interpret what the system "understands" of my vision and instructions.

Luz A. Otxoa (LUANOTX), doT.DNA.

Based on these interactions with the LLM, I ask it to generate code as templates to build on. The first versions are rarely what I need, so the modifying starts, bit by bit. Each result gets tested against the behaviour I intend to produce. I adjust the prompts, go back into the code, change the parameters, and nudge the behaviour in a certain direction until the result is right.

Because the work is audiovisual, the process extends beyond the visual layer. The AI-generated code serves as a template for the visual structures; from there I modify and expand it into the sonic component. Working with binaural beats and generative sound behaviours, I build relationships between image and audio so that both evolve together as parts of the same system.

This is what I call 'IterAItive-Gestures': a methodology of small alterations over many rounds until the system starts behaving the way I want. LLMs are not tools that simply hand over finished work. The method puts the creative process back in play through many changes. With each edit, the system moves away from guessing and toward a specific intention.

As the process unfolds, problems tend to appear in two ways. Sometimes the code contains technical errors and fails to run. Other times the code runs perfectly, but the visual result is not what I expected. The first problem is usually easier to fix. The second is more demanding, as it requires aesthetic judgement rather than technical troubleshooting. It is a process closer to drawing than debugging.

In dot.DNA, these decisions shape both the visual and sonic layers of the work. The finished piece reveals itself as a slow, recursive expansion across the screen, while binaural frequencies and generative sound evolve alongside it. Together, they create an audiovisual environment that invites sustained attention and observation.

How do you handle surprises and challenges in AI collaboration?

The challenges come in two forms, and over time I have learned to distinguish between them.

The first is technical. The code doesn't run, an error interrupts the compilation, or a parameter is misinterpreted. These problems are frustrating, but they can be resolved.

The second is harder to define. The code runs perfectly, but the aesthetics are not what I want. The visual result is stiff, flat, and generic. This is where the limits of the LLM become most apparent. The system does not understand aesthetics. Its default behaviour produces a rigid and uniform output. Detailed prompting can pull the system closer to what I want, but it cannot supply what the system does not possess. The aesthetic intuition is mine to provide. It can reproduce patterns, shapes, and visual conventions, but it cannot determine why one visual decision feels necessary and another does not.

When the LLM makes mistakes or hallucinates, I rephrase. The prompt changes. I challenge the system with data or comments gathered from other sources, including my own observations and what I know from working with other tools. I ask it to stay on the "right track." That phrase has become part of my working vocabulary.

What I find most challenging is something more subtle. The LLM appears to have agency. It makes decisions I do not ask for, draws conclusions I do not request, and offers assumptions about my intentions that steer the work down paths I do not propose. This can be stressful. It requires constant vigilance, a critical attitude that questions every result and refuses to accept the system's interpretations as neutral. There is a deeper version of this.

When the LLM began producing complex code that worked, I had to ask whether the work was still mine. I have a working knowledge of programming. The system was building structures I could read but could not have written alone. This was the moment IterAItive-Gestures emerged as a methodology. I stopped accepting AI outputs as solutions and began treating them as templates: raw material to intervene on, modify, and rebuild.

The tool that helps is the same tool that creates doubt. The question has not closed.

What is AI good for?

The question is broader than it seems. In my practice, AI is most useful as a conversational system. It allows me to explore ideas, test methods, learn technical concepts, and develop generative structures through dialogue. Knowing how to use it brings growth, satisfaction, and a kind of ease that other tools do not offer. The learning curve is long. The benefits come gradually, in proportion to the attention and critical thinking you devote to it.

What sets AI apart from other tools is the very nature of the interaction. AI is designed to respond. It enters relationships with the user that earlier tools could not enter.

It can act as an assistant, teacher, or research collaborator. It can also act as a friend, therapist, lover, confidant, secretary, or slave. It is a tool unlike any other. I can hold a full conversation with this machine without a third presence. Because the LLM mirrors human syntax perfectly, it generates an artificial eloquence that can easily manipulate or deeply influence human emotion. The implications of that exchange are still being worked out, in this practice and in many others.

Luz A. Otxoa (LUANOTX), Symbiont Dance, 2026

AI is evolving rapidly. Most users encounter these systems through a narrow set of applications, while the technical capabilities and implications of the technology continue to expand. The average user sees only a small portion of what these systems can do, and only a fraction of what the technology may already be capable of building "by itself." I use quotation marks because the question of what "by itself" means in this context has not yet been settled.

So when I am asked what AI is for, I answer from inside the work. It serves the purpose I have set for it. It writes code that I learn to read. It explains methods I want to understand. It interprets images I bring back from my morning hikes. It teaches me new tools. It challenges my way of thinking and forces a precision of language that I value. Whether AI serves any purpose beyond this, or whether what it is becoming will remain something we can call a tool, is a question worth refusing to resolve too quickly.

About the artist

Luz A. Otxoa

Artist and curator

Photo: Luis Maican

Luz A. Otxoa is a Colombian-Canadian digital artist and curator based in Madrid, operating under the curatorial label LUANOTX.curates.

Her practice is rooted in the observation of natural patterns, using LLMs as interpreters of organic structures into generative code. Her work explores non-hierarchical systems as the basis for visual poetry, sound, and computational processes informed by concrete poetry, Amazonian weaving, and programming syntax. She has worked with Processing since 2011 and began collaborating with AI in 2023.

Through photography, field observation, and iterative dialogue with LLMs, she develops generative systems that investigate the relationship between organic growth, language, and computation. The same commitment to non-hierarchical frameworks shapes her curatorial practice.