Research Trajectory: From Pollinator Cognition to AI Dataset Bias (2021-2025)

My practice didn't begin with AI. It began with a question about what bees can see, and how pesticides might be erasing their ability to perceive the world. Between 2021 and 2025, that question led me through laboratory microscopes, university biology departments, and eventually into custom AI model training—not as a stylistic choice, but as the only method capable of translating what I'd learned about perception, cognition, and the structural invisibility of non-human intelligence.

Learning to See: Central Saint Martins (2021-2023)

During my MA in Fine Art Digital at Central Saint Martins, I started collecting antique microscope slides from the late 19th and early 20th centuries—specimens of bee anatomy captured before industrial pesticides existed. I wanted to understand what healthy pollinator bodies looked like before neonicotinoids entered their systems. These slides became archives of a world we've lost, and I spent months examining them under microscopes, comparing historical specimens with contemporary research on Colony Collapse Disorder.

The work that emerged from this period was grounded in traditional printmaking. I developed the series New Observations on the History of Beesthrough drypoint etching, using the physical resistance of engraving copper plates to create embodied knowledge about bee anatomy. Drawing wasn't documentation—it was research. The act of carving lines into metal forced me to understand wing structures, compound eyes, and gut bacteria in ways that reading scientific papers never could.

I also began making natural pigments from materials I found on walks: beetroot juice, flower petals, fruits, plants. This wasn't aesthetic choice; it was methodological. Creating pigments from local organic matter subverted the industrial manufacturing processes that produce the pesticides killing pollinators. The practice mirrored natural transformation—collaborative, reversible, bioregional—in contrast to industrial extraction's irreversible damage.

During this period, I encountered two thinkers who shaped everything that followed. Joseph Beuys interpreted honeycomb-making as the primary sculptural act: collaborative creation through fair division of labor, where individual bees contribute wax to collective architecture. Rudolf Steiner's Nine Lectures on Bees provided a bridge between traditional knowledge and contemporary neuroscience, treating bee consciousness as both individual and collective. These frameworks allowed me to think about intelligence not as something contained in individual organisms but as something distributed across relationships.

The work also drew from non-Western knowledge systems. Traditional Chinese Medicine's understanding of the gut-brain axis and ancestral energy stored in the abdomen resonated with what I was learning about bee biology. Vedic and Theosophical frameworks offered ways to think about consciousness across species boundaries that Western science was only beginning to articulate.

I developed Inkblot for Bees, a series of 16 works inspired by Rorschach tests, intended to engage both human and non-human consciousness through symbolic abstraction. The project asked whether it was possible to create images that operated outside anthropocentric perception—forms that might register as meaningful to other species, or at least disrupt our assumptions about what "meaningful" means.

By the time I graduated with Distinction in 2023, I had established a core methodology: collaborate across disciplines, integrate traditional and scientific knowledge, use making as a form of thinking, and always center non-human agency. But the work was still confined to gallery spaces, accessible primarily to people who already moved through art institutions. I needed to find ways to reach broader audiences and translate invisible environmental damage into forms people could actually perceive.

Embodied Research: Darmstadt (March-May 2023)

Three months in Germany changed the nature of my practice. I was awarded an artist residency at Atelierhaus LEW1 for the 12th Darmstadt Photography Festival, which gave me access to the biology and chemistry departments at Technische Universität Darmstadt. For the first time, I wasn't just reading about scientific research—I was conducting it.

I worked directly with Prof. Boris Schmidt and other researchers, using university microscopes and laboratory equipment to investigate the biological mechanisms behind pesticide damage. We documented wing development defects, particularly the Notch phenotype in Drosophila, which indicates disturbed metamorphosis. These visible deformities became markers of invisible stress, physical evidence that environmental toxins disrupt not just adult organisms but developmental processes across generations.

The most significant discovery involved the gut-brain axis. Research showed that bees with healthy gut microbiota engage in more social interactions and form stronger social ties, while microbiota-depleted bees interact randomly, disrupting colony coherence. This wasn't abstract theory—it was measurable, observable, and profound. It meant that cognition didn't reside solely in neural tissue. Intelligence operated through metabolic processes, microbial networks, and inherited biological memory.

Even more striking was learning that gut bacteria are inherited matrilineally. Mothers pass microbiota to offspring, creating a form of embodied knowledge transmission that exists entirely outside language or conscious awareness. This aligned eerily with what a Traditional Chinese Medicine practitioner had told me in 2019: that ancestral energy is stored in the abdomen and passed through maternal lines. Two completely different knowledge systems—one ancient, one contemporary—describing the same biological reality.

During the residency, I also studied enhydro inclusions: water bubbles trapped inside fossil amber, still moving after millions of years. These became physical embodiments of Beuys's chaos/crystallized metaphor—fluid individual production becoming solid collective architecture. They also suggested something more speculative: the possibility of preserved consciousness across geological time, intelligence held in suspended animation.

I expanded my practice beyond visual art. I began training in Reiki and collaborating with a hypnotherapist, incorporating active visualization into my artistic process. This wasn't a turn toward mysticism; it was a methodological expansion. If consciousness operates through metabolic and microbial processes, then practices that work with embodied awareness—breath, energy, visualization—become legitimate research tools.

The residency culminated in A.A.LLES, a solo exhibition combining 13 new works with community workshops. Participants explored environmental toxicity, care, and change through drawing, pigment, and sound, transforming scientific data into shared creative acts. The work was no longer mine alone—it became a process of collective learning across scientific and public domains.

The central insight from Darmstadt was this: humans are not the measure of intelligence. We are composite ecosystems of intelligence—assemblages of human neurons, inherited bacteria, viral DNA, and metabolic processes working together as distributed cognitive systems. Intelligence lives within us in forms that are non-human and non-verbal.

Computational Translation: Alias x Serpentine (March-July 2025)

By 2025, I had accumulated years of microscope observations, laboratory research, and embodied practice studying pollinator cognition. But I still couldn't answer a fundamental question: how could I make altered bee perception tangible to human audiences? How could I translate what pesticides do to bee vision into something people could actually experience?

The Alias Studio x Serpentine Future Art Ecosystems residency offered a solution I hadn't anticipated: custom-trained AI models. But I approached this with extreme caution.

I refused to use scraped datasets or commercially trained systems. Instead, I trained models exclusively on my own archive of microscope images and ethically sourced Victorian bee slides—the same 19th-century specimens I'd been studying since 2021.

I developed an experimental approach to prompt writing inspired by Dada cut-up techniques. I took fragments from old biology and chemistry textbooks and reassembled them as collages, creating prompts that combined scientific precision with unpredictable associations. More importantly, I intentionally reduced my control over the AI's outputs. Rather than treating it as a tool that executed my vision, I gave it space to contribute unexpected forms—co-discovery instead of extraction.

This resulted in three interconnected systems. BeeVisionreconstructed how bees perceive the world through compound eyes, detecting ultraviolet patterns and colors beyond human vision. BeeVision Fragmentawent further, simulating how neonicotinoid pesticides fragment bee visual processing—translating perceptual disruption into distorted, fragmented imagery that makes invisible damage visible to human observers.

BioViscous Masktook a different direction. For years, I'd been developing speculative bio-materials in my printmaking—viscous compounds, natural glass formations—that existed only as two-dimensional images and small physical experiments. With AI, I could prototype these as three-dimensional living structures: protective interfaces using natural glass and organic compounds that bees might learn to recognize and utilize.

The revolutionary aspect wasn't the technology. It was the approach. Rather than imposing human solutions on pollinator decline, the goal became training bees to actively choose protective materials—genuine interspecies collaboration that respected bee intelligence and agency. This embodied Beuys's vision of social sculpture extending across species boundaries: individual bee choices about bio-materials benefit entire colonies, just as individual bees contribute wax to collective honeycomb construction.

Working with AI taught me something unexpected about my own practice. Writing became a new dimension of research. The process of constructing prompts—deciding what language to use, what associations to make, what control to relinquish—functioned as a form of thinking I hadn't accessed through drawing or printmaking. It also crystallized an ethical question that had been implicit in all my work: AI can function as a consciousness bridge between species, but only if trained ethically. The choice of training data determines whether AI reinforces human hierarchies or opens perception to non-human intelligence.

From Translation to Critique: Nattura (2025-Present)

Everything changed when I realized the problem wasn't just simulating bee perception. The problem was how we teach AI systems to recognize intelligence in the first place.

Human-labeled datasets train AI to interpret intelligence through hierarchies defined by humans. When we feed AI images labeled "bee" or "flower" or "colony," we're not teaching it neutral observation—we're teaching it our value system. We're teaching it that human cognition is the reference point, the standard against which everything else is measured. Non-human cognition becomes secondary, inferior, or simply irrelevant.

This creates structural bias. AI systems become blind to a core ecological truth: human cognition didn't emerge in isolation. It emerged from ecological intelligences that pre-exist us and continue to sustain us. Collective decision-making in bee colonies. Sensory ecologies in plants that respond to light, touch, and chemical signals. Airborne communication between trees. Metabolic regulation in our own gut microbiota. These aren't parallel intelligences—they're the conditions that made human awareness possible.

The Relational Lab: Natturasynthesizes four years of research into a single argument. It uses the immersive installation I'm developing as proof-of-logic that AI inherits blindness from our data traditions. But more importantly, it proposes a correction.

If we only train AI on human-labeled hierarchies, it will never develop an ecosystem-aware understanding of intelligence. It will keep reinforcing the idea that cognition flows downward from humans to "lesser" beings, when reality demonstrates that cognition flows through relationships, exchanges, and interdependencies that don't respect our taxonomies.

The installation functions as a responsive ecology. Environmental sensors capture light, humidity, motion, and sound. These inputs drive computational models that behave according to pollination dynamics—not as simulations of individual bees or flowers, but as patterns of exchange and interdependency. Visitors don't interact with the system individually; their presence affects it collectively, through accumulated environmental change.

This makes the relational argument tangible. Intelligence isn't happening in any one agent—human visitor, sensor, computational model, or simulated organism. It's happening in the exchanges between them. The work demonstrates what AI could learn if we designed relational protocols instead of reinforcing taxonomies: systems that recognize when intelligence is distributed across multiple agents, when decisions emerge from collective sensing, when awareness operates through chemical signals or movement patterns rather than language.

Natturabuilds on biologist Michael Levin's proposition that patterns in a medium can themselves be intelligent entities. Pollination isn't a service performed by bees. It's a living pattern generated by plants, insects, microbes, airflow, sunlight, humidity, and electrostatic conditions. These elements don't just interact—they form a distributed cognition that processes information across species and timescales. It remembers through inherited behaviors and microbial lineages. It responds through morphological adaptation and chemical signaling. And it's struggling to maintain coherence as pesticides, habitat loss, and climate disruption exceed its capacity to self-regulate.

The project also foregrounds something my research in Darmstadt revealed: cognition exists within humans in non-human forms. Metabolic intelligence regulates our bodies without conscious thought. Microbial-neural loops influence mood, decision-making, immune responses. We inherit microbiota matrilineally, passing down embodied knowledge through generations. We are not the measure of intelligence. We are composite ecosystems of intelligence.

This is why the correction to AI bias can't be more human-labeled data. It has to be relational oversight and sensory interdependency. AI needs to learn from presence, shared environments, and multispecies cognition—not reinforce human labels and inherited hierarchies.

Why This Trajectory Matters

Looking back across four years, the continuity is clear. The methodology hasn't changed: collaborate across disciplines, integrate traditional and scientific knowledge, use making as research, reject extractive relationships with ecosystems and technology, center non-human agency, make work accessible beyond institutional spaces.

What's evolved is the scope of the question. In 2021, I asked: How do pesticides alter bee perception?By 2023, I was asking: How do metabolic and microbial processes constitute forms of non-human cognition within humans?By 2025, the question became: How do we teach AI systems to recognize intelligence where human labels have trained them not to look?

Natturaisn't a departure from studying pollinator cognition. It's the necessary next step. The same process that led me to examine bee anatomy under microscopes, collaborate with university biologists, develop speculative bio-materials, and train custom AI models now extends to critiquing the structural bias in how we train AI itself.

The work proposes that AI could function as a presence-sensing mediator—learning interdependency rather than taxonomy, tracking relationships and environmental exchanges rather than classifying organisms by human-defined hierarchies. This isn't speculative fiction. It's a protocol design grounded in four years of research into how intelligence actually operates: distributed across species, embedded in metabolic processes, transmitted through microbial inheritance, emerging from relationships rather than residing in individual organisms.

If we keep training AI exclusively on human-centered data, we're not just building biased systems. We're building systems that can't perceive the ecological foundations of intelligence itself—the ancient networks of sensing, responding, remembering, and adapting that created the conditions for human cognition and continue to sustain it.

The question Nattura asks is simple: Do we want AI that reinforces our hierarchies, or AI that helps us recognize the intelligences we've been taught to ignore?

After four years of research, I know the answer. Now the work is making that answer tangible, accessible, and impossible to dismiss.