Tom White, an artist whose work resides within the domain of computational abstraction and neural network-generated art, offers an instructive glimpse into contemporary practices that merge artificial intelligence and visual aesthetics. His collection, ‘Synthetic Abstractions’, serves as a critical extension of his ongoing exploration into the complex relationship between human perception and machine cognition.

White’s methodology is consistent and expansive, often employing machine learning algorithms as integral tools to distill visual forms into essential abstractions. His process reveals a distinct interplay between the organic intuition of human vision and the calculated precision of computational output. The resulting works are characteristically minimalistic, subtly suggesting recognizable forms without explicitly defining them. This interaction places White’s pieces within a broader discourse about perceptual ambiguity and computational interpretation.

The ‘Synthetic Abstractions’ series specifically explores how neural networks decode and recode visual stimuli, effectively creating compositions that oscillate between the familiar and the abstract. Observers are confronted with shapes that, while resonant of real-world objects or organisms, resist conclusive interpretation. This tension emphasizes the interpretative variability inherent in human perception when mediated through artificial intelligence systems.

Previous work

White’s earlier collections similarly embody his experimental ethos. In ‘Perception Engines’, the artist investigates how neural networks visualize everyday objects, rendering them into simplified geometric and abstracted forms. The collection presents a systematic deconstruction of recognizable items into forms barely identifiable, interrogating the threshold of recognition and abstraction. This collection situates itself in a dialogue with cognitive science and visual psychology, implicitly questioning the neural underpinnings of human perception.

In the collection ‘Mechanical Eye’, White introduced an interactive series: each piece employs a custom text-to-image pipeline, wherein the viewer's mouse movements dictate the gaze direction of a singular, centrally placed eye. This interactivity transforms the observer into an active participant, changing the relationship between creator and audience. Each artwork in this series is generated using a unique neural network, trained on specific title texts provided by White – ensuring that every piece possesses its own visual language and behavioral nuances. ‘Mechanical Eye’ attests to White's exploration of machine perception and the dynamics of viewer engagement, asking questions about the evolving relationship between humans and intelligent systems.

With ‘Sparse Latents’ the artist went on a cartographic exploration of interpretable latent features within large language models. Utilizing the Gemma Scope – a suite of open sparse autoencoders trained on Google's Gemma 2 models – White investigated twelve features within the Gemma 2 2B model. By clustering and rendering thousands of prompts that maximally activate these features, he visualized the internal representations of the model, offering insights into how specific concepts are encoded within its architecture. This approach aligns with contemporary efforts in mechanistic interpretability, aiming to demystify the decision-making processes of complex AI systems.

Moreover, his ‘Pixray Genesis’ project explicitly broadened participation in generative art-making. Through the Pixray interface, users can create their own neural network-generated imagery and it allows them to mint their outputs on the Tezos blockchain directly. That way, White has democratized generative art and blurs distinctions between artist, tool, and audience. This approach marks a departure from traditional conceptions of authorship and the concept of “originality”, now situating art-making within a shared and distributed digital environment. Pixray is accessible through the web interface pixray.gob.io and serves as an approachable introduction for participants new to generative art – which underscores White’s interest in inclusive digital creativity.

Further examples of White’s collaborative and experimental approach are visible on Replicate, an online platform facilitating accessible neural network experimentation. Through platforms like Replicate, White extends his role beyond creator to facilitator, enabling other artists and technologists to engage directly with computational techniques. This stance positions him within an emerging context of artistic practice deeply intertwined with technological inquiry and collaborative experimentation.

Synthetic Abstractions

White’s ‘Synthetic Abstractions’ continues to invite critical reflection rather than the idea of passive appreciation. The works act as conduits for contemplation, encouraging viewers to scrutinize how perception itself is reshaped and reframed through computational processes. They aim to subtly question how far the boundary between human cognition and algorithmic interpretation can stretch, before dissolving entirely.

Computational perception

The underlying themes across White’s oeuvre consistently return to the notion of perceptual duality – ambiguity and clarity, familiarity and abstraction, human and machine. His work suggests an implicit dialogue with historical abstraction movements while explicitly situating itself within contemporary discussions surrounding artificial intelligence, blockchain art platforms, and digital participatory practices.

Thus, Tom White’s works can be seen as case studies in contemporary generative practices. The implications of his projects reach further than aesthetic considerations, resonating instead with broader cultural, philosophical, and technological contexts – which we see happening now and that continue to redefine the nature of artistic creation, and also the human perception in the digital age.

Tom White’s ‘Synthetic Abstractions’

Electric Chronicles

Tom White’s ‘Synthetic Abstractions’

V.Ruins’ Dark Tales

Electric Chronicles