Latent Potential

What if our engagement with artificial intelligence unfolded not through the familiar, extractive binary of prompt and response, but through spatial navigation — a dimensional wandering across latent space? Latent Potential is a research project that emerges from this question. It presents speculative navigation objects intended to radically rethink our relationship to AI’s compressed mathematical terrain, moving from comfortable autopilot toward explorative journeys into the remote wilderness of existing and emerging latent spaces.

When complex visual data enters the training sets of generative AI, it is compressed into essential features within what is termed latent space — an abstract, mathematical domain where vast amounts of texts and images are transformed into lower dimensional representations. In this space, vectors map data points across multiple dimensions, capturing patterns, structures, and relationships embedded in training data. Latent space is not a static archive built for retrieval; rather, it is a continuously generative and metamorphic territory for interpolation and transformation. Here, the images and patterns we encounter are not direct traces of human intention or cultural memory, but the emergent products of mathematical operations. Our primary interface to this near infinite dimensionality remains the prompt — a crude mechanism that limits genuine exploration.

What if, instead, we could move beyond the prompt? What if new strategies for navigation allowed us to probe remote areas of latent space and even influence the formation of emerging ones?

Latent Potential proposes speculative navigation devices as alternative interfaces for wandering, rather than prompting, latent space. How can this space be imagined in biomorphic terms, suggesting not only artificial intelligence but also artificial embodiment? These devices — responsive to pressure, gesture, temperature, or biometric input — could translate embodied or sensory actions into multidimensional movement through uncharted areas between training data points. Far from mere retrieval tools, they are designed to reveal emergent aesthetic territories and what Antonio Somaini calls "synthetic" forms — configurations that could not have been generated by cultural convention alone.

Functioning as both critical artefacts and functional speculations, these devices materialise alternative engagement models with AI, challenging familiar prompt-response logic and speculating on new forms of human–AI collaboration. The project is particularly intrigued by the emergence of an “alien" realism — aesthetics that arise uniquely from algorithmic processes and defy traditional cultural references.

What aesthetic knowledge becomes possible in the spaces between our cultural references? What if partnerships with AI were oriented toward situated, open-ended discovery, not just optimisation or extraction? By making tangible the navigational potential of latent worlds, Latent Potential is an invitation for experiments in aesthetic literacy, collaborative agency and speculative re-imagination within an ecology of machine learning that is as much metabolic as it is computational.