Forget the steady ground – reality floats on a dynamic ecological river of evolution and potential. This intricate “tangled weave” isn’t static, but alive with possibility. This dynamic dance, this shifting reality, invites transformative adaptation, a defining characteristic of art: it offers access to unforeseen potential, serves as a resonant mosaic of inspiration, and holds the power to transform.
AI are artifacts of potentiality. Transcending mere mechanisms, these AI-powered generative systems weave a rich tapestry of information exchange, acting as both vessels of insight and catalysts for future possibilities.
The Adjacent Possible in Evolutionary Dynamics
Theoretical biologist, Stuart Kaufmann’s concept of the “adjacent possible” offers a framework for understanding the evolutionary process as a dynamic exploration of what is potentially achievable within a given system. In the biological context, the adjacent possible represents the set of all feasible evolutionary trajectories that a biological system can take from its current state.
The concept of the adjacent possible invites parallels with the latent spaces of generative art. The artist, as image alchemist, sifts through the adjacent possible (latent spaces) to bring forth novel and meaningful creations, drawing from the expansive artistic possibilities that always exists — in a nascent state.
Latent space isn’t simply a code for “unexpected” outputs. Think of it like a compressed map of the data the model learned (e.g., animal pictures). Each point on the map holds key features (like fur, color, leg shape, etc). Moving these points lets you generate new animals based on those learned features. Moving these points lets you generate entirely new forms based on this nuanced lattice of visual memory. It’s not a perfect replica of the real world, but it unlocks diverse possibilities within the model’s knowledge.
Or imagine a kaleidoscope, swirling with fragments of visual information—objects, scenes, styles—drawn from a vast and diverse source. Each fragment whispers stories, emotions, and ideas learned from countless images. This is the essence of AI-generated images: not mere copies, but tools that capture and amplify the knowledge embedded within their training data.
Weaved by neural networks, these tools become living archives of visual memories. They remember not just shapes and colors, but the deeper associations, emotions, and cultural nuances found within massive datasets. These memories are multi-layered and interconnected, echoing the delicate details and diverse perspectives present in the training data.
The dynamic interplay between the past and the future is a defining feature of AI image tools. Informed by historical datasets, these memory fields project forward by generating images that transcend mere replication and enter the realm of creative synthesis. This forward projection becomes a collaborative act between the AI and the human creator, as the tool’s capacity to inform is harnessed to explore new aesthetic territories, challenge artistic conventions, prompting new visual narratives.
Generative art is a means to make the artist less a creator and more a composer. More a seer who relies on an intuitive approach to a palette of compositional material than a maker marching out to create what is an individuated remarkable marking. Generative art functions as the role of editorial observation angling a quotational stacks and the selection of results from a uncountable stack of possibles.
Maybe the rise of generative art has come to our particular age not only because our tools allow for such encyclopedic and potent analytical and compositional creations but because we are the ancients, we are old recordists, the archivists, the hoarders of knowledge who have brought us to this render.
“Evolution is a process of creating patterns of increasing order. I believe that it’s the evolution of patterns that constitutes the ultimate story of our world.” (Kurzweil, 1999)
Parallels
Selective Innovation and Co-evolutionary strategies:
- Evolutionary processes involve the selective innovation of traits that confer a survival advantage.
- Similarly, in AI image creation, the artist-curator selects and refines the generated images to produce aesthetically pleasing and conceptually resonant creations. Generative models don’t operate in isolation. They learn and adapt based on the information they receive, be it training data or some other manner of feedback. This creates a dynamic co-evolutionary process, where both the model and the artist exchange information, potentially leading to the emergence of novel information patterns and aesthetic sensibilities.
AI-generated images can be seen as sophisticated tools that encapsulate and radically extend the information embedded in their training data. These tools draw upon vast datasets, meticulously remembering patterns and features, and effectively encoding knowledge from diverse sources. By doing so, they act as repositories of information, embodying the collective tropes and subtleties present in their training data. This data is granular, and its interpretation is influenced by both syntagmatic (sequential) and paradigmatic (systematic) factors.
Emergence of Novelty:
- The core essence of both evolutionary dynamics and AI generative image creation lies in the emergence of novelty. Biological evolution in exploitation of adjacent possibilities brings forth new species and adaptations, while AI latent space exploration generates novel images and artistic expressions.
Conclusion:
While conventional tools may retain the imprints of wear and tear from repeated use, AI image tools remember not only the patterns inherent in their training data but also learn from the iterative feedback provided by human creators. This adaptive memory allows them to refine and evolve their output over time, mirroring the continuous learning and improvement characteristic of intelligent, evolutionary systems.
References:
Pascal, Blaise. Pensées de Pascal. Bibliothèque Nationale de France, http://www.penseesdepascal.fr/XXIII/XXIII12-approfondir.php: Accessed February 10, 2024.
Huxley, T. H. (1854). On the educational value of the natural history sciences. The Westminster Review, 61(230), 1-47.
Wilden, Anthony, System and Structure (Second Edition), London: Tavistock Publications, 1980. “Tools are artifacts, but they are not in essence objects. Since they qualitatively increase a species possibility of organizing and controlling the matter–energy in their ecosystem, their primary characteristic is that of information. They are forms which inform; they are informed because they remember the past and make possible new types of projection into the future.”
Kauffman, S. A. (1993). The origins of order: Self-organization and selection in evolution. Oxford University Press.
Ray Kurzweil (Kurzweil, 1999, p. 64) The Age of Spiritual Machines: When Computers Exceed Human Intelligence. Penguin Books