SoTA

State-of-the-Art (SoTA) refers to cutting-edge AI models, analogous to achieving an artistic pinnacle. Early SoTA architectures (CNN/RNN) were designed to mimic human cognition, echoing the representational approach. Yet, as Rich Sutton’s “Bitter Lesson” argues, ever-growing GPU power now eclipses human-designed intricacies, nudging SoTA from carefully crafted forms toward vast, computation-driven abstractions.

This shift parallels how art moved from representation into Modernism’s more abstract modes, especially after photography’s rise. A century ago, Benjamin spotted the profound influence of mechanical reproduction on aesthetics. Decades later, abstract expressionism emerged at the extreme of modernism, epitomised by Pollock’s action paintings and Greenberg’s emphasis on flatness. Likewise, today’s AI reproduction propels neural networks beyond human comprehension—a realm of “AI for AI’s sake”—hintable from Eric Schmidt’s warning.

SoTA is an XAI(Explainable AI)-driven interactive multi-device web artwork that makes our evolving metaphor tangible. Two projectors, four PCs, and audiences’ mobiles present 118 neural architectures in a 3D, interactive, and representational form—initially mirroring SoTA’s earlier pre-modernist era (the Frontend). But once audiences stop interacting and leave the site, unprecedentedly, the screens flip to chaotic monochrome visuals that abstractly depict LLM token cross-similarities (the Backend). Like Pollock’s expressive drips, the swirling lines of connectionism immerse audiences in a vortex of high-dimensional complexity no longer intelligible to us.

Mirroring Modernism’s shift, SoTA reveals how AI can reshape society, identity, and creativity. It questions whether massive models will foster inclusive, human-centred progress or slip into self-referential abstraction—a Black Box. By uncovering AI’s hidden layers, SoTA urges us to reflect on the implications of State-of-the-Artificial Intelligence and responsibly guide these systems.