Interpretability and Research Probes

Bridging Lens logic with AI interpretability and research tool use.

How Large Models Fake Seeing, and What Artists Detect

Large models simulate visual coherence, but miss structural intent. Artists detect pressure and failure where machines default to surface gloss. This paper explains how breakdown, not finish, reveals real visual reasoning. The piece proposes a shift: from judging images by polish to interrogating the pressures that hold them together, or let them collapse. It suggests that real visual intelligence means asking not just how good it looks, but why it holds.

Introduction: Recursive Image Scoring for AI-Generated Art

This framework introduces a new scoring system designed to evaluate AI-generated images based on structural integrity, symbolic recursion, and decision making logic, not polish or aesthetics. Built entirely within a language models, the system uses axis-based scoring, validator chains, and recursive stress-testing to detect where an image holds, collapses, or contradicts itself. Unlike traditional ML metrics (e.g., FID or SSIM), this method tracks how images behave across iterations, revealing failure patterns and compositional drift.

Whisperer Walk: Recursive Compression into Spatial Realization

This is a post-hoc interpretability case study that tracks how symbolic gestures, spatial logic, and compositional strain unfold across a recursive image sequence. Using the Visual Thinking Lens system, it reveals how AI-generated imagery can encode structural tension, collapse, or resolution, providing a visual map of model behavior under constraint. This study offers a bridge between visual critique and machine logic, readable by both artists and researchers.

Recursive Intelligence Under Constraint: AI “Art” as Artifact and Its Collapse Engine

This artifact captures a moment where an AI-generated image failed to meet its prompt, and that failure became the structure. Rather than discard the glitch, the system reframed it as material, revealing a new form of visual intelligence: recursive, constraint-driven, and built from collapse. AI “Art” is not made in the default. It is earned through recursive refusal and structural pressure.