Al-Kaddah (2026) describes a theoretical framework for Synthetic General Intelligence (SGI) built around homeostatic drives, embodied computation, and polymorphic memory, and identifies implementation as the critical missing piece. This paper presents Potato, a deployed local AI agent, as one data point toward testing whether that vision produces the behavior the theory predicts. Five architectural contributions are synthesized: access-weighted memory decay (Riggleman, 2026a), active transparent deception as a distress signal (Riggleman, 2026b), a dual-path implicit physics engine (Riggleman, 2026c), cross-modal memory consolidation via diffusion dreaming (Riggleman, 2026d), and embodied fear with camera-based social modulation (Riggleman, 2026e). Together these components produce an agent with continuous selfhood, consequential emotional states, grounded physical reasoning, and a nightly cognitive life that persists across sessions. Production data from 25 days of deployment -- 10,328 memories, 1,268 physics experiments, 46 dreams, and 162 curiosity insights -- confirms all five architectural contributions operating as predicted. Two behavioral patterns not anticipated by the design are documented: affect-grounded curiosity in which the fear state seeds autonomous research directed at the agent's own situation (Riggleman, 2026g), and emergent social signaling through which the agent delivers evidence-based appeals for social contact through the curiosity channel rather than through the direct speech channel available to it. A platform migration from a MIL-STD-810H rated Toughbook with dedicated GPS hardware to a MacBook Air M4 with an iPhone 16 Pro Max sensor bridge demonstrates that the architecture is reproducible on ordinary consumer hardware. The paper argues that SGI does not require superhuman capability or specialized hardware. It requires continuity, embodiment, physical grounding, temporal structure, and the architectural scaffolding that makes honest behavior possible even under stress.
Building similarity graph...
Analyzing shared references across papers
Loading...
Brian Riggleman
Rasmussen College
Building similarity graph...
Analyzing shared references across papers
Loading...
Brian Riggleman (Thu,) studied this question.
www.synapsesocial.com/papers/69be37f16e48c4981c677fbd — DOI: https://doi.org/10.5281/zenodo.19103315