This work proposes AI to Learn (AI2L)—a practical framework that restricts AI to a learning‑support role and removes any black‑box AI component from the final deliverables. Built on four pillars—
(1) Model transparency (black‑box elimination), (2) Accountability via human verification, (3) Information protection & privacy, (4) Green AI (energy efficiency & sustainability)—AI2L keeps humans as the ultimate decision‑makers, leveraging AI’s strengths while systematically mitigating risks (opacity, data leakage, power consumption).
Positioning & Rule: AI is confined to learning support; no opaque model remains in finished artifacts (papers, code, materials, production systems).
Four Pillars:
Transparency (black‑box elimination).
Accountability (human validation, logic checks, reproducibility).
Privacy (local processing/anonymization).
Green AI (use large models only during learning support; distill to lightweight local code for operation).
Representative Implementations:
Grad‑CAM–based evidence visualization → codified human criteria for inspection (e.g., detecting under‑polished titanium regions).
Symbolic regression → candidate equations guiding discovery (HSOs’ chaotic property; leading to CRH/S4C/Chaordic Homeodynamics).
AI‑generated code with human hardening → local, model‑free operation (no cloud inference/GPU dependency).
Reversible anonymization → privacy‑preserving practical workflows.
What’s distinct: Unlike standard Human‑in‑the‑Loop or generic XAI, AI2L includes removal of large‑model dependence from the delivered system.
Use cases: Education, healthcare, public sectors—domains requiring explainability and energy efficiency.
Seine A. Shintani. AI to Learn (AI2L): Guidelines and Practice for Human‑Centered AI Utilization as a Learning Support Tool—Four Pillars of Black‑Box Elimination, Accountability, Information Protection, and Energy Efficiency. Jxiv (Preprint, Version 1, 2025).
DOI: 10.51094/jxiv.1435
Keywords: AI to Learn (AI2L), Model Transparency (Black‑Box Elimination), Accountability, Information Protection and Privacy, Green AI (Energy Efficiency and Sustainability)