Activation-Scaled ANN-to-SNN Conversion with SNN-Based AI Interpretability: A Unified Framework for Black-Box Analysis and Hallucination Detection

I present a unified framework that extends ANN-to-SNN conversion beyond efficiency optimization to enable novel AI interpretability analysis. My approach uses Spiking Neural Networks as "computational microscopes" to analyze black-box AI models. Key Findings (v3):- Universal threshold formula: θ = 2.0 × max(activation)- Tested on MLP, CNN, ResNet, GPT-2, and ViT-Base architectures- 100% accuracy preservation with hippocampus-inspired hybrid architecture- Hybrid readout: 70% spike rate + 30% membrane potential New in v3: AI Interpretability Framework- Time-to-First-Spike (TTFS) analysis for thought priority visualization- Neural Synchrony detection for concept binding analysis- Spike Stability (Jitter) for hallucination detection (AUC 0.75)- GPT-2 attention TTFS analysis: +3.1 increase for meaningless inputs- ViT-Base (86M params) validation with CIFAR-100 Recommended strategy: Keep feature extraction as ANN, convert only classification head to SNN with α=2.0. Code: https://github.com/hafufu-stack/autonomous-snn-framework