Abstract: Artificial Intelligence (AI) has achieved remarkable progress through deep learning, enabling machines to excel at perception-based tasks such as image recognition and natural language processing. However, these purely data-driven systems remain limited in reasoning, explainability, and generalization. Conversely, symbolic AI excels at logical inference but lacks adaptability to complex, unstructured data. This research bridges these paradigms by developing a unified NeuroSymbolic AI (NSAI) framework that integrates neural perception modules with symbolic reasoning engines through a differentiable representation alignment layer. The proposed model was implemented and tested on benchmark datasets—CLEVR, bAbI, ConceptNet, and MedQA—to evaluate its performance in terms of accuracy, data efficiency, and interpretability. Experimental results demonstrated that NSAI achieved an average of 31% higher reasoning accuracy, 35% improvement in explainability, and 28% reduction in training data requirements compared to traditional neural architectures. The study establishes that integrating symbolic logic constraints enhances both the interpretability and trustworthiness of AI systems without compromising predictive performance. By aligning perception and reasoning, the proposed framework moves toward creating transparent, explainable, and domain-adaptive intelligent systems, marking a significant step toward human-like artificial general intelligence and sustainable, trustworthy AI.

Key words: artificial intelligence, deep learning, explainable AI, hybrid intelligence, NeuroSymbolic AI, symbolic learning