Artigo - Atena Editora

Artigo

Baixe agora

Livros

Bridging intelligence and trust: A unified framework for AI and Blockchain integration

The rapid co-evolution of Artificial Intelligence (AI) and blockchain technology has exposed a persistent gap between intelligence—the ability to extract insight from data—and trust—the assurance that data, models, and decisions are transparent, verifiable, and tamper-proof. This study introduces the Unified Trust-Intelligence Framework (UTIF), an end-to-end architecture that natively fuses AI and distributed-ledger technologies to deliver auditable, privacy-preserving, and energy-aware intelligent services. A systematic review compliant with PRISMA guidelines (167 peer-reviewed sources, 2018-2024) reveals four critical deficiencies in the current literature: (i) the lack of formal on-chain model certification, (ii) opaque immutability of operational logs, (iii) limited cross-chain and cross-domain interoperability, and (iv) sub-optimal energy footprints. UTIF addresses these gaps through: On-chain algorithmic certification that fingerprints model weights and training metadata via cryptographic hashing. Federated data governance that combines privacy-preserving federated learning with zero-knowledge proofs (ZK-SNARKs) for regulatory compliance (GDPR, EU AI Act). An AI-assisted hybrid PoS-BFT consensus that dynamically tunes fault-tolerance parameters under varying network conditions. A self-verifiable MLOps pipeline deployed on Hyperledger Fabric with Layer-2 rollups, providing continuous integration, delivery, and audit trails. Experimental validation uses two open-access benchmarks—MIMIC-IV (clinical) and ECB-SDW (financial)—executed on a 20-node heterogeneous testbed. UTIF reduces transaction latency by 38 % and operational energy consumption by 27 % compared with Fabric 2.x and PoA baselines, while enhancing adversarial robustness (F1 + 12 %) through on-chain model attestation. A perception survey of 46 domain experts reports a statistically significant boost in trustability (+1.27 ± 0.31 on a 5-point Likert scale, p < 0.01). Stress tests show 98 % valid throughput under Sybil scenarios with 1,000 malicious nodes, maintaining a carbon footprint below 0.25 kg CO₂ e per 1,000 transactions. The findings demonstrate that deep, native convergence of AI and blockchain can simultaneously achieve measurable trust guarantees, competitive performance, and sustainability. The article concludes with regulatory implications, identified limitations (network scale, oracle dependencies), and a research roadmap toward edge-to-cloud, 6G-ready, Web3-compliant intelligent infrastructures.
Ler mais

Bridging intelligence and trust: A unified framework for AI and Blockchain integration

  • DOI: https://doi.org/10.22533/at.ed.317582501104

  • Palavras-chave: AI-Blockchain integration; digital trust; on-chain model certification; federated learning; zero-knowledge proofs; energy-efficient consensus; MLOps.

  • Keywords: AI-Blockchain integration; digital trust; on-chain model certification; federated learning; zero-knowledge proofs; energy-efficient consensus; MLOps.

  • Abstract: The rapid co-evolution of Artificial Intelligence (AI) and blockchain technology has exposed a persistent gap between intelligence—the ability to extract insight from data—and trust—the assurance that data, models, and decisions are transparent, verifiable, and tamper-proof. This study introduces the Unified Trust-Intelligence Framework (UTIF), an end-to-end architecture that natively fuses AI and distributed-ledger technologies to deliver auditable, privacy-preserving, and energy-aware intelligent services. A systematic review compliant with PRISMA guidelines (167 peer-reviewed sources, 2018-2024) reveals four critical deficiencies in the current literature: (i) the lack of formal on-chain model certification, (ii) opaque immutability of operational logs, (iii) limited cross-chain and cross-domain interoperability, and (iv) sub-optimal energy footprints. UTIF addresses these gaps through: On-chain algorithmic certification that fingerprints model weights and training metadata via cryptographic hashing. Federated data governance that combines privacy-preserving federated learning with zero-knowledge proofs (ZK-SNARKs) for regulatory compliance (GDPR, EU AI Act). An AI-assisted hybrid PoS-BFT consensus that dynamically tunes fault-tolerance parameters under varying network conditions. A self-verifiable MLOps pipeline deployed on Hyperledger Fabric with Layer-2 rollups, providing continuous integration, delivery, and audit trails. Experimental validation uses two open-access benchmarks—MIMIC-IV (clinical) and ECB-SDW (financial)—executed on a 20-node heterogeneous testbed. UTIF reduces transaction latency by 38 % and operational energy consumption by 27 % compared with Fabric 2.x and PoA baselines, while enhancing adversarial robustness (F1 + 12 %) through on-chain model attestation. A perception survey of 46 domain experts reports a statistically significant boost in trustability (+1.27 ± 0.31 on a 5-point Likert scale, p < 0.01). Stress tests show 98 % valid throughput under Sybil scenarios with 1,000 malicious nodes, maintaining a carbon footprint below 0.25 kg CO₂ e per 1,000 transactions. The findings demonstrate that deep, native convergence of AI and blockchain can simultaneously achieve measurable trust guarantees, competitive performance, and sustainability. The article concludes with regulatory implications, identified limitations (network scale, oracle dependencies), and a research roadmap toward edge-to-cloud, 6G-ready, Web3-compliant intelligent infrastructures.

  • Raul Jaime Maestre
Fale conosco Whatsapp