Machine Learning for 6G Wireless Networks: Paradigms, Enablers, and the Pursuit of AI-Native Trustworthiness

Abstract

The sixth generation (6G) of wireless communication systems is envisioned to be inherently AI-native, integrating intelligence into every network layer to support unprecedented capabilities, including terabit-per-second data rates, sub- millisecond latency, and pervasive sensing . This ambition re- quires managing extreme complexity introduced by revolutionary technologies such as Terahertz (THz) communication, Ultra- Massive MIMO (UM-MIMO), and Reconfigurable Intelligent Surfaces (RIS) . Machine Learning (ML) is recognized as the computational backbone for this transformation, enabling adaptive, self-optimizing, and context-aware wireless environments that fundamentally redefine how networks operate . This paper presents a systematic review, mapping ML across three progressive integration paradigms: AI for Network (AI4NET), Network for AI (NET4AI), and AI as a Service (AIaaS). We detail ML’s pivotal role in enhancing the physical layer through deterministic Wireless Environment Control (WEC) and robust channel estimation using generative models . Furthermore, we elaborate on distributed intelligence architectures, such as Federated Learning (FL) and Split Learning (SL), which are essential for balancing high computational demands with data privacy and resource constraints in the emerging Computing Power Network (CPN) . Finally, we argue that the core viability of 6G depends on embedding trustworthiness into its architecture, emphasizing the mandatory roles of Explainable AI (XAI) for operational accountability and Distributed Ledger Technology (DLT) for immutable data provenance .

Introduction

The rapid evolution of wireless communication has progressed from 5G to the emerging vision of 6G, expected around 2030, driven by the increasing demands of immersive applications such as holographic communication, digital twins, remote robotics, and real-time human–AI interaction. These applications require performance that far surpasses what 5G can offer.

6G aims to create an AI-native, cyber-physical continuum integrating the human, physical, and digital worlds. Guided by principles of sustainability, trustworthiness, and inclusion, 6G will rely on advanced technologies including Terahertz (THz) communication, Reconfigurable Intelligent Surfaces (RIS), Integrated Sensing and Communication (ISAC), and Ultra-Massive MIMO. Owing to their complexity and non-linearity, these systems require machine learning (ML) and artificial intelligence (AI) as intrinsic architectural elements. AI will enable real-time sensing, learning, prediction, and optimization across the entire communication stack.

6G also introduces unprecedented performance requirements, especially for mission-critical Industrial IoT (IIoT). Targets include 100× higher spectral and energy efficiency, reliability up to 1–10??, sub-millisecond latency, centimeter-level positioning accuracy, ultra-low jitter, and device lifetimes up to 40 years. Achieving these key performance indicators (KPIs) depends heavily on AI-driven resource management, prediction, channel modeling, and deterministic control.

AI integration in 6G evolves through three paradigms:

1. AI for Network (AI4NET) – AI enhances existing network functions such as air-interface optimization, intelligent resource allocation, and predictive operations and maintenance.

2. Network for AI (NET4AI) – The network becomes an AI-supporting platform, providing distributed computing, data handling, and security infrastructure needed for large-scale AI workloads.

3. AI as a Service (AIaaS) – AI becomes a built-in service that industries and users can consume directly, enabled through a Quality of AI Service (QoAIS) framework and scalable, secure AI functions.

To meet the varied demands of 6G, multiple ML paradigms are used:

• Supervised Learning for traffic prediction, beam management, CSI feedback optimization, and threat detection.

• Unsupervised Learning for clustering, dimensionality reduction, and anomaly detection in massive, heterogeneous datasets.

• Reinforcement Learning and Deep Learning for dynamic resource orchestration, wireless control, path optimization, and physical-layer operations.

• Specialized architectures, such as Federated Learning for privacy-preserving edge intelligence, Graph Neural Networks for modeling network topologies, and Digital Twins for safe AI training and optimization.

Conclusion

The integration of Machine Learning is the decisive factor transforming the sixth-generation (6G) communication system from a mere technological upgrade into an autonomous, cog- nitive ecosystem . This structural shift is necessitated by the extreme complexity and dynamism introduced by fundamen- tal enablers such as Terahertz (THz) communication, Ultra- Massive MIMO (UM-MIMO), and Reconfigurable Intelligent Surfaces (RIS), problems that classical model-based solutions are incapable of managing efficiently . The realization of per- vasive intelligence is structured by the systematic progression through three core paradigms: from using AI as an external enhancement tool (AI4NET), to redesigning the infrastructure for native AI support (NET4AI), and culminating in the commercialization of measurable intelligence as a core product (AIaaS) . Achieving the unprecedented 6G Key Performance Indi-cators (KPIs)—particularly ultra-high reliability (1 ? 10?9), ultra-low latency (< 1 ms), and microsecond-level jitter (< 0.1µs)—relies entirely on ML-driven technologies . AI is essential for transforming the traditionally stochastic physi- cal layer by enabling deterministic control of the wireless environment through RIS (Wireless Environment Control or WEC), thereby mitigating blockages and ensuring predictable channel conditions (Figure 1). Furthermore, ML ensures system robustness by deploying advanced generative models for channel estimation that maintain high accuracy even when exposed to unpredictable Out-of-Distribution (OOD) scenarios. Within the core network, ML algorithms, primarily Deep Reinforcement Learning (DRL), power specialized function- alities like Proactive Resource Management (PRM) and End- to-End Optimization (E2EO) (Figure 2), ensuring scheduling precedes the demand curve and addressing the complex, multi- dimensional optimization required for joint communication and computing resource allocation . The architectural foundation of 6G intelligence (NET4AI) is built on distributed ML architectures, driven by the need to integrate computation closer to the edge . Federated Learning (FL) and Split Learning (SL) are indispensable for balancing high computational demands with communication efficiency, resource constraints, and data privacy across the Comput- ing Power Network (CPN) architecture . Looking forward, Large Language Models (LLMs) are anticipated to be the cognitive core, enabling *Intent-Based Networking (IBN)* by translating high-level human objectives into executable network strategies . However, successfully deploying this cognitive layer requires developing specialized, low-parameter *on-device LLMs* to manage computational complexity at the terminal and edge servers . Complementary technologies like Quantum-Assisted Machine Learning (QML) hold radical potential for accelerating the training of vast AI models and solving computationally hard, high-dimensional optimization problems with a quadratic speedup . The core long-term viability and commercial adoption of AIaaS is conditional on establishing architectural Trustwor- thiness . This mandatory requirement is multi-layered: it integrates technical reliability via robust, adaptive ML mod- els, and safeguards security and privacy through Distributed Ledger Technology (DLT) . DLT (e.g., Blockchain) provides essential features like *immutable data provenance* and se- cure, decentralized model storage, mitigating advanced threats such as malicious aggregation injection attacks in distributed learning environments (Figure 3) . Critically, Explainable AI (XAI) is mandatory for governance, solving the ”black-box” problem by providing transparent, auditable justifications for autonomous decisions . XAI ensures that operational efficiency is balanced with human accountability, thereby translating technical reliability into societal trust (Figure 4) . Future research must prioritize the standardization of the *Quality of AI Service (QoAIS) framework to measure the commercial delivery of intelligence. Furthermore, accelerat- ing the development of secure, domain-specific **on-device LLMs* for autonomous orchestration and refining XAI tech- niques for real-time applicability in safety-critical domains are essential steps to ensure that 6G’s network autonomy is consistently transparent, auditable, and inherently trustworthy . The continued synergy between ML and physical/architectural design will ultimately determine the success of 6G in realizing a sustainable, hyper-connected, and fully intelligent cyber- physical continuum.

References

[1] M. A. Uusitalo et al., ”6G Vision, Value, Use Cases and Technologies From European 6G Flagship Project Hexa-X,” IEEE Access, vol. 9, pp. 17799–17812, 2021. [2] Q. M. Cui et al., ”Overview of AI and communication for 6G network: fundamentals, challenges, and future research opportunities,” Sci China Inf Sci, vol. 68, no. 7, pp. 171301:1–171301:61, 2025. [3] N. H. Mahmood, G. Berardinelli, E. J. Khatib, R. Hashemi, C. De Lima, and M. Latva-aho, ”A Functional Architecture for 6G Special-Purpose Industrial IoT Networks,” IEEE Trans. Ind. Informat., vol. 19, no. 3, pp. 2530–2540, Mar. 2023. [4] A. A. Puspitasari and B. M. Lee, ”A Survey on Reinforcement Learning for Reconfigurable Intelligent Surfaces in Wireless Communications,” Sensors, vol. 23, no. 5, pp. 2554, 2023. [5] U. Demirhan and A. Alkhateeb, ”Integrated Sensing and Communication for 6G: Ten Key Machine Learning Roles,” IEEE Commun. Mag., vol. 61, no. 4, pp. 113–119, Apr. 2023. [6] Y. Wang et al., ”Transformer-Empowered 6G Intelligent Networks: From Massive MIMO Processing to Semantic Communication,” IEEE Wirel. Commun., pp. 1–9, 2022. [7] S. Lohani, E. M. Knutson, and R. T. Glasser, ”Generative machine learning for robust free-space communication,” Commun. Phys., vol. 3, no. 1, pp. 177, 2020. [8] S. Liang, Q. Cui, X. Huang, and Y. Wang, ”Efficient hierarchical federated services for heterogeneous mobile edge,” IEEE Trans Serv Comput, vol. 18, no. 1, pp. 140–155, 2025. [9] S. K. Jagatheesaperumal et al., ”The collaborative impact of AI and Big Data in various applications of Industry 4.0,” International Journal of Intelligent Networks, vol. 3, pp. 102748, 2022. [10] G. Durisi, T. Koch, and P. Popovski, ”Toward massive, ultrareliable, and low-latency wireless communication with short packets,” Proc. IEEE, vol. 104, no. 9, pp. 1711–1726, Sep. 2016. [11] A. Maatouk, N. Piovesan, F. Ayed, et al., ”Large language mod- els for telecom: forthcoming impact on the industry,” arXiv preprint arXiv:2308.06013, 2023. [12] C. Wang and A. Rahman, ”Quantum-Enabled 6G Wireless Networks: Opportunities and Challenges,” IEEE Wireless Commun., vol. 29, no. 1, [13] pp. 58–69, Feb. 2022. [14] B. R. Das et al., ”A Comprehensive Survey on Emerging AI Technolo- gies for 6G Communications: Research Direction, Trends, Challenges, and Opportunities,” International Journal of Intelligent Networks, vol. 6, [15] pp. 113–150, 2025. [16] J. Passerat-Palmbach et al., ”Blockchain-orchestrated machine learning for privacy preserving federated learning in electronic health data,” in Proc. IEEE International Conference on Blockchain, 2020, pp. 550–555. [17] Y. Heng, J. Mo, and J. G. Andrews, ”Learning site-specific probing beams for fast mmWave beam alignment,” IEEE Trans Wireless Com- mun, vol. 21, no. 7, pp. 5785–5800, Jul. 2022.

Copyright

Copyright © 2025 M L Sharma, Sunil Kumar, Namita Gupta, Suteekshn Manchanda, Aditya Srivastava, Kamran Ahmad, Akshat Aggarwal, Priyanshi Tayal. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.