Source-linked AI summary
AI for Next Generation Computing: Emerging Trends and Future Directions
Sukhpal Singh Gill, Minxian Xu, Carlo Ottaviani, Panos Patros, Rami Bahsoon, Arash Shaghaghi, Muhammed Golec, Vlado Stankovski, Huaming Wu, Ajith Abraham, Manmeet Singh, Harshit Mehta, Soumya K. Ghosh, Thar Baker, Ajith Kumar Parlikad, Hanan Lutfiyya, Salil S. Kanhere, Rizos Sakellariou, Schahram Dustdar, Omer Rana, Ivona Brandic, Steve Uhlig
TL;DR
Next-generation computing still struggles to make AI/ML decisions practical, explainable, and executable in complex systems. This article synthesizes trends, challenges, and opportunities across emerging computing paradigms and proposes a conceptual integration framework, concluding that significant obstacles and risks remain alongside promising research directions.
Problem
AI/ML predictions for computing systems remain insufficiently practical, explainable, and executable despite applications in resource allocation and energy efficiency.
Method
The article synthesizes AI/ML trends across cloud, fog, edge, serverless, and quantum computing while developing a conceptual integration framework.
Results
The article identifies emerging developments, implementation obstacles, and potential risks for AI/ML-integrated next-generation computing.
Takeaways & Limitations
AI/ML-enabled autonomic computing offers broad opportunities across modern computing platforms, but substantial research challenges remain before effective integration.
Takeaways & Limitations
Quantum computing remains constrained by computational noise and the substantial qubit and logic-gate costs of quantum error correction.
Abstract
from arXiv · showhide
Autonomic computing investigates how systems can achieve (user) specified control outcomes on their own, without the intervention of a human operator. Autonomic computing fundamentals have been substantially influenced by those of control theory for closed and open-loop systems. In practice, complex systems may exhibit a number of concurrent and inter-dependent control loops. Despite research into autonomic models for managing computer resources, ranging from individual resources (e.g., web servers) to a resource ensemble (e.g., multiple resources within a data center), research into integrating Artificial Intelligence (AI) and Machine Learning (ML) to improve resource autonomy and performance at scale continues to be a fundamental challenge. The integration of AI/ML to achieve such autonomic and self-management of systems can be achieved at different levels of granularity, from full to human-in-the-loop automation. In this article, leading academics, researchers, practitioners, engineers, and scientists in the fields of cloud computing, AI/ML, and quantum computing join to discuss current research and potential future directions for these fields. Further, we discuss challenges and opportunities for leveraging AI and ML in next generation computing for emerging computing paradigms, including cloud, fog, edge, serverless and quantum computing environments.
1. Introduction
Autonomic computing enables systems to achieve desirable behaviours with limited human intervention through self-* capabilities and varying levels of autonomy. This review examines how AI/ML can support autonomic computing across cloud, fog, edge, serverless, quantum, and IoT-enabled environments while identifying emerging challenges and directions.
- Autonomic Computing: Autonomic computing studies how systems achieve desirable behaviours independently, including self-configuration, self-optimization, self-protection, and self-healing.These behaviours are collectively described as self-* capabilities.
- Autonomic Computing: Autonomic systems realise self-management with varying autonomy and human intervention through managers that sense environments and act using knowledge-based rules.IBM’s Autonomic Manager interacts through sensors and effectors, with administrators configuring its operation.
- AI/ML for Autonomy: AI/ML supports autonomic behaviour by discovering workload patterns for resource optimization and adaptively identifying system dynamics to enable self-learning.The passage identifies recurrent neural networks as an example of an adaptive dynamical-system identification method.
- Emerging Computing Paradigms: AI/ML-based autonomic computing is increasingly applied across cloud, fog, edge, serverless, and quantum paradigms, especially where configuration spaces are large.Larger parameter spaces increase the potential value of optimizing searches over configuration options.
- Challenges: Although autonomous management is growing in web services and data centers, autonomic capabilities remain inconsistently integrated into schedulers and workflow systems.Existing applications can manage node failures, network updates, and limited performance optimization, but coverage is incomplete.
- Review Scope and Contribution: The paper presents a review combining AI/ML with cloud, fog, edge, serverless, and quantum computing, using IoT, edge, and computing-continuum applications as an exemplar.Example application areas include smart agriculture, environmental monitoring, industrial digital twins, smart cities, and renewable-energy management.
2. A Prospective Model for Next Generation Computing Systems · 2.1. IoT Applications · 2.2. Resource Manager
The prospective model integrates AI/ML with autonomous computing to deliver effective services for diverse IoT applications. Its resource-management architecture coordinates resources, workloads, QoS, SLAs, AI-based scheduling, and explainable guarantees.
- 2. A Prospective Model for Next Generation Computing Systems: The proposed software architecture integrates AI/ML and autonomous computing technologies to provide effective computing services for varied IoT applications.The model is presented as a prospective architecture linking AI/ML with autonomous computing systems.
- 2.1. IoT Applications: Gateway devices connect IoT/edge devices and end users to computer systems while abstracting sensor and actuator interactions at the edge.This abstraction supports communication between edge-connected devices and computing systems.
- 2.1. IoT Applications: The system supports multiple IoT applications or digital twins, including healthcare, smart cities, farming, and weather monitoring.These applications receive AI and other autonomic services through the proposed system.
- 2.2. Resource Manager: The resource-management module adaptively and fault-tolerantly manages scheduling while tracking available, reserved, and desired resources, constraints, and task QoS.Tracked resource information includes CPU utilization, memory, price, resource type, and resource quantity.
- 2.2. Resource Manager: The QoS manager estimates workload completion time, while SLA-based priority queues organize critical cloud workloads according to violation probabilities and penalty rates.Priority queues target workloads with urgent deadlines in the execution state.
- 2.2. Resource Manager: AI predicts workload schedules and resources to meet task budgets and timeframes, while the resource manager can provide explainable guarantees under uncertainty.Explainable AI methods may support these guarantees where possible.
2.3. Autonomic Model
The autonomic model uses IBM’s four self-management capabilities—self-healing, self-configuring, self-protecting, and self-optimizing—implemented through interacting autonomic elements that monitor, analyze, plan, execute, and communicate adaptations. AI/ML supports prediction, fault and threat detection, adaptive scheduling, resource configuration, and QoS-aware optimization.
- 2.3. Autonomic Model: IBM’s autonomic model emphasizes self-healing, self-configuring, self-protecting, and self-optimizing capabilities.These capabilities target uninterrupted operation, security, automated component updates, and improved performance.
- 2.3. Autonomic Model: AI/ML enables automated configuration, threat and fault detection, adaptive scheduling, and resource addition to improve QoS and meet deadlines.The executor can add a backup-pool node when needed, while adaptive scheduling adjusts to changing environments and data-intensive workloads.
- 2.3. Autonomic Model: Autonomic elements independently manage resources and exchange performance messages to maintain distributed-system performance.Their interactions support models that self-heal, self-configure, self-optimize, and self-protect.
- 2.3. Autonomic Model: Sensors and monitors collect QoS, fault, update, and security data, comparing AI-predicted outcomes with observed performance against recorded thresholds.The knowledge base stores QoS thresholds, including the highest SLA-violation value, to identify deviations and trigger responses.
- 2.3. Autonomic Model: The Analyze and Plan unit predicts QoS metrics and selects responses such as restarting failed nodes or adding resources after alarms.Before adaptation, formal guarantees from control theory can support plan validation, while AI can anticipate user goal updates and assess plans in advance.
- 2.3. Autonomic Model: The knowledge base records current and previous system states alongside user- or administrator-defined desired functional and nonfunctional states.Centralized information exchange supports distributed tasks, while replication may improve reliability across the system.
2.4. Service Management Layer
The Service Management Layer combines database, security, blockchain, and serverless management functions to support IoT task execution and resource control. It also encompasses FaaS, SaaS, PaaS, and IaaS services, including dynamically scalable function containers.
- Service Management Layer: The database manager handles IoT application data, while AI-based security systems predict and guard against external threats during task execution.Blockchain services can secure application-data transmission during execution.
- Service Management Layer: The serverless manager controls cloud resources consumed by IoT applications at runtime.The passage also identifies integration of serverless data pipelines with quantum computers as an emerging direction, but the supplied text is truncated.
- Service Management Layer: The layer includes four service categories: function (FaaS), software (SaaS), platform (PaaS), and infrastructure (IaaS).Function containers provide dynamically scalable virtual environments; SaaS uses VM-based virtualization, while PaaS can be delivered through Microsoft Azure, AWS, or OpenStack.
3. AI for Computing: Current Status and Open Challenges
This section examines research opportunities for leveraging AI and ML in next-generation computing, focusing on emerging trends and open challenges across multiple computing paradigms.
- 3. AI for Computing: Current Status and Open Challenges: The section identifies research opportunities for leveraging AI and ML in next-generation computing.These opportunities are presented as a central focus of the discussion.
- 3. AI for Computing: Current Status and Open Challenges: It covers AI and ML integration across cloud, fog, edge, serverless, and quantum computing environments.These emerging computing paradigms are identified as the application contexts.
- 3. AI for Computing: Current Status and Open Challenges: The section discusses new trends and open challenges in AI-integrated next-generation computing.Figure 4 presents the emerging-paradigm scope referenced in this discussion.
3.1. Cloud Computing
Cloud computing and AI reinforce each other by improving performance, efficiency, agility, insight, and cost savings. Realizing these benefits requires substantial integration, high-quality data, and strong security and privacy protections.
- Benefits: Combining AI tools with cloud computing enhances enterprise efficiency, strategic insight, responsiveness, adaptability, flexibility, agility, and cost savings.Cloud-hosted data and applications support more responsive operations while reducing costs.
- Cloud-native management: Cloud-native microservices improve application update efficiency, but their growing number and time-sensitive features make effective management challenging.AI/ML-based solutions, including neural-network approaches, are presented as ways to address these management challenges.
- Benefits: AI in the cloud improves data management through real-time insights, data cataloguing, fraud detection, and anomalous-trend identification.These capabilities help organizations handle large data repositories and support day-to-day operations.
- Benefits: AI-cloud integration enables intelligent automation and predictive infrastructure management, reducing downtime while improving customer experience and operational outcomes.Algorithmic models use historical data and other patterns to automate business processes and infrastructure management.
- Challenges: Adopting AI in the cloud is constrained by difficult technology integration, inadequate data quality, and security and privacy risks.Businesses must migrate applications to the cloud, ensure data is accessible and clean, and protect sensitive and financial information from breaches.
- Implementation requirements: Proper AI-cloud integration requires substantial knowledge, resources, and financial investment before businesses can use advanced capabilities such as image recognition and natural language processing.The passage presents effective integration as necessary for realizing these machine-learning capabilities.
3.2. Fog Computing
Fog computing supplements cloud services for IoT by processing data and supporting time-sensitive decisions closer to sensors. AI and ML enable decision support in fog environments, while deployment remains challenged by latency, mobility, resource, energy, security, and reliability requirements.
- Motivation: Fog computing supplements cloud computing to handle massive IoT data and support applications with minimal reaction-time requirements.It provides basic network services for time-sensitive IoT applications.
- AI Enablement: AI’s ML and DL subfields provide useful data insights and decision support for AI-enabled fog computing technologies.These capabilities are discussed as foundations for fog applications.
- Architecture: Distributed fog devices co-located across the network accelerate IoT data analysis and decision making beyond the data centre.Decision-making and action-taking must occur near the sensors rather than solely in the cloud.
- Challenges: Fog application deployment remains constrained by execution time, mobility awareness, dynamic resource scheduling, energy efficiency, security and privacy, and fault tolerance and availability.The literature identifies these issues across application placement and fog infrastructure management.
- Applications: Emerging fog applications include image processing, video, NLP, and robotics, with placement and processing requiring efficient AI-based methods.Image applications classify and discriminate objects or people, while NLP scenarios emphasize careful and fast speech processing.
3.3. Edge Computing
Edge computing brings processing and storage closer to users and enables AI-enabled applications, but its adoption faces scalability, performance, security, governance, integration, and investment challenges. AI-integrated edge computing requires stronger frameworks, standards, monitoring, and integration with cloud architecture.
- Edge Computing: Edge computing moves processing and storage closer to end users, while big data still requires cloud-based storage.This positions edge as a distributed layer between devices and cloud infrastructure.
- Edge Computing: Edge computing enables AI applications through distributed architectures, device-specific security strategies, and reduced data movement.Sensitive personal data requires multiple encryption levels and dynamic encryption processes.
- Edge Computing: 1% of monitoring data is estimated to provide business insights such as anomaly identification or future event prediction.Edge computing can reduce cloud and network-bandwidth expenditures by processing relevant data nearer its source.
- Edge Computing: Edge environments face scalability and performance problems, especially for mission-critical applications, alongside difficult monitoring across many remote locations.Real-time endpoint-to-cloud performance management and failure handling are also critical issues.
- Edge Computing: AI-integrated edge adoption is constrained by security vulnerabilities, missing standards, heterogeneous platforms, limited edge capacity, cloud integration complexity, investment, and limited awareness.Proposed remedies include new software frameworks, recognized monitoring technologies, and faster research and development in security, governance, and standards.
3.4. Serverless Computing
Serverless computing abstracts infrastructure management through event-driven functions, managed services, autoscaling, and execution-based pricing. Its AI/ML applications benefit from simplified deployment and flexible resource use, but vendor lock-in remains an open challenge.
- 3.4. Serverless Computing: Serverless abstracts operational management, enabling cloud providers to automate provisioning while developers focus less on infrastructure maintenance.The paradigm is expected to expand as infrastructure responsibilities shift away from developers.
- 3.4. Serverless Computing: Event-driven functions execute code automatically as events occur, connect services, and support pay-per-execution billing based on time and resources used.This model can accelerate development by linking multiple services within an application.
- 3.4. Serverless Computing: Serverless Kubernetes automatically scales developer-provided containers up or down to handle exceptional traffic and fluctuating workloads.The scaling function operates within Kubernetes-managed clusters.
- AI/ML Integration: Serverless platforms can simplify AI-based applications by providing dynamic scalability, security, uninterrupted storage and message delivery, and support for machine-learning workflows.Machine-learning workflows include data analysis and preprocessing, model training, and model fine-tuning.
- AI/ML Integration: AI/ML-integrated serverless architectures provide execution-based pricing, independent functions, autoscaling, pay-per-use, reduced server-management burdens, and high availability.These benefits allow teams to work with limited intervention while systems adjust to changing workloads and provide fault tolerance.
- Open Issues and Challenges: Vendor lock-in makes switching cloud providers difficult, while the lack of industry-wide standards contributes substantially to cloud-computing adoption challenges.The passage identifies vendor locking as an open issue for serverless AI applications.
3.5. Quantum Computing
Quantum computing processes information through quantum states, using superposition and entanglement to offer capabilities beyond classical processing and potential influence on AI. Its development is constrained by fragile hardware and noise, with NISQ devices and quantum error correction representing successive steps toward fault-tolerant computation and practical applications.
- Quantum Computing Fundamentals: Quantum computing encodes information in quantum states and uses superposition and entanglement, enabling processing opportunities unavailable to classical systems.Its basic information unit is the qubit, represented in a complex Hilbert space with basis states |0⟩ and |1⟩.
- Quantum Computing Fundamentals: Quantum computing can be more powerful than classical computing for certain tasks, but composite-system phase spaces scale exponentially with the number of component systems.The same quantum features that provide computational power also make efficient simulation and control difficult.
- Hardware Challenges: Quantum advantages diminish exponentially as quantum hardware grows in size and complexity, requiring ultra-low temperatures that create additional materials and engineering challenges.The devices need temperatures at fractions of a Kelvin.
- Development Roadmap: NISQ devices are a foreseeable near-term target for simulating complex composite quantum systems, while reliable fault-tolerant computation will require quantum error correction and thousands or more qubits.Quantum error correction imposes substantial costs in qubits and logic gates.
- AI and Applications: Quantum computing may accelerate AI development and support applications including faster genomic analysis, personalized medicine, drug discovery, finance, transportation, sustainability, and cybersecurity.An experiment reported improved and accelerated reinforcement-learning agent training using a quantum channel, while IBM reported quantum algorithms outperforming classical Monte Carlo simulations.
4. Modern Autonomic Computing with Embedded Intelligence
Modern autonomic computing embeds intelligence in IoT, edge, and cyber-physical systems to process growing data closer to devices and support adaptive decision-making. Key challenges include limited edge resources, complex AI workloads, and the design of reliable sensing-actuation loops.
- IoT and Intelligent Edge: Over 75 billion IoT devices are expected to be in use by 2025, generating massive data that is challenging to manage.Connected automobiles, smart metres, and in-store sensors contribute to this data growth.
- IoT and Intelligent Edge: Edge computing processes data on-the-fly, reducing network capacity use and saving time through local or nearby processing.Fog computing can combine cloud and edge computation rather than relying exclusively on a central cloud location.
- IoT and Intelligent Edge: Future networks must process data in real time at the edge while adapting to far greater numbers of smart devices.The edge can help enterprises become more nimble and decrease expenses by avoiding unnecessary cloud transfers.
- IoT and Intelligent Edge: Intelligent edge supports decision-making for task offloading, edge caching, and resource scheduling despite constrained computing power.Data-stream mining adapts with finite memory and retrains only when performance falls below a threshold.
- Embedded Intelligence: IoT applications increasingly deploy intelligence, but limited device capacity makes running complicated AI applications difficult and leaves system- and algorithm-level challenges.Examples include natural language processing, face recognition, bio-medicine, and autonomous driving.
- Cyber-Physical Systems: Cyber-physical systems combine sensors, actuators, processing, and autonomous control loops to monitor and respond reliably to physical environments.Examples include transport systems, buildings, electricity grids, and water infrastructure.
5. Explainable AI (XAI) for Next Generation Computing
Explainable AI is important for next-generation computing because intelligent decisions determine the success of computing initiatives. Key concerns include system stability, workload robustness, causal explainability, and the high cost of inaccurate decisions.
- Intelligent decisions are critical to the success of computing initiatives.
- Computing systems must be assessed for stability and robustness before executing workloads.
- Trained models raise concerns about whether they are black boxes or causally explainable.
- Inaccurate decision-making is expensive in both cost and resource usage for complex technologies.
6. Potential Risks of AI-integrated Computing
AI-integrated computing can reduce costs but introduces upfront workforce expenses and risks involving connectivity, data privacy, errors, and over-reliance on models. Effective deployment therefore requires attention to latency, regulation, trust, control, and data engineering.
- Workforce costs: AI can save money, but a highly trained workforce makes adoption expensive initially.The passage identifies workforce expense as an upfront drawback.
- Internet Connectivity Issues: Slow internet connections can create latency between cloud, fog, or edge transfers and replies, hindering autonomous ML/AI computing.Server ML methods are especially vulnerable because forecasting speed is important.
- Privacy of Data: AI’s extensive data requirements raise ownership, security, compliance, and privacy challenges involving customer and provider information.Autonomic computing with AI requires privacy regulations and data security.
- Possibility of Errors: AI systems may generate problematic statements and make mistakes while exploring many possible solutions, requiring established trust and control.The passage cautions against further use before trust and control are established.
- Over-reliance on AI Models: AI/ML algorithms are only a small part of complex software-intensive systems, and model quality depends on representative training data.Strong data engineering processes are required to select appropriate and representative datasets.
7. Emerging Trends and Future Directions
The section identifies emerging AI/ML-integrated computing technologies across three maturity levels and outlines unresolved research directions for cloud, fog, edge, serverless, quantum, and Cloud-to-Things environments. Priorities include autonomic resource management, security, dependability, scheduling, quantum scalability, and decentralized coordination.
- Emerging Trends and Future Directions: Research classifies computing fields into three maturity levels: 5 years, 5 to 10 years, and more than ten years, illustrated by a next-generation computing hype cycle.The selected technologies could make efficient use of AI/ML-integrated next-generation computing.
- Cloud Computing: New ensemble machine learning approaches should govern user-based QoS-based container clusters across Docker Swarm and Kubernetes while preserving cloud service dependability and QoS.Advanced machine or deep learning techniques are identified as necessary for dependable cloud services.
- Cloud Computing: AI/ML research should address cost-effective network virtualization, energy minimization, dependability, timely scaling decisions, and thermal-aware task and resource scheduling in cloud environments.Cloud-based Big Data analysis may identify client-behavior trends and support better decisions, while thermal-aware methods target scheduling improvements.
- Fog and Edge Computing: Fog and edge research should forecast security vulnerabilities, estimate geographic resource requirements, schedule tasks, protect privacy, and automate decisions under decentralized and resource-limited conditions.Blockchain-enhanced security, IoT-device patching, runtime analytics, and AI/ML-based decision-making are proposed for edge systems.
- Serverless Computing: Serverless research includes AI-enhanced task delay and reaction times, ensemble deep learning for automatic heart disease detection, and AI/ML threat mitigation through clustering-based security analysis.These directions target IoT and Serverless computing contexts.
- Quantum Computing: Quantum computing priorities include scaling chips while controlling noise, integrating quantum error correction toward FTQC, developing cloud quantum infrastructures, and applying AI to IoT-scale data.NISQ devices could simulate complex quantum-system dynamics, while FTQC could support novel-material design and AI applications.
- Cloud-to-Things Computing: Cloud-to-Things research highlights Decentralized Autonomous Organizations, blockchain-enabled smart contracts, Smart Oracles, and AI/ML knowledge management for distributed heterogeneous infrastructures.Smart Oracles can provide decentralized monitoring across vertically and horizontally distributed infrastructures.
8. Conclusions and Summary
Modern computing systems underpin services across cloud, fog, edge, serverless, and quantum environments, while AI, ML, and DL advances create growing demands for scalable, QoS-aware computing. The article surveys emerging AI/ML trends, challenges, and future integration directions across these computing paradigms.
- Computing Foundations: Modern computing systems provide services across Cloud, Fog, Edge, Serverless, and Quantum Computing and support low-latency solutions to real-world problems.These systems have also enabled start-ups to access large computing capacity for challenging problems and scientific progress.
- AI/ML Drivers: AI, ML, and DL have gained popularity through accuracy advances in computer vision, natural language processing, and related applications.Their training has been supported by massive data collection and advances in state-of-the-art computing hardware, including GPUs.
- Emerging Demands: AI and ML demand computing systems that handle huge heterogeneous IoT and sensor data streams while delivering customized, QoS-aware services across diverse hardware.The stated QoS dimensions include latency, energy, cost, and response time, alongside software-level and financial constraints.
- Article Contribution: The article presents a holistic vision of AI and ML trends for cloud, fog, edge, serverless, quantum, and other computing platforms.It brings together breakthroughs, identifies obstacles to AI/ML implementation in modern computing, and develops a conceptual framework for future technology integration.
Authors Biography
The authors are academics and researchers across cloud, adaptive systems, quantum technologies, cybersecurity, AI, and mobile computing. Their biographies reflect substantial teaching, research, publication, editorial, and professional experience across international institutions.
- Sukhpal Singh Gill: Sukhpal Singh Gill is a Lecturer in Cloud Computing at Queen Mary University of London, with prior research roles at Lancaster University and the University of Melbourne.He has co-authored 70+ peer-reviewed papers, holds an H-index of 30+, and researches cloud, fog, software engineering, IoT, and energy efficiency.
- Minxian Xu: Minxian Xu is an associate professor at the Shenzhen Institute of Advanced Technology whose research focuses on cloud resource scheduling and optimization.He has co-authored 30+ peer-reviewed papers and earned his Ph.D. from the University of Melbourne in 2019.
- Carlo Ottaviani: Carlo Ottaviani is an Associate Lecturer and Research Fellow at the University of York working on quantum information, quantum technologies, security, quantum-key distribution, and quantum networks.His earlier work included quantum-phase gates, quantum memories, and quantum repeaters using nonlinear quantum atom-optics systems.
- Panos Patros and Rami Bahsoon: Panos Patros leads the Cloud and Adaptive Systems ORKA Lab at the University of Waikato, while Rami Bahsoon researches software architecture, cloud and services engineering, and self-adaptive software.Bahsoon is a Reader at the University of Birmingham and has co-edited four books on software architecture.
- Other authors: The remaining authors contribute expertise in cybersecurity, AI and cloud computing, computer science, and mobile cloud and deep learning research.Arash Shaghaghi is a Senior Lecturer in Cyber Security at RMIT University; Muhammed Golec is a Queen Mary University Ph.D. student; Vlado Stankovski is a Full Professor at the University of Ljubljana; and Huaming Wu is an associate professor at Tianjin University.