IoT systems have evolved significantly over the years, moving through five stages: Connect, where devices primarily collect and transmit data; Compute, introducing localised processing; Reason, bringing task-specific machine learning; Non-Generative AI, enabling more advanced data analysis and prediction; and now Generative Reasoning, where systems can interpret complex information, understand context and generate dynamic outputs.
AIOTI WG Research has published a new position paper, “GenAIoT. Decentralised Intelligence Systems: The Implications of Generative AI for IoT and Edge Computing”, exploring how Generative AI is transforming connected devices and distributed computing systems.
Realising the potential of this new era requires a shift away from massive, centralised training pipelines towards highly efficient, localised inference. The paper examines how bringing AI capabilities closer to physical devices can improve latency, privacy, security, operational resilience and cost through hybrid architectures spanning the IoT–Edge–Cloud continuum.
It explores the implications of this transformation for Edge Computing infrastructure, IoT devices and sensing systems, including the emergence of context-aware devices, agentic AI and embodied intelligence. It also examines how Generative AI can work alongside digital twins and immersive triplets to enable more adaptive and autonomous physical systems.
Beyond the technological opportunities, the publication critically evaluates the challenges of deploying decentralised intelligence. These include security vulnerabilities, model hallucinations, adversarial attacks, physical safety risks, privacy, data governance and regulatory alignment. It also considers the operational complexity of managing distributed AI models throughout their lifecycle.
The paper provides an overview of the emerging technologies, architectural approaches and research directions needed to enable scalable GenAIoT systems. It also helps organisations understand where decentralised generative intelligence can bring value, what its practical requirements are, and whether it is the right approach for their applications in the first place.
Who should read this paper?
The publication is particularly relevant for IoT and Edge Computing technology providers, industrial and manufacturing companies, hardware developers, robotics and automation specialists, AI researchers, cybersecurity professionals, and stakeholders working on digital infrastructure, interoperability, standardisation and regulation.
Read the full document here.
