Artificial intelligence is moving beyond the digital world and into the physical environments in which we live and work. As connected devices become capable of interpreting data, making decisions and taking action, the Internet of Things (IoT) is entering a new phase. The implications extend well beyond the technology itself, creating new opportunities for enterprises while raising important questions about connectivity, security, business models and the responsibilities of the organisations involved.
These themes will be explored in the forthcoming fireside chat webinar, ‘Physical AI: What does it mean for enterprise IoT deployments when things start making decisions?’, featuring me, and Jon Connet and Suresh Sathyamurthy of Aeris.
The discussion will examine how AI is changing the IoT landscape, where the most tangible opportunities lie, and what enterprises, mobile network operators (MNOs) and connectivity platform providers need to do to prepare.
The relationship between AI and IoT can be understood through three dimensions: IoT feeding data into AI systems, AI running on IoT devices, and AI optimising IoT deployments. These are complementary developments, rather than mutually exclusive approaches.
The value of AI-enabled IoT lies in its ability to process more operational data, derive greater insight from it and make new applications viable. Precision agriculture, smart grids and predictive maintenance are among the areas where this could create tangible benefits. Better analysis can also allow systems to perform tasks that would previously have required specialist human expertise.
Video is particularly important in this context. The ability to analyse visual information more effectively could support applications that have previously been constrained by the volume of data involved or the cost of processing it. As AI makes it possible to extract more value from operational data, it also creates demand for more data, potentially expanding the range of viable IoT applications.
However, AI does not necessarily need to run in one place. The choice between device, edge, network and cloud processing depends on the application, the immediacy of the decision required, the sophistication of the model and the associated costs.
Processing at the edge can improve performance by enabling higher-resolution inputs, reducing delays and increasing resilience. It can also support compliance through local processing and anonymisation, while reducing the costs associated with connectivity, storage and processing. Cloud environments, meanwhile, offer greater computing resources and can be easier to manage and augment.
The challenge is to determine the right balance. Real-time applications, such as drone collision avoidance, have different requirements from batch-processing applications such as retail inventory management. The architecture must reflect those differences.
For enterprises, AI-enabled IoT does not necessarily mean an entirely new business case. In many instances, the opportunity remains familiar: using sensor data more effectively, improving operational efficiency and deriving greater value from existing deployments.
Nevertheless, the introduction of AI makes product design and deployment more multidisciplinary. Devices may require different processing capabilities, sensors and power characteristics. The business case may also become more critical to the organisation, potentially involving greater security requirements and new budget owners.
The technical requirements are evolving too. Devices running AI models may not always generate more data. In some cases, more effective processing could reduce the volume transmitted, even as the devices become more capable. Latency requirements may also be lower for certain applications.
But managing these deployments introduces new responsibilities. Over-the-air updates, model compression, lifecycle management, model drift and feedback loops become important considerations.
Enterprises will also need middleware capable of bringing AI and IoT together. This must support model updates and redistribution, adaptation to individual deployment environments, and the maintenance of software bills of materials, compliance and auditability.
Ultimately, responsibility for an autonomous device making a bad decision remains with the enterprise operating it.
For MNOs, the growth of AI-enabled IoT presents both an opportunity and a challenge. Greater demand for IoT data could support connectivity revenues, while making previously unviable business models possible. Video monitoring is one example.
However, edge processing and redaction can reduce the amount of data transmitted, potentially changing the basis on which connectivity is priced. The implications should not be overstated: many deployments have not had the capacity to process all their data in the first place.
The more fundamental requirement is consistency. AI systems depend on reliable data, and connectivity problems can create gaps in historical records, introduce biases into datasets or result in incorrect classifications. In some cases, anomalous behaviour may be linked directly to poor connectivity.
For real-time operational AI, continuous and predictable data flows can be essential. Coverage gaps, network congestion, cell handovers and differences in network capabilities can all affect performance. Delayed data can be just as problematic as incorrect data, potentially undermining confidence in the system.
Latency, large model downloads and data sovereignty requirements also make traffic routing and the selection of computing environments increasingly important. Local breakout can support latency and compliance requirements, but it requires local network infrastructure.
The opportunity for operators is therefore not simply to sell more connections. It is to provide the resilience, compliance and service assurance required to support increasingly sophisticated IoT deployments.
Connectivity platform providers will have an important role in this transition. Beyond orchestrating resilient connectivity, platforms may need to support model management, security verification and proactive service assurance.
AI introduces additional security threats, including model theft, adversarial manipulation, data poisoning and compromised update mechanisms. At the same time, edge redaction can reduce the amount of sensitive information crossing corporate boundaries.
There is no single solution that guarantees success. Operators need to strengthen the underlying connectivity proposition, while platform providers and other partners help manage the complexity of increasingly intelligent deployments.
For enterprises, the priority is to understand how AI changes the requirements of their IoT applications. For operators, it is to ensure that connectivity remains valuable as the market evolves.
These are the questions that will shape the conversation during the webinar ‘Physical AI: What does it mean for enterprise IoT deployments when things start making decisions?’. The discussion will look beyond the promise of Physical AI to examine what it means in practice, and what organisations across the IoT ecosystem should be doing now to prepare. We hope you can join us.
