Different people define digital transformation in different ways. But at its core, digital transformation is about profitably using data from smart devices to pursue your business objectives such as increasing efficiency of operations, improving customer experience and even creating new products and services. Artificial intelligence (AI) and the Internet of Things (IoT) are, thus, jointly advancing digital transformation.

With IoT, technologists seek to digitise the world by embedding billions of sensors in our consumer gadgets and homes, our vehicles and industrial equipment—practically almost everywhere and in everything. On their part, AI applications have picked up momentum and started having a significant impact in our lives and on our businesses in the last decade because of the availability of large amounts of data, vastly greater computing power at lower costs and important breakthroughs in machine learning (ML) methods.

Today, in almost all cases, IoT and AI work together in cloud computing—because the vast computing power required to train real-world ML models is available only in the cloud. Data from IoT devices is transmitted back to a central hub in the cloud where it is analysed and stored, and actionable insights are sent back to the device. For example, consumer and enterprise IoT apps that use an IoT cloud from Amazon, Google or Microsoft all follow this model of ‘train machine learning models and do inference’ centrally and push out analytics to the end-points. Industrial IoT platforms such as GE Predix, Siemens MindSphere and the Bosch IoT Suite follow a similar cloud model.

Cloud AI works well for many use cases as long as there is network connectivity. On your phone, Apple Siri and Amazon Alexa work very well if you are connected but become unresponsive when you are out of coverage area—you would have experienced this yourself first hand. Not just in India, in several places across the world, reliable network coverage is not a given.

Further, many applications such as autonomous vehicles and drones need to make real-time decisions with near-zero latency. The quality of immersive experience provided by augmented/virtual reality applications is marred by network and application latency. Remote sites, such as offshore oil rigs, have to rely on satellite communications and cannot easily transmit vast amounts of data to the central hub.

Industrial IoT devices generate humongous amounts of data and it is simply not possible to cost-effectively transmit and store all that data in the cloud. Currently, less than 1% of the data generated by smart devices is centrally analysed—there is no way the central servers can process all the data as IoT devices proliferate even more in the coming years.

Clearly, the cloud AI model does not work for many categories of usage scenarios.

In edge computing, data is processed and analysed at the edge or end-devices, close to the data source. Thus, the amount of data transmitted to the central hub is minimised. As data is not stored centrally, consumer data privacy concerns are also alleviated to an extent. And since there is no round-trip of data between the hub and the edge, performance time is improved.

Edge AI offers powerful digital transformation opportunities because of real-time analytics and the ability to directly analyse more contextual information. In fact, edge AI is a prerequisite for self-driving cars, mission-critical industrial IoT applications and even immersive consumer experiences. We can expect to see AI move from a centralised model to the front lines very soon.

Kashyap Kompella is the CEO of rpa2ai, an analyst firm specialising in automation and enterprise AI.

Close