Intelligent Cloud services & Smart IoT
IoT devices started entering the early adopter’s lives a decade ago. This first generation was, of course, connected, but usually only to a dedicated app or controller. A connected light would only respond to a single switch, and a Bluetooth heart monitor would send data exclusively to a mobile phone that would only process it locally. Today’s devices promise much more sophisticated behaviour that requires them to exhibit intelligence. For a modern consumer to consider a device truly worthy of the adjective “smart”, Cloud-based AI is also required. We build beneficial services that involve IoT devices with both centralized infrastructure that uses big data in the Cloud, and processing on local devices and the latest AI techniques must enable self-learning. Indeed, as with biology, intelligence is also defined by a capacity to adapt. We’ll see that distributed intelligence is the architectural goal. IoT devices need to work together intuitively and be extremely simple to set up. Connectivity is a central promise here, and the service operator has an opportunity to help devices interact seamlessly with both users and other devices. Indeed, the local operator can even leverage an advantage here by providing pre-configured devices that work out-of-the-box unlike their retail versions that require at least some configuration.
So, what is IoT intelligence?
I like comparing a sophisticated IoT setup at home, made of many IoT devices, to a colony of ants, where each can only perform a straightforward task, but the whole is resilient, showing intelligent behaviour. The ant colony is autonomous, and the intelligence is in the individual’s DNA. Modern IoT devices have the advantage of also being able to leverage a backend for analytics and other processing. Straightforward examples of IoT intelligence include having the lights on and the temperature set at home when arriving from work. The self-learning system could, for example, adapt to the fact that on Friday’s people come home earlier or use biometric sensors to offer AI-driven sleep technology. Use cases that may need backend analytics include identifying security risks on a device or ever-improving speech processing.
An operator-enabled IoT service is uniquely suited to exploit and benefit from centralised processing. The operator can also play a crucial role in securing devices, with even stronger security coming soon from blockchain-based solutions.
The behaviour of an IoT environment today is intelligent, from a user’s point of view, because it obeys rules they can understand. The challenge is for those rules to be managed with simplicity and to be readily accessible to users. A premise for that is, of course, having a user-friendly way to capture user intent into a set of rules. Part of the perceived intelligence of the whole system also then comes from its flexibility to easily incorporate new devices with seamless connectivity, offering another low hanging fruit for service operators. Implementing IoT-friendly protocols in their home gateway, for example, makes connecting the new device that much more seamless.
Intelligent behaviour is sometimes just simplicity.
Installing a new connected light bulb, for example, must be a trivial task. The basic rules must be available on the home network so that such tasks are possible even if there is an Internet outage. Technical aspects like finding the right radio and network protocols and identifying the IP address must all happen behind the scenes. However, an “intelligent” IoT setup also recognises when a device’s password is too weak or maybe not even set. The system should be able to qualify what level of maintenance is likely to be required and what the new device’s expected lifespan might be. Once new devices have been identified, they must be managed so we must be able to discover their capabilities. It is only then that they can be added in a meaningful way to scenarios that now include voice control.
Where is the intelligence?
The available processing power of the myriad of devices entering both people’s homes and factory floors alike is growing in earnest. The power of any single device is higher than a powerful PC 20 years ago. With improved protocols, this now represents the third wave of the computing model. With 50BN devices coming online, the network is going to have to be “intelligent about data”. Processing needs to happen closer to the source enabling “edge intelligence”. The Fog paradigm sees application come to data, no longer the other way around. One good use for the local processing capability is to filter and aggregate data before it is sent up to the Cloud. A security camera can, for example, send just occasional frames when the image is still, but send much more data when there is activity in front of it.
5G networks decentralise. The need to lower energy consumption and to make use of new high-frequency wavelengths in the radio spectrum implies smaller cells than in previous cellular generations, and the equipment in a cabinet at the base of any modern cell tower has both storage and processing capabilities. After years of promises, 5G can help edge processing to start delivering real value.
Massively resource-intensive processing still requires Cloud infrastructure to manage voice recognition as well as natural language processing. Cloud-based solutions also offer the only cost-effective architecture for pattern recognition, where the learning process picks up elements from multiple sources. There are use cases where a simple proxy in the Cloud can make a solution more robust, for example, to buffer commands when a user is travelling towards home, but this can be a simple link without real intelligence.
Distributed Intelligence is about the interaction between different agents, enabling automation that needs contextual intelligence. Multi-system interoperability will make for much smarter systems. In our White Paper last year, we discussed Edge Computing in detail, which although a different issue, poses similar architectural challenges.
With today’s leading IoT solutions, the “intelligence” isn’t so much spread out as the decision making is pushed out to the edge of the network. The centralised part uses analytics to improve the autonomous behaviour of devices by improving their rules.
A set of rules might, for example, require accessing both the temperature with the outside lighting conditions and accessing a personal calendar to optimise the heating system by autonomously creating an exception when the family is on holiday. Simple things can be resolved locally, but complex ones need remote computing power and access to other information sources. Cloud-based solutions can seamlessly interact with other countless other Cloud-based systems. For example, just starting with the random letter ’N’, the Cloud services of Nest, Netflix or Nuance can cooperate to deliver a smart service.
Some scenarios need a mixed approach like with security management. Physical security with a connected door lock today requires a local hack-proof architecture but detecting a DDOS attack on a security camera can only be done from the Cloud. A proactive monitoring approach should, for example, be able to identify a security flaw if the connected camera’s login ID is “admin” with a password set to “admin”.
Although connectivity is orders of magnitude more reliable than it was just a decade ago, it is still not stable enough for mission-critical tasks. Those of us with home-fibre, the most reliable connectivity to date, still suffer outages from time to time, and the penetration of this access technology still has a long way to go. We still require local processing for these, and it is unacceptable not to be able to turn a light on or off or print a document locally if there is an issue with Internet connectivity.
In the end, intelligence must be both local and global, and some call it ambient. SoftAtHome has been investing for many years in reaching the right compromise where central processing complements local solutions, but also where local processing and storage make for more responsive and robust home solutions. SoftAtHome’s Wifi’ON and Things'ON products both use algorithms running in the Cloud to improve those in home-devices, while Secure’ON uses different DDOS-protection components in the Cloud and on operator devices that collaborate to protect against attacks.
At SoftAtHome, we are always striving to position each component of a solution in just the right place, for optimised architectures and the most cost-efficient solutions available.
