The Internet of Things and artificial intelligence are coming together and the edge is playing a big part.
Recently a colleague directed my attention to a post by Maynard Williams, Welcome to the Internet of Thinking. I was interested to see Maynard explore a world in which sensors and intelligence are tightly coupled together, instead of separate as is the de facto standard today.
This is also something I have believed for a while. Some years ago (2016), I wrote a forward-looking post on the Artificial Intelligence of Things. By this I meant that the intelligence derived from the data generated by Things would become so important that the evolution of IOT platforms would be head in that direction.
A year later, I wrote about IOT coming down from the cloud, to play a more prominent role on the edge. Now, in 2018, the consensus seems to be that both trends need each other.
I see a number of reasons, but here are the two most obvious ones:
- Autonomy: When use cases mature, in general more intelligence is needed. When more intelligence is added to digital processes that are supported by data derived from real-time production processes, the dependency on connectivity to the cloud becomes a liability. Production has to be able to continue even when under a DDOS attack causing the Internet connection to be interrupted. So the trend I am seeing now is deployment of fully functional IOT platforms on the edge (including advanced analytical processing capabilities like complex event processing) which previously where available in a cloud-only fashion, to fulfil these needs.
- Operational Efficiency: AI and other analytical tools, like machine learning and time series analysis, need vast amounts of data. Moving all this data into the cloud is a costly solution, especially if the data is generated in remote areas. The edge is desirable if there is a need to minimize solution latency, data traffic and data storage (in the cloud). The edge can help to aggregate and pre-process data for centralized enterprise-wide solutions. This means that the analytical tools are close to the source, with only the necessary data transferred to the cloud for enhanced insights.
Some word of warning is warranted: Much of the learning by machines can only be done if example data is available in abundance, and for this you need scale. The cloud is the ideal place to collect and train the models, and currently there is no good reason why this could be done better at the edge (except maybe for extreme security/data protection issues).
So, like Peter Levine in his February 2017 podcast, my opinion is that the cloud is ideal for modelling and training purposes. It gives you the ability to centralize the data collection and figure out the relevant patterns you want to train your advanced analytical models on. At the same time, I deem it very likely that the execution of the model, in operational terms the scoring against the model, is preferably done at the edge.
In order to make that happen, the only way forward is a distributed IOT platform, which can deploy the relevant components at the right place according to the needs of the use case. In order to reduce the complexity for developers it is absolutely mandatory that the technology used in such a distributed setup is the same, so that they don’t have to keep the deployment scenario in mind while doing the development. Something which some IOT vendors forgot about! But luckily not all…