A decade ago, cloud computing was creating a ton buzz in the technology marketplace. Today there is a similar amount of excitement around edge AI.
Edge analytics is a fundamentally different approach from what we’ve seen in previous years. During the cloud era many enterprises focused on creating centralized cloud-based data processing and analytics systems. In contrast, edge analytics architectures store, process and analyze data in real-time. This approach offers benefits like reduced network burden and connectivity costs, reduced storage and database management costs and, most importantly, real-time data crunching and analysis.
There is huge potential for this emerging technology across industries and a mix of different IoT applications.
For example, an energy management company using different sources of power — such as renewable, traditional grid generators and battery — can use edge analytics to optimize use. If solar energy supply falls in one location, the system automatically increases power supply from another source. If a cheaper source is available, the system shifts to a more optimal source for that location. A similar value can be seen in medical tracking and monitoring tools, which provide the patient and a medical professional with instant notifications if any issues arise.
Edge AI is ideal for instances in which mission critical data requires near real-time decision making at the device level. In these cases, the end device is not only a collector of data, but an almost autonomous system often equipped with machine learning, so that is to be able to make decisions on its own. Connected cars and vehicle-to-everything, or V2X, communications require edge AI capabilities for automated hazard warning, collision avoidance and congestion avoidance systems. These instances can’t afford the time lag thatdata takes to travel up to the cloud for analysis to make the right split-second decision.
Edge AI is also effective in cases where constant cloud connectivity is simply not available. The ship engine agnostics company MAN PrimeServ uses AI on the edge to monitor and evaluate data from ship computers on-board while at sea, because sending this mission-critical data to the cloud would be too expensive using satellite connectivity. When a ship returns to shore, computers transfer the data to the cloud using cellular connectivity.
While there is clearly great potential in edge AI, organizations must also overcome challenges. The biggest question organizations must answer is whether IoT devices that are already deployed have the capability to become more intelligent. That is, are they able to support edge AI applications and take actions based on that intelligence?
The other major concern with edge AI is security. When you give end points more control over data, they become a target for cyberattacks. To overcome this issue, organizations can equip SIM cards with solutions that would improve device authentication, network policy controls to limit what data sources the devices can reach and ensure security for IoT data in motion.
Solving the security issue alone will require a lot of additional compute, and the more functionality you put on the end point, the more power it will consume. All of this requires significant investment, so you need to determine if the benefits of real-time intelligence and decision making will outweigh the cost of turning existing IoT solutions into edge AI solutions.
Ultimately, it all depends on the use case, including desired objectives and value creation gained. If real-time decision making isn’t required, the investment isn’t likely to be worth it. If it does, you may be better off using an IoT gateway as an intermediary between the existing IoT end points and deploy edge AI on that gateway instead. In industrial IoT applications, this gateway sits between the machine-to-machine endpoints on the factory floor and the cloud. The machine-to-machine endpoints can then leverage the data analytics and decision-making capability of the IoT gateway in real time.
Because we’re dealing with nascent technology, it’s important to balance the excitement with some pragmatism. The cost of adopting edge AI may outweigh the benefits of real-time intelligence and decision making in some use cases, so this is the first point to consider. You must ask other questions to before adopting edge AI including: what do you want your IoT application to do – just collect data or make decisions; does that decision making need to be immediate; or do you need data analytics on an hourly, weekly or daily basis?
All IoT Agenda network contributors are responsible for the content and accuracy of their posts. Opinions are of the writers and do not necessarily convey the thoughts of IoT Agenda.