Sponsored content by: sponsored by

The limitations of IoT-AI integration

Friday, June 22, 2018

By Wojciech Martyniak, M2M/IoT Product Manager at Comarch

In Hollywood, artificial intelligence is often portrayed as something to be feared. It is synonymous with a world dominated by machines, in which humans are an endangered species fighting for survival. The truth, though, is far more benign. In the Internet of Things - a global network of connected devices in our homes, workplaces, cars and cities - artificial intelligence and machine learning processes operate, often unnoticeably, smoothing the path of everyday life and facilitating business in every industry.

The reason for the stark contrast between Hollywood's doomsday narratives and reality is simple: artificial intelligence operates according to goals set by humans, and, while machines are capable of learning, they do so within the framework pre-defined by these goals.

Data - the foundation of artificial intelligence

The foundation of artificial intelligence is data analysis. For example, organizations such as Google, Netflix, Facebook and Amazon all use algorithms that gather and examine data on user behavior, on which they base certain aspects of content delivery. Our banks and mobile network providers do something very similar - and in these cases, their systems are set up to compare data with established patterns and take certain actions if anomalies appear. In addition, consider the robot vacuum cleaners that have seen spikes in popularity, and even the advances in self-driving cars - in both instances, data are being gathered and analyzed, so that lessons can be learned and operations can become more efficient and safer.

IoT sensors and the power of AI for predictive maintenance

Where we have taken a huge leap forward recently is in harnessing the power of artificial intelligence not just to automate processes related to past and current events, but also to predict future activity for physical and virtual devices. For example, companies in various industries commonly deploy solutions that use information obtained from automated, in-depth analysis of network data over time and over vast geographical areas. These companies use devices, such as sensors connected to the Internet of Things (IoT), for predictive maintenance. Based on data analysis, the sensors predict the conditions under which a device is likely to malfunction - or simply wear out - and can then trigger actions automatically or issue an alert that gets an engineer on site before a problem arises.

Such solutions can also uncover interesting patterns in equipment data. For instance, by using a variety of sensors in the marine shipping sector, companies find means to correlate information from fuel meter readings and the amount of power used by on-board refrigerated containers. Data obtained in such a way can be used to optimize generator output, which led, for one company, , to savings of $30 per hour - or $6.5 million over the course of a year.

Similar IoT-based sensor solutions can be employed on production lines to monitor and predict potential issues. Healthcare providers are already using remote monitoring for patients, via devices that give early warning of life-threatening changes in vital signs. Transport and logistics operators keep track of goods and vehicles with solutions that can also learn routes, personnel availability and more. Even entire cities are becoming smart, with applications such as IoT-based and AI-powered smart-camera systems that can count vehicles or recognize license plate numbers.

The limitations of IoT-AI integration

The complex integration of IoT and AI via data analytics also has some limitations. Often, using self-learning AI to extrapolate all the required steps to achieve a certain goal is simply counterproductive - the process can be extremely time-consuming, especially in cases of new or large data sets.

In these instances, human intervention is required to optimize learning mechanisms and direct the solutions toward the required operational capabilities. It is quicker, easier and more cost-effective to give a device or network a pre-defined, algorithm-based course of action, which still automates and optimizes operations but saves money by cutting the time required for the machine to examine its rules, and subsequently to self-learn and act accordingly.

Consider an example of a logistics company that needs to account for changes of home locations: in large data sets, having AI figure out something like this may take a significant amount of time and would affect the quality of services delivered. For humans, this fact is so self-evident that it immediately springs to mind when considering logistical issues. Applying a simple algorithm - stipulating that going back to one location three times and staying there for more than two days updates the home location - is the correct course of action in such cases.

Comarch IoT Connect

Being able to manage all of these aspects - data analytics, IoT devices, AI and meaningful human intervention when necessary - is what will set key players ahead of their competition across many sectors. At Comarch, we implement this philosophy through Comarch IoT Connect, where we test the feasibility of various AI approaches (DBSCAN, K-Means and Linear Regression) within its framework. Results show the AI approach to be very effective in several areas, including service quality assurance and predictive maintenance - but particularly so in the anomaly detection field, where AI is trained to detect unexpectedly high (or low) device activity in a specific location. However, in many other instances - such as the case of changing home locations described above - we settle for a simpler algorithmic approach. In these instances, implementing AI takes much longer, costs more, and yields negligible benefits.

A platform designed in such a way offers the optimal solution to facilitate the business strategies of IoT service providers, effectively tailored for the needs of any given organization/industry. Its architecture is not only flexible and scalable, but also integrates smoothly in multi-national, multi-level and multi-operator environments. An award-winning solution respected by industry experts at Berg Insight and Gartner, Comarch IoT Connect has already been implemented by major players such as Telekom Austria Group and Saudi Telecom Company, to optimize IoT connectivity management and assist in generating offers to verticals in the automotive, consumer electronics, retail, energy, finance, healthcare, manufacturing, transport and security sectors.

Such solutions clearly have the capacity to revolutionize the way we work, do business and live in our everyday environments. And while machines can certainly learn, AI still relies heavily on a strong foundation of real, human intelligence - which is why we're unlikely to see the Hollywood interpretation of the AI world any time in the near future. Then again, it wasn't so long ago that mobile telecommunications - let alone 5G and the IoT - were unimaginable ...