The merging of artificial intelligence (AI) technology with IoT systems is already impacting businesses and consumers, securing huge benefits for all. IEC International Standards help to overcome some of the unavoidable flaws.
AI-enabled IoT systems already enable businesses to better monitor the supply and demand chain and deliver goods, with less human intervention, which in turn generates cost-savings. Such systems detect safety issues in smart manufacturing plants and are increasingly expected to automatically deal with them.
Called the Industrial Internet of Things (IIoT), these systems not only enable plant managers to identify faults that may not have come to light, they also save precious time – a bonus for employees and workers, as well as customers.
In the home, these intelligent systems will be expected to enable consumer connected devices to not only notify the owner or ring alarms, say when a fire starts in the oven, for instance, as most smart systems already do, but also shut off the oven, the whole power system in the home and call the fire brigade.
Using powerful algorithms and machine learning software, the Intelligence of Things will allow a multitude of areas and businesses to not only detect problems but act on them and solve them with minimal human intervention.
Quality over quantity
One of the drawbacks of these new systems is that machine learning is only as good as the data provided. IEC and ISO together develop international standards for information technologies through a joint technical committee (ISO/IEC JTC 1). “It’s the rubbish in/rubbish out quandary.
If you feed a learning system data that is corrupt you will not have a good result no matter how powerful the algorithms are. But that is where performance standards can help, by enabling users to monitor the quality of the data, for instance”, explains François Coallier, who chairs one of the subcommittees of JTC 1 which prepares standards for the IoT.
It publishes several key documents that help to standardize the industries impacted by the IoT and the intelligence of things. For instance, ISO/IEC 30141 provides a global reference architecture and common vocabulary for the IoT.
Algorithms can be biased
Algorithms are only as good as their developers. Machine learning can reproduce sexist and racist bias from the real world. Examples include image recognition software that fails to identify non-white faces correctly. This occurs when the scientists who develop the algorithms unwittingly introduce their own prejudices into their work.
Biases can influence the way a medical sample is collected by not including some members of the intended statistical population, for example. This could result in building an algorithm used for medical diagnosis, trained only on data from one subset of the population.
ISO/IEC JTC 1/SC 42 is addressing many of these concerns. It is looking into a wide range of issues related to trustworthiness as well as robustness, resiliency, reliability, accuracy, safety, security and privacy within the context of AI.
Read more in The Intelligence of Things.