How to get ahead of the curve when IoT converges with AI
IoT devices collect vast amounts of data, making AI essential for sifting through the noise to produce actionable insights. Here are four points you should think about when developing or deploying an AI-IoT solution.
Like unrefined gold ore, the raw data accumulated by millions of Internet of Things (IoT) devices means nothing without some form of analysis—also known as ‘data mining’. Rather than perform data mining manually, many organisations are marrying IoT with artificial intelligence (AI), essentially automating the process of distilling insights from what would otherwise just be noisy information.
During a discussion titled ‘Where AI Meets Industrial IoT—Key Considerations for Instrumenting & Integrating Intelligence for Business’ at the IoT Asia 2019 conference, a group of experts shared ideas on how to effectively leverage AI and IoT to create value. We bring you four key takeaways from the session.
Putting AI on edge
Typically, most IoT systems stream their data directly to the cloud, where the computational infrastructure for analysis also resides. As more and more devices come online, digital traffic becomes a problem.
“Because you are sending all the data [from sensors, for example] to the cloud, you end up choking up the whole bandwidth of the system,” said Mr Vishal Goyal, senior technical marketing manager for ASEAN, Australia, New Zealand and India, STMicroelectronics. To avoid this issue, organisations looking to implement IoT solutions should consider computing at the edge, which could mean building AI into the IoT sensors themselves to process the data collected there.
“As we bring AI down from the cloud to the devices, we are now only uploading the meaningful information,” explained Mr Goyal. “This reduces the bandwidth requirement, and not only does it make system implementation much easier, but also more scalable since it is less likely to be overloaded when more sensors are subsequently added.”
Moving data faster
Even though edge computing allows us to do more with the limited bandwidth, we would still eventually require greater network connectivity, especially for real-time decision making with IoT data, said Mr Damien Kopp, co-founder of Envolve Data. “With the ‘data-fication’ of any form of process and business comes the need to distribute and send the data collected, [and it] will continue to put some pressure on the networks,” he noted.
This is where 5G wireless network infrastructure comes in to play. Promising higher bandwidth and almost negligible latency, 5G would allow even more rapid and high-volume data transfer among devices, as well as between devices and the cloud. South Korea and the US have already launched commercial 5G wireless networks, and a host of other nations are looking to follow suit. Therefore, IoT solutions providers ought to make sure that their devices are compatible with 5G when it becomes available.
Finding the right fit
Given the buzz around AI and IoT, one of the myths that have surfaced is that such cutting-edge solutions are expensive. However, Mr Kopp noted that fit-for-purpose systems could be developed at low cost and also made available to small businesses—even to humble mom-and-pop shops.
For example, in the retail sector, one could obtain valuable data insights with a system built for just US$120, Mr Kopp claimed. Such a system would consist of a US$40 video camera to capture the movement of people (or product sales) and the video data can then be analysed at the edge with publicly-available software run on a Raspberry Pi mini-computer that only costs US$80. Therefore, by defining a specific use case, organisations may be able to benefit from AI and IoT without breaking the bank.