The advent of publicly available generative AI tools like ChatGPT, Claude,
and Perplexity brought with it a massive wave of excitement and
investment. Amazon, Google, Microsoft and Meta have all placed large bets on AI
partnerships, acquisitions, and branded AI capabilities. As excitement took hold,
industry analysts and participants projected global investments reaching over $200
billion globally by 2025 and over $1 trillion overall.
Kenta Yasukawa, Co-Founder and CTO, Soracom explores further.
More recently, as the initial hype dies down, observers have begun to express scepticism regarding when, or even if, those investments will pay off. For those of us who have been working within the Internet of Things for some time, this cycle of lofty expectations and inevitable recognition that change will take time, ingenuity, and hard work may sound familiar.
Yet IoT technology is in fact more deeply embedded than ever before across industries from finance and healthcare to logistics and transportation. While the technical innovations and business impact have been tremendous, most applications still tend to stop at an ‘Intranet of Things’ stage, where devices communicate primarily with data repositories inside an organisation.
The term IoT originally envisioned a truly connected world, where things could interact and collaborate with each other to improve our lives and environments. Collecting data securely and remotely actuating devices is a step in that direction. The emergence of Generative AI (Gen AI) tools provides the missing piece: an enabling technology that can make decisions based on incoming and historical data and trigger things to take actions as appropriate.
Wasn’t it doable with traditional machine learning by training models? Yes, however, one big difference is that Gen AI can perform the job without training models. Even when a new use case is added to the system or a new set of parameters are added, you would not need to re-train your models.
Gen AI is fundamentally designed to synthesise data and analytical inputs and make, or guide decisions based upon that data. The potential of this computational process goes well beyond what we’ve seen to date (i.e. chatbots, image or video synthesis, etc.). Because Gen AI can analyse data and images to make suggestions or even decisions and remotely control devices based on natural language and instructions without training machine learning models, the opportunity for applications in IoT is limited only by our imagination.
As one example, Gen AI now makes it possible for even non-technical members of IoT project teams to create applications that connect data, events, and actions using only natural-language prompts. In a manufacturing environment, this might look like a process that actively monitors real-time video data to identify potential cases of missing protective equipment on the factory floor, or even to place an emergency call when a fall is detected. What until recently might have taken weeks of complex coding can now be developed in hours.
Similarly, we know that Gen AI can rapidly label data to accelerate training and testing for Machine Learning (ML) models, but it can also be used to make effective decisions on the fly with zero previous learning required. This has immediate applications in manufacturing, logistics, and even large-scale HVAC. Once human operators confirm the improved HVAC control, an ML model can then be trained accordingly to reduce the computational cost in achieving the result.
Another promising application of Gen AI for IoT is the ability to generate code for effective management of large data sets. In 2023, analyst firm IoT Analytics wrote about the power of code generation for IoT through large language models. Today, Gen AI assistants like GitHub Copilot are already beginning to help technical staff develop and optimise their own code more quickly and let them focus on creating new tools rather than performing rote tasks.
As an example, one common problem organisations face in IoT is how to collect and leverage rich data sets while complying with privacy regulations like GDPR. AI-generated code can effectively anonymise large data sets while maintaining their statistical validity and segmentation to preserve privacy while generating new insights and even new profit centres. Another example is to use AI to generate complex SQL queries to a large data set on behalf of humans so anyone can get insights from it.
As we have experienced in IoT, it’s when the hype dies down that the real work begins. GenAI is not a feature in and of itself, but its transformative impact will be felt as tomorrow’s IoT architects roll up their sleeves and get down to the business of building. We may not see an overnight ‘AI boom’ any more than we saw an ‘IoT boom,’ but as those architects integrate Gen AI capabilities to optimise large deployments, improve connected experiences, and solve real problems for end users, we can prepare for a step-function advance into the next phase of IoT.
Author: Kenta Yasukawa, Co-Founder and CTO, Soracom
This article originally appeared in the October 24 magazine issue of IoT Insider.
