I am currently reading the book “IoT – the hype no one knows about” by Afzal Mangal. A short book that I digest in chunks and analyse chapter by chapter. In the book, Afzal uses parallels between AI and IoT, suggesting that AI has a steeper success curve than IoT. The question is why — and, of course, is it really true?
In my opinion, IoT is a name for a set of disruptive technology solutions primarily used for digitalisation. In its purest form, the solutions would be, for example, sensors — devices aimed at creating insights, insights that can transform businesses and create a better world. Yes, you are right, the transformation can lead to cutting costs, saving the planet, or completely transforming companies into new business models that provide better value while also saving the planet. Yet IoT’s superpowers sometimes fall short of the AI hype.
ChatGPT isn’t AI — it’s just the face of the hype
Let us just understand two things about AI. AI is not just ChatGPT, but ChatGPT has been a huge success factor for anything AI-related. Can we name similar successful IoT projects? In short, no — not yet. Or perhaps we can, we just do not think of them as IoT devices. Apple Watch is one such device, with a huge user base, but it is labelled a smartwatch, not an IoT device — even though it is like a sensor on your wrist.
Back to AI. ChatGPT is a useful tool, I have to admit that — anyone can use it — but ChatGPT still struggles with illustrating humans with the right number of fingers or totally throws out the wrong answer, and we’re somehow okay with that.
An example: I asked ChatGPT to summarise a web page as a tag cloud in an image. Got an image that seemed reasonably right.
“Do you want it in other colours?” My answer was yes, and I provided the colours in hex format and the fonts. I got an image of butterflies in return.
I asked what it was doing, and the response was, “Oh, it was more *fluttering butterflies* than *buzzwords in tech*! 😅 I misinterpreted the question.”
If I were to send out a sensor that essentially guesses values or doesn’t know its own boundaries (reporting 500-degree outside air temperatures), I’d be out of business in no time. Still, we are somewhat OK with the algorithmic overconfidence and machine-assured guessing. So is AI more mature than IoT? The answer: no. The difference is that AI hit the ground running with the launch of a ready made solution that was easy to use, but IoT as a concept was introdued before anything successful and useful on a broader scale was in place. And if it was, it was not labeled an IoT solution.
IoT and AI: Better together
Even though AI is riding a steeper hype curve than IoT, they still need each other more than you think. Few phrases are as transformative as “AI meets IoT.” The combination of the two technologies is a paradigm shift. IoT generates loads of data, and AI loves to analyse data. The two are highly complementary: one collects, the other interprets. IoT throws off massive amounts of raw data — way more than any person could handle — and AI is built to chew through that kind of volume. On their own, IoT just collects, and AI just analyses. But together? That’s where the real value starts to show.
Beyond the graphs — data meets storytelling
IoT data provides valuable insights, and devices send data that is eventually graphically presented to the user. However, data and diagrams are not always easy to understand, and they have to be interpreted by experts. By making data more understandable, we create a more equal society where more people can benefit from data — for example, from sensors.
In an early experiment, we wanted to see if data would be more understandable. We decided to combine the superpowers of the Internet of Things with the superpower of AI.
AI can find patterns, flag issues, and help make sense of the noise in theory — but what about in real life? In a test with a partner, we let IoT sensors collect data from our office. Then we let AI interpret the data and present it in an understandable way. The focus is no longer on presenting values as diagrams or bars, which far from everyone can interpret, but on presenting easily understandable data that everyone can grasp.
Here’s an example of how the data can be presented:
Summary of the Last 24 Hours
Office:
• Battery Level: showed significant variation between 11% and 87%, with an average of 29% at night and 42% during the day.
• Temperature: ranged from 19°C to 23°C, with average temperatures around 21°C.
• Humidity: was between 35% and 44%, with average values near 40%.
Outdoors:
• Battery Level: constant at 100%.
• Temperature: was significantly lower, between 4.5°C and 12°C, with a marked difference between day (average 8°C) and night (average 6°C).
Key Observations:
• Indoor temperatures are somewhat higher during the day, which may be an effect of sunlight or building usage — but it is within the recommended limits for an office.
• Outdoor temperature is significantly lower at night, as expected.
• Indoor humidity is relatively constant between day and night.
This should be taken into account when adjusting heating and air conditioning settings to optimise comfort and energy consumption.
The above is quite rudimentary, but it was our first try. With data available, AI can provide understandable recommendations, summarised as text, which is clear and easy for everyone to understand.
When AI finds patterns sensors collect
In a test in Linköping Science Park, sensors revealed that the indoor air quality index (IAQ) increased at the same time every day — something we discovered with an AI analysis. Although the particle levels and CO₂ concentrations in the premises were within normal ranges, a significant increase in TVOC was detected, which can indicate the presence of chemical compounds in the air. It was discovered that the cleaning company had been using the stronger cleaning agents previously used during the pandemic, inadvertently, on a few occasions. Thanks to the insights we provided, the cleaning company acted promptly and reverted to using products better suited to the current post-pandemic work environment.
AI can make data more accessible, understandable, and valuable to more people within an organisation — but AI needs data in order to provide the right insights.
AI is in the house: When basil meets big data
The Swedish Public service TV company, SVT, recently aired a feature about Svegro, a company using IoT and AI to grow food with strong results. In one of Sweden’s largest greenhouses, herb cultivation is monitored and images are analysed by AI, which knows exactly what the plants need.
On Ekerö, six million basil plants are grown each year. They are monitored by cameras, and an AI model determines when it’s time for harvest. According to Svegro’s CEO Kristin Orrestig, the AI model has helped reduce food waste and cut energy consumption by 15 percent. One of the main benefits is that AI avoids overcompensating the plants — something people tend to do. We might keep it a bit warmer than necessary, water a little too often — but AI is spot-on in diagnosing what’s truly needed. I can imagine that for more valuable assets, one could use IoT devices such as sensors.
Smarter decisions, faster -intelligence at the EDGE
The integration of AI into IoT goes beyond data analysis. AI is increasingly being embedded directly into IoT devices, enabling even simple devices to make smarter decisions locally. By processing data close to its source, latency can be minimised — a critical factor for applications that demand uninterrupted operation and rapid response times. Think autonomous vehicles or industrial automation, where high reliability and resilience are essential. In smart city applications, for example — where sensor networks control traffic lights or monitor air quality — local AI algorithms can quickly process data and act without relying on or overloading centralised servers. This also improves energy efficiency, since data processing occurs closer to the source and less energy is spent on data transmission.
Edge computing is particularly well-suited for distributed AI systems that require decentralised decision-making. However, despite its advantages, edge also comes with limitations. It’s not always ideal for handling complex models that need continuous updates from centralised data sources. AI models are typically trained on aggregated datasets — something edge systems rarely support. In addition, smaller edge devices often lack access to advanced resources such as GPU acceleration, making compute-intensive tasks less efficient when run locally compared to in a data centre.
The more we want edge devices to handle, the higher the hardware cost becomes. Therefore, at present, local AI in a basic IoT device should be seen as a potential enhancement rather than a replacement for cloud-based AI. A hybrid approach — combining the real-time benefits of edge computing with the large-scale analytical power of the cloud — enables a balanced and flexible AI ecosystem that can meet both local and global data processing needs.
Can we trust AI decisions if we don’t understand them?
When AI meets IoT, it creates a powerful combination. But with this development comes a significant risk: bias — systematic distortions in data-driven decisions. Bias in AI systems can occur at several stages. The most common is during data collection — sensors that don’t accurately reflect reality, a lack of representative data, or incorrect labelling.
If an IoT system is trained only on data from specific geographic areas, age groups, or environmental conditions, the model may draw conclusions that aren’t universally valid. During the analysis phase, bias can be amplified by algorithms that lack explainability or prioritise optimisation over fairness. Even the interpretation of results carries a risk of human bias if decision-makers rely blindly on AI without understanding its limitations.
Addressing bias in IoT-based AI systems requires multiple actions. It starts with ensuring diverse, high-quality, and representative data from sensors. Standardising data collection, metadata, and transparency in how data is processed is essential. Furthermore, explainable AI models (XAI) are needed to make it possible to understand why a conclusion is reached — not just that it is. Continuous evaluation of algorithm performance in real-world environments is also crucial, along with combining technological development with ethical guidelines.
IoT and AI will shape the future of decision-making. But for the technology to be fair, reliable, and useful, we must treat bias as a design challenge — not an afterthought. By systematically working with data quality, transparency, and accountability, we can create intelligent systems that both perform and include.