As I reflect on the Internet of Things, I have been working on a personal definition of the capabilities of devices and objects that participate in IoT. I want to have a framework for thinking about how these objects interact with the world and each other.
Note: When I first wrote this back in 2014 (12 years ago!), the state of the art of automation and machine learning was very different than it is in 2026. We have crossed a boundary into a deeper exploration of "AI", lead by the push to use Large Language Models, generatitve systems, and agents. There have also been significant improvements to the SOA for robotics, with complex "humanoid" robots creeping closer to real-world applications. How much of this technology is likely to make it into real world applications at scale is still an open question.
What is certain is that there is a lot of buzz about Physical or Embodied AI and that the term IoT is somewhat out of fashion. Even though there is a clear throughline from one to the next, we have seen the creation of new buzzwords in an attempt to create excitement about what is not an overnight innovation, rather the result of decades of accreting knowledge and experimentation.
The idea of smart, connected physical objects, with computing at the edge and in the cloud, has resurged a bit, despite somehwat modest real-world success. Digital twins at scales from personal devices to cities, self-driving vehicles,advanced autonomous robots, and human-comparble machine intelligence (for some tasks) are all a reality, though not globally available and certainly not anything like mature. It will most likely take decades more development for theses technologies to become fully operable, whatever the boosters say. It is also impossible to ignore that they here, being deployed, and very disruptive, if largely due to perception rather than reality.
I think there a few distinct properties of IoT objects:
- Communicating
- Sensing
- Controlling (Actuating)
- Interacting
- Orchestrating
- Autonomy
The first property, Communicating, might be thought of as the minimum requirement for a IoT object. By definition, IoT devices are connected in some way to the larger Internet. There are many ways that these devices can communicate. They may broadcast over the public Internet, communicate via private networks, or something in between, but these devices are explicitly connected. They may be passive communicators—such as objects with RFID tags attached or embedded—and discoverable by near-field or long-range systems. They may also be active communicators capable of broadcasting information directly to other devices or networks.
The second and third properties, Sensing and Controlling (Actuating), might also be thought of as first-level concerns for IoT objects. An IoT object must be able to either sense a condition in its environment or control an acutator, or both. Without one or the other of these properties, an IoT object is essentially inert. IoT objects have direct control over the interface that connects to sensors or actuators. The interface itself might allow for extension, such as the standard power outlet of a Belkin WeMo Switch, or integrated directly into the IoT device, like the LED lighting in a Philips Hue bulb.
Passive, inert objects may still play a part in IoT, but this form of "ambient inventorying" is less interesting unless the network itself has relatively intelligent end points or sensors. For example, some RFID antenna systems can be set up to provide advanced, precise location information and other metadata about tagged items. However, this type of system is only useful when the tagged items are within the envelope of the antennas. The object has a deep dependency on physical location and an external system. While this is somewhat true of all IoT devices — e.g., some devices require a hub or available wifi to connect — RFID-type devices are particularly limited and bound to external systems.
Another key property of IoT objects is the ability to interact with each other. This can be thought of as anything from an IoT sensor sending data to a phone, to complex systems built on many IoT objects. For example, the Belkin WeMo Motion sensor can trigger the behavior of a WeMo Switch.
IoT object interactions may be direct and local — the switch in your house make connect to the bulb via LAN or PAN, avoiding a round trip to the Internet proper — or the interactions may be remote. When remote, interacting devices are often communicating indirectly, with a central system managing messaging from one device to another.
A common characteristic of IoT object interactions is their automated nature. The components, thanks to the integrated sensing and actuating capabilities, may be endowed with the ability to act autonomously. At the basic level, this can be accomplished with very simple rules, e.g. "when the rain sensor detects rainfall, it signals the window to close". As IoT devices get smarter , more complex intereactions become possible.
Orchestrating could be thought of as a deeper version of Interacting. An emerging class of IoT devices are appearing that can coordinate and manage the actions of other devices. An example of this would be a central home automation hub that mananges a network of other devices. The Sen.se Mother and Motion Cookies system is an orchestrated system.
Orchestrating becomes particularly powerful aspect of IoT devices as scale and interoperation increase. The more devices we have and the more connected to each other they are, the greater the possibiblity for orchestration and wide-ranging interactions and effects. When we add increased autonomy to these growing networks, IoT shifts from point-to-point or hub-and-spoke topologies, and begins to resemble a richer fabric.
Need to capture some thoughts about autonoumous behaviors of IoT devices.
I wrote the comment above in 2014, then never came back to it. A lot has changed about the possibilites for automony in IoT systems when we look at the specific technologies available, but the gist of the capability to act independently has not shifted that much.
Autonomy is the property of an IoT system to operate independent of outside assitstance to achieve one or more tasks. It would be fair to say that a fully autonomous IoT system would be a "set it and forget it" affair: fully agentive and self-regulating.
That said, there are clearly levels of autonomy that can be achieved. Many IoT systems incorporate human components-so-called "human in the loop"-to handle tasking, adjustments, validation, redirection, and critical decision making. We can consider such IoT systems as cybernetic systems, with humans acting as the hand on the tiller.
Autonomy may also be partial, where as system has ful control of an enviroment in one way, but limited control in others. For example, the autonomous nature of an IoT system might be limited to a virtualized version of the physical components, as in a simulation or emulation expressed as a digital twin; the system may be autonomous within the model, but restricted in it ability to affect the real world through actuation and orchestration. In such situation, the IoT system may really on other mehtods to put the physical system into the correct state, such as a human operator informed of the need to adjust the system in some way.
Based on the talk by Sanjay Sarma of MIT, I want to add something here about identity. The idea is basically that the baseline requirement for IoT is a tag (say, RFID), a sensor that can read that tag, and a digital record that corresponds to the thing somewhere online.
In other words, IoT may be indirect: the "thing" does not have to communicate actively; something else can do so on its behalf.