IoT has made it easier to connect physical devices up to digital platforms. Sensors can be used to capture temperatures, vibrations, location, movements, pressures, status of machines or other information about the real world.
However, getting data to understand and create actions are still difficult problems.
This is where AIoT (converging AI and IoT) becomes interesting.
From Capturing to Intelligence
An IoT architecture could resemble something like:
Device -> Sensor -> Connectivity -> Data Platform
Which could give insight into what is going on in the physical world.
An AIoT architecture builds on top of this by adding AI/ML capabilities:
Device -> Sensor -> Connectivity -> Data -> AI/ML -> Insight -> Decisions -> Action
Being able to process raw data from the physical world enables extracting deeper insights.
Context Matters
In many industrial settings, systems operate with data from a variety of sources.
A manufacturing environment for example, could have data coming from machinery, the production line, vehicles, location, inventory systems and a variety of sensors.
By looking at this data in isolation, context is lost.
An AIoT approach would unify this information to give a better look at the actual situation.
For example, a machine could capture data about its own vibrations.
Combined with production insights about the current load of that machine, and other information from the environment, such as temperature changes and historical averages, a more complete picture emerges of what is happening.
That can help identify irregularities, or provide a reference point to understand what is going on in relation to other data points.
Building an AIoT System
Creating an AIoT system requires more than just building a model to consume data from IoT devices.
There are several steps involved in the pipeline:
1. Collecting Data
Sensors or connected devices need to provide data of value or interest.
2. Connectivity
Making sure that the data captured can reach the systems that will process it. This may involve using different connectivity approaches depending on the scenario.
3. Processing
Massaging the data so that it is consumable by downstream systems.
4. AI and Analytics
Identifying patterns, classifying data, anomaly detection or other approaches to get value out of the data.
5. Decisions and Actions
Ultimately, decisions need to happen for actions to be taken once there is value in the data. This may involve presenting information to an operator, or directly interacting with other systems. Without this step, the system may not provide concrete value.
AIoT systems can be seen as multi-disciplinary systems that need to deal with software, data processing, connectivity, AI, and the realities of the physical world.
AIoT and Physical AI
AIoT is related to the new concept of Physical AI.
As described by some in this interesting article, while most software today deals with logical information, Physical AI deals with systems that operate in the physical world.
Physical World -> Sensors -> Connectivity -> Data -> AI -> Decisions -> Action
Physical AI builds a feedback loop between the physical and logical worlds.
From a developer's perspective, building software around Physical AI (and by extension, AIoT) requires thinking about not just the accuracy of their models, but also the impact of latency, data quality and other variables.
Additionally, system design involves questions about how to make decisions, how to operate under constraints, what is the impact of making the wrong decision, and how the system can operate reliably in the real world.
Starting from the Problem
When designing systems, one of the most common mistakes is to think about the tool first, and the problem later.
In the case of AIoT systems, it is important to think about the physical situation and what needs to happen.
Some possible questions to guide the design of a system include:
Physical Event to Understand What type of data do I need to capture? How often do I need to capture that data? Where will that data be processed? What decisions does the system need to make? Who or what will consume that information? What actions do I need to take?
By focusing on these questions, it is possible to avoid capturing data that is not needed, or adding unecessary complexity to the system.
Organizations that are building ventures around solving physical problems are looking at these questions. Aperture Venture Studio is an example of a venture studio that focuses on bringing AI, IoT and related technologies to industrial and physical systems to create value.
The Future of AIoT
AIoT systems ultimately converge the worlds of sensors, data, decisions and actions.
They enable systems that provide value beyond simple observations about the physical world.
For developers, the space represents an interesting opportunity to build systems that operate at the intersection of cloud and edge software, sensors, AI, and the physical world. As these systems evolve, the question "How can we connect this device?" will be replaced with "How can this system sense and act upon the world?"
That is the future of AIoT.
Top comments (0)