Encapsulating the world of AIoT (AI+IoT=AIoT)

“Any sufficiently advanced technology is indistinguishable from magic” — Arthur C Clarke

We all must have come across two trends that are dominating the technological industry, Internet of Things (IoT) and Artificial Intelligence (AI). The convergence of AI and IoT is all set to redefine the future of industrial automation. First let us try to understand these two technologies independently, how do they work and finally integrating them will be like adding icing on a delicious cake.

Artificial Intelligence (AI)

Artificial Intelligence is ubiquitous in today’s world. It can be found in your smartphone in the form of a virtual voice assistant which is always eager to solve your queries on hearing a wake-up call like “Hey Google”. Upon browsing a website, a chat-bot popping up to offer better customer experience (sometimes much to our annoyance). The recommendations you receive on Netflix helping you select a suitable web series for spending your holidays fruitfully (while binge watching). Alongside AI, certain terms like Machine Learning and Deep Learning are used interchangeably. So, lets clear the distinction between these terms…

This Venn diagram will provide a little intuition about how these terms are related to each other. First coined in 1956 by John McCarthy, AI involves machines that can perform tasks that are characteristic of human intelligence. While this is rather general, it includes things like planning, understanding language, recognizing objects and sounds, learning, and problem-solving.

AI is defined as the study of “intelligent agents”. Now arises the question what is an intelligent agent? Anything that perceives its environment using sensors and takes optimal actions that maximize its chance of successfully achieving goals. Machines get the ability to think and act like humans as well as make rational decisions.

Machine Learning (ML)

Machine Learning, a subset of AI, is one of the ways to achieve it. On hearing this, my friend asked me are there other ways to achieve AI besides ML? Yes, there are Rule based Expert Systems but they involve millions of lines of codes with complex rules and decision trees.

The above proverb explains that providing a solution can only solve the present problem in a particular environment. But if the principles are learnt that lead you to the answer then you will have the ability to solve any related problem. Extending this philosophy to computers, Arthur Samuel defined Machine Learning as the ability to learn without being explicitly programmed.

So instead of hard coding functions in order to accomplish a particular task, the computer learns an algorithm on its own. The primary aim is to allow the machine to learn to take decisions automatically without much human intervention and adjust actions accordingly.

The system relies on observations or training examples in order to identify patterns, extract features and become capable of drawing inferences and making predictions when they encounter a similar situation. Thus, Machine Learning includes certain statistical techniques that enable machines to improve at tasks with experience (just like us).

Deep Learning and Artificial Neural Networks (ANN)

Now when we are mimicking the human attribute of intelligence in a machine, why not simulate the way our brain operates in the form of a mathematical model?

This gave rise to the study of Artificial Neural Networks. Our biological nervous system comprises of multiple interconnected units called neurons, which collect inputs and based on electro-chemical reactions and connections between the neurons, information is passed and we gain an understanding of our surroundings.

Inspired by biological systems, ANN comprise of “nodes” (artificial neurons) part of discrete layers and having connections with other “neurons”. Each layer extracts a specific feature to learn, such as detecting edges while performing face recognition. It’s this layering that gives deep learning its name, depth is created by using multiple layers as opposed to a single layer.

Comparing the approach of Machine Learning with Deep Learning

As the above quote signifies, due to the hype you should not straight away dive in for deep learning (pun intended). Depending upon the data-set and application, even a simple ML model might do the trick.

Conventional ML is typically used for projects that involve predicting an output or uncovering trend like stock price prediction or e-mail spam filtering. In these examples, a limited body of structured data is used to help the machines learn patterns that they can later use to make a correct determination on new input data.

Let’s have a peek into how humans gather data…

Sight or vision is the most developed of the five human senses. We use it daily to recognize our friends, detect obstacles in our path, signs to guide us, to complete tasks and to learn new things. Another developed ability of humans is our understanding of languages which help us to communicate with each other. Extending these capabilities to machines opens up the doors for two fields namely-

1)Computer Vision — Focuses upon capturing and storing an image, videos or any visual data, and then transforms those frames into information that can be further acted upon.

2)Natural Language Processing — Enables a computer to communicate successfully in human language using text, speech in different human languages. For example- the functioning of “See translation” we often use to understand a person’s comments written in a language we do not know, on social media.

Here the data in the form of images/video frames (comprises of millions of pixel intensities) or text which is a collection of words or audio is a bunch of wave signals. The point to be noted is the real-world observations tend to be enormous and unstructured. Thus, it is a tedious task to extract features to be supplied to a machine learning model. And this is where deep learning comes to our rescue by automating the process of feature extraction using multiple layers to extract features partially in each layer.

Internet of Things

In the journey of artificially building a nervous system an integral component is being left out. Our sense organs also play a crucial part by providing us with the ability to perceive our surroundings. We can also communicate with other fellow humans, form social networks, distribute larger tasks between our teammates. Can these abilities be added in machines or in more general terms in the objects/devices surrounding us?

Internet of Things has a lot to offer for this purpose. Machines can collect data through sensors which act as sense organs in this case. Cameras can be thought of as the eyes, mechanical sensors (temperature, pressure, etc) as sensitive skin and microphone as ears. There has been a lot of advancement in sensor technology and recently there has been tremendous research in the development of smart sensors.

Our devices or “things” are now connected with each other via the internet enabling them to talk to each other (Machine to Machine or M2M communication). A host of network protocols like Wi-Fi, Bluetooth, Zigbee, BLE and other forms of wireless communication, have made it easy to form a network of things.

We can now achieve distributed computing by sharing the processing tasks locally where data is collected (on the edge) and on the servers (cloud). With this, the dream of having connected homes, connected vehicles, smart cities, smart agriculture will definitely be achievable in the coming years.

When both come together…

The relationship between the two is just like a digital peripheral nervous system and brain. The field of IoT deals with creating, gathering, transmitting data from all possible things around us and AI is responsible for embedding intelligence into IoT components. It helps us create value and gain meaningful insights out of the raw data which is just a mere junk of numbers for a computer.

A few amazing Applications…

Autonomous vehicles — Tesla CEO Elon Musk’s view regarding what makes Tesla’s self-driving cars unique: “The whole Tesla fleet operates as a network. When one car learns something, they all learn it. That is beyond what other car companies are doing…”

“Prevention is better than cure”

Until now, we just considered it to be applicable to humans. But people are now thinking from a broader perspective by including machines as well. With this, comes an interesting application called as “Predictive Maintenance”.

It uses data from various sources like maintenance records, sensor data from machines, and weather data to determine when a machine will need to be serviced before any damage takes place. By leveraging the power of AI, operators can make more informed decisions about when a machine will need repair. This helps the system become proactive not reactive.

Artificial Intelligence of Things is bound to impact almost every industry including automotive, aviation, finance, healthcare, manufacturing and supply chain. Of course, there are a few concerns regarding security and scalability. Nevertheless, many other emerging technologies like Blockchain and Cloud Computing are merging with AIoT to come up with innovative solutions which will change our lifestyle. Certainly, the impact is going to be profound. Concluding with an apt phrase about Industry 4.0-

AI is the new electricity, data is the new coal, and IoT the new coal-mine.