Edited at

My thoughts on why AI x IoT is a killer combination.


Iteration 1

It seems trivial that AI x IoT are always together for the conferences like AI/IoT or IoT M2M treat them as friends. But with a little bit of closer look at these things, we can understand they're actually competing technologies each with their own drawbacks.


How are they even competing technologies?

Let's start with Authentication by Apple. Apple used IoT (fingerprint sensor) until iPhone X for authentication. But from iPhone X, it is using AI (face recognition) for authentication. See how AI took over IoT's project.

Not yet convincing?

Now let's take the case of indoor navigation. They're infinite number of use cases of this application. 3 years ago, everyone's thoughts were to solve this problem using extra sensors to track people. But today's solution (used in retails shops) uses computer vision to track people and what they have picked (Video Link).

Or consider self driving cars. While google already solved it with extra sensors and camera, Elon Musk through Tesla said, they're trying to solve the same problem just with computer vision 3 years ago.

Or consider battery life on phone, there are two approaches, IoT way i.e., adding a Thing (battery) or using AI i.e., AI to save power. Google through Pixel 3 is using AI to save battery consumption.


Competing Right?

Let's have another iteration of thought.


Iteration 2

Now let's understand one thing,


Why did it almost every time start with IoT and later got taken over by AI?

It got taken by AI because IoT uses extra sensors which are in general expensive. But AI (Computer Vision) just uses camera and a GPU to understand. But AI models taken time to be developed for these models needs to be trained on large amounts of data.


Can AI & IoT be together for a more automated world?


Yes, IoT can solve the problems faced by AI and take have it's own position.

Yeah, IoT knows the solution but is expensive. So, if IoT can teach AI, AI becomes better. For example take this,

Every day at Uhuru Kamiyacho Office, every one swipes RFID card while going in and going out. Now, if we take a photo and add the tag from the RFID, and train an AI model, and use this trained AI model for door opening or authentication, we get costs saved for RFID card reader and so on and get the problem solved.

Or consider, Yamamoto san, Katsu san, Nishimura san, Akama san's fun project of finding if chair is occupied or not using enebular (IoT)? Can AI solve it? Yes, but requires data. Use the same techniques. Tag pictures using IoT project's sensors and train AI using it, to use the model elsewhere. What will we achieve? Save IoT infrastructure costs and AI data collection costs.


IoT can be a teacher for AI. But IoT like all teachers shouldn't be jealous of it's student AI. Together for a better future.