Once, Artificial Intelligence was exclusively synonymous with Skynet and fictional futures, but those days are long gone. Now, AI has quickly become woven into the fabric of modern technology. Over the last decade, AI has been more readily accessible and integrated than ever before. People interact with AI nearly every day; it’s already ingrained into society. But, it’s capable of even more. This article outlines the many other things the infamous Raspberry Pi pocket-sized computer can do. In addition to its ability to run an arcade emulator and create a media server, Raspberry Pi can be used to develop AI technology itself.
AI Processing Power & Pi Limitations
Artificial intelligence development requires a LOT of processing power. Take AI networks and AI model development, for example; it can take a server farm days to process a single computational task in these systems. However, once these models and networks are trained, they can be deployed on significantly less powerful devices. This is an important step in reaping the benefits of AI without the required exorbitant computation ability. While the infamous Raspberry Pi has far less computational power than a modern-day server, it can be utilized in a variety of ways in order to run AI models over a short amount of time.
Google AI for Raspberry Pi
Google has been a famous champion of AI in many sectors, and it’s also been a key player in bringing AI into the DIY and Raspberry Pi communities. The Google subgroup, AIY Projects, releases most of the AI and DIY content and kits, and they have several products that make deploying Google brains onto a Pi device extremely easy.
Google AIY Voice Kit Projects
The DIY intelligent speaker, Google AIY Voice Kit, was the first product to be released by AIY Projects. Similar to an Amazon Alexa, Google Home or Apple Homepod, this intelligent speaker is packed with voice detection technology that is capable of understanding the human voice, translating it into text, and then initiating action or solving a query. The AIY Voice Kit can be assembled in minutes (check out this AIY Assembly Video), and can be immediately used to create any DIY project where voice detection can be valuable. This kit comes with everything you need to get started, including a dedicated Voice Kit Pi and other hardware. The kit uses the Google Assistant API and Google Cloud -- the same tech used in Google Home — to allow you to explore the tech behind all things Google Voice.
Google AIY Vision Kit Projects
The Google AIY Vision Kit is the younger, stronger sibling to the Voice Kit. The general configuration of this kit is the same. It features all the required hardware needed to start, and uses various Google Cloud and API tech to power a smart camera out of the box. The Vision Kit comes equipped to identify up to 1000 different items through their object classification demo. The Vision Kit utilizes a live camera feed to identify (with confidence intervals) various objects. Face detection can be run using this kit, and you can even run a demo that takes a picture every time a face is detected. DIY home security system, anyone? The Vision Kit takes things even further by allowing you to employ Google Cloud to run TensorFlow and train your AI vision model with new image detection.
Let’s pause and appreciate how amazing this is. The AIY Projects Vision kit is not only an edge-deployed AI model running on a Raspberry Pi, it is also an edge-deployed portal to Google Cloud-powered image detection model training which, you guessed it, runs on a Raspberry Pi.
We Need More Power! Raspberry Pi AI Accelerators
If you are truly an overachiever, and you want to deploy more complex models that even the Raspberry Pi cannot handle — or if you want to try local model training – then, know that the Raspberry Pi is compatible with accelerators. These include models such as Google’s Coral Edge TPU Accelerator or Intel’s Neural Compute Stick 2. This means that you can easily equip your Raspberry Pi with additional, more powerful hardware, which can accelerate its ability to run complex AI models and neural networks. For example, you can use an accelerator to run scaled-down versions of AlphaGo's famous Go-playing neural net, or run complex image classification models. Therefore, you can achieve an even higher level of confidence than when running the model exclusively on the Raspberry Pi.
Five years ago, Deepmind’s AlphaGo technology bested the world's greatest Go player, Lee Sedol, in a triumph of artificial technology development and use. At the time, that triumph was the very pinnacle of AI development. In a profound nod to Moore’s Law, it seems only fitting that a scaled-down version of that same technology can fit into a pocket-sized computer system — one which costs less than $200, with Raspberry Pi at the heart of it.
