An experiment designed to give everyone the chance to experience the enormous potential of machine learning: it is the Google Teachable Machine. When it comes to machine learning, thought goes straight to a complex technology, made of algorithms and codes, by its very nature difficult to understand to non-professionals. Google wants to make it easy to approach one of the most interesting fields of computer evolution, from the enormous potential, double-stranded to the development of solutions in the field of artificial intelligence.
For this reason, the Mountain View group has put online Teachable Machine, an open-source experiment, to be run directly into the browser interface even without having any programming or coding skills: the only requirement in hardware terms is a webcam connected to the computer to capture the video signal. Operation is well explained: clicking on the green, purple, and red buttons define three classes of actions that the system will respond by playing a certain GIF, an audio file, or activating the voice synthesizer to pronounce words or user-defined phrases. The more sample images of the AI are available for use – at least 30 are recommended – the more accurate and faster the response is. GIF, sound and speech are available.
What happens in the background is the use of a new library named deeplearn.js, made available to the entire developer community. The system analyzes in real time the images recorded by the webcam, comparing it with the ones previously acquired and searching for common items. It will then be understood when you are lifting a hand, thus activating the corresponding action.
This is obviously a simple example, reduced to the minimum terms when compared to what is the enormous potential of machine learning, but fully achieves the objective of facilitating the understanding of the dynamics that make up the foundation of technology. The code of the Teachable Machine experiment is open source and anyone can draw inspiration or modify it to give life to the most disparate projects. The web application is similar to experiments like ‘Quick, Draw!’ or ‘Autodraw’, quite easy to use.