Neural networks and machine learning (or "automatic learning") are topics that appear frequently in the tech press, but it's not always possible to interact with them or understand them more deeply. That's where Google's Teachable Machine experiment comes in, allowing us to easily train a machine to perform different actions based on the gestures we make in front of a camera.
Understanding the Experiment
Communicating its achievements is one of the biggest challenges every company faces. Some concepts are too advanced, and if they aren't shared in a friendly way, they eventually get lost in digital noise. One of them is machine learning, also known as automatic learning. What is it? How does it work? What are its limitations? While we know artificial intelligences need to be trained, it would be ideal to participate directly in that training. Thanks to Google's new experiment called Teachable Machine, we can now do that.
How It Works
A browser, a webcam, and a lot of movement. Teachable Machine divides learning into three different colors (or "classes"), and each one is associated with a gesture. The machine requires a good number of images to identify them (30 at minimum), and once it reaches an adequate confidence level, it can play a GIF, a sound file, or phrases with the speech synthesizer. If you don't want to make gestures, an interesting alternative is to use objects. The key is to give the machine something to work with. The rest is a matter of combining all its training, or driving it crazy with half gestures and objects similar to the originals.
Try It on Mobile and Beyond
If you don't have a webcam connected to your computer, don't worry: Teachable Machine works with most mobile browsers, using the phone's front camera. Finally, Google assures that the experiment runs entirely locally without sending personal data to its remote servers, and if you want to see the experiment's code, just click here.
