Google has launched a new browser experiment, Teachable Machine, that allows users to get some basic, hands-on experience with machine learning. The program uses a computer’s webcam to let a user train a basic AI software. There are three input channels, namely Green, Purple and Orange, that could be used teach the program three different criteria. After the machine has learned enough, it would respond to inputs through GIFs, speech or sounds. For those who are skeptical about using their computer’s video camera for online programs, Teachable Machine lets users train a neural network located on their devices and hence, does not send any images to a server.
Teachable Machine is fun to use and an important step towards understanding the mechanism of AI. It is not about how perfect it is. It is about us getting to understand the amount of information it needs to be fed to deliver the most accurate output possible. It is about realising that such software does not perceive our gestures in the same way we do. Its perception of inputs is artificial and hence, easily modifiable.
Google says “keep playing around.”. Although we have made significant advances in this field in recent years, we still have a long way to go until we see more sophistication in software like these. It is recommended to use different approaches to help Teachable Machine learn different inputs and gestures in depth.