A group of members at the University of Wisconsin–Madison, the US, have developed an AI-based glass that can identify images and differentiate them with the help of light. The glass doesn’t need any kind of charging or power to operate.
In the proof-of-concept demonstration, the glass could differentiate between handwritten single digit numbers. In real-time, the glass can also identify those numbers which are presented with changes, for example, if 3 is altered to an 8.
There are small bubbles and some embedded impurities incorporated in the glass which power the AI to work. The light waves are bent in different techniques by the bubbles and impurities, once it is reflected off the handwritten numbers pass through the glass AI.
Depending on the numbers presented, on the other side of the glass, the bending procedure focuses the light onto one of ten specific spots. The process similar to a key and a lock, the ten spots then correspond to the digits 0 to 9.
According to one of the members, while training the AI, the team presented various pictures of handwritten numbers to the glass. The team had to adjust the size and impurities if the light did not penetrate through the glass to fall in the right area. After training the AI a thousand times, it grasped to bend the light to the right spot. The impurities act like artificial neurons.
The team showcased the proof of concept but it’s still unclear about implementing the glass AI in real life.