MIT DEVELOPED AN AI SYSTEM THAT CAN LEARN LIKE CHILD
If we talk about teaching something to AI system, we need to use annotations that will describe how words work.
But Researchers in MIT have developed a parser that imitates kids learning processes by observing scenes and making connections.
System analyse caption videos and learns to link words to objects and action by determining the accuracy of description. It turns the potential meaning into logical mathematical expression, picking the expression that most closely presentment what it thinks is going on.
The method developed by MIT can speed up that learning process and will make AI that can handle uncommon languages that get AI-friendly annotations.