Baidu is making deeper inroads into Artificial Intelligence. The Chinese tech company is doubling down upon AI and in a major breakthrough, has managed to teach a virtual agent living in a 2D environment. The teaching part is not particularly tough, however the trick and the breakthrough lies in teaching the AI to navigate using natural language commands.
Both negative and positive reinforcements were used and surprisingly, the AI managed to develop a basic sense of grammar. How was it similar to teaching a baby? Well, I was watching my niece today. She is one and a half and is just starting to learn things. So she doesn’t have a sense of numbers or what they are yet, but she has started trying to unlock iPhone and even recognizes where a particular number lies. How? By watching others touch those particular spots on the iPhone and speak out the corresponding numbers.
So that is how kids learn. They watch people around them and learn to associate names with pictures. That is positive reinforcement. Baidu’s system has even started taking the things it has learned and applying them to totally new situations.
This is what makes Baidu’s systems different. Predictability is an essential part of computer systems but Baidu’s systems are breaking the mold and are generalizing the skills they have picked up and are applying them to different situations.
Applying past knowledge to a new task is very easy for humans but still difficult for current end-to-end learning machines. Although machines may know what a “dragon fruit” looks like, they can’t perform the task “cut the dragon fruit with a knife” unless they have been explicitly trained with the dataset containing this command. By contrast, our agent demonstrated the ability to transfer what they knew about the visual appearance of a dragon fruit as well as the task of “cut X with a knife” successfully, without explicitly being trained to perform “cut the dragon fruit with a knife.