The robot presented on the video can chat with other humanoid robots via the Internet. Moreover, when faced with a new situation, the robot can search the World Wide Web and gather appropriate information in order to fill an identified knowledge gap. Afterwards, it incorporates new knowledge under its own power, and is ready to execute diverse new tasks. Present chatbots have rather limited reasoning, inflexible behavior and they lack learning abilities. They perform entrusted tasks in strict accordance with programmed procedures. What if chatbot developers equipped new chatbots with an advanced online-learning mechanism applied recently in intelligent humanoid robots?
Researchers from Hasegawa Lab in Tokyo Institute of Technology developed the online-learning mechanism basing on the method of Self-organizing Incremental Neural Network (SOINN). They implemented this mechanism in intelligent humanoid robot which processes diverse tasks flexibly, is able to learn and make its own decisions while interacting with a real environment.
The robot’s general purpose tasks are executed by application of already known knowledge. Additional learning of insufficient knowledge is made possible through the SOINN unsupervised online-learning method. Thus, the overall intellectual level of the robot is improved. See the following video with Self-Organizing Incremental Neural Network at work 
