In recent years, chatbots have become increasingly common, finding application in various sectors and it seems that the Google team is working on a solution capable of guaranteeing a model that is able to offer a human-type chatbot that is able to speak of anything.
The Google Brain team of researchers in the past few hours has revealed the details related to Meena, a neural conversation model trained to try to correct the “critical flaw” of current chatbots, which often say conflicting and non-context-specific things or demonstrate that not having common sense and basic knowledge.
Google’s Meena model focuses on understanding the context of a conversation to provide a reasonable response, all with the aim of creating something that can “chat on practically everything a user wants.”
The Meena project focuses on intelligence
Meena was created on the basis of 341 GB of text of conversations on social media in the public domain, or 8.5 times more data than the existing cutting edge generative models.
Google has created a new quality benchmark for chatbots and human evaluators are asked to judge whether a response is “reasonable in context”. According to this benchmark, Google’s solution is far better than those of the competition:
For more information, please refer to the page dedicated to the project.