Those who use a language assistant know how frustrating it can be to talk to Siri or Alexa. The ability of the voice assistants to talk is often still very limited. They can understand, but the conversation is still far from a complete conversation. With their new chatbot, according to Google, they have solved this problem.
Meena, as the new wonder weapon of the tech giant is called, is supposed to communicate with people via neural conversation networks with 2.6 billion parameters much better than any other artificial intelligence on the market.
The Google development team trained the new Cahtbot model with more than 40 billion words. The 341 GB of text data also included social media conversations to make the flow of conversation seem even more natural. The neural network used by Google is called Seq2seq and compares words in a paragraph to understand the relationship between them.
Meena sensitivity and specificity mean value (SSA) in comparison (Source: Google)
To make the conversation more meaningful, the Google research team has also developed a new matrix called the Sensitivity and Specificity Average (SSA) to make the conversation more enjoyable. So far, Meena has not yet been released to the open source community, but as soon as that changes, the potential of the new bot can be better tested.