- Meena is a neutral network that has about 2.6 billion parameters.
- Meena is multi-turn open-domain chatbot trained end-to-end on data mined and filtered from public domain social media conversations.
- Google in its blog post also stated that the other AI-based assistants lack sense.
- Google has introduced Meena, which claims to be the greatest chatty of all existing AI chatbot available in the market.
From a long time ago, Google is working on its greatest digital assistant that is efficient in having a conversation based on Artificial Intelligence (AI) model known as Google Meena.
Google Meena, a neural system that has 2.6 billion parameters, is a multi-turn open-domain chatbot and claimed to be better than any other AI chatbot available in the market. The team educated the model with 40 billion words — 341 GB of text data including social media conversations — using the seq2seq model.
Seq2seq is an alternative to Google’s Transformer — a neural network that resembles words in a paragraph to each other to understand the relationship between them.
Conversational AI is a comprehensive term used to natural language models for AI that understand human language, speak and completes duties given in easy phrases. According to Google, Meena can carry on a conversation at a level that feels far more like talking to another person than any existing chatbot.
Nevertheless, speaking and conducting a Conversation are both different duties as AI assistants like Alexa, Siri, Google doesn’t really chat — they just complete tasks and answers authentic questions.
Occupying a human-like conversation is still very difficult for smart assistants available today.
For checking the efficiency and capacity of a conversational agent, Google released a system of measurement called Sensibleness and Specificity Average (SSA).
SSA helps to keep up answers in a conversation and makes sure that they are relevant and understandable.
SSA ranks the conversation by evaluating the static performance mixed with genuine interaction — The efficiency outcomes are primarily based on the percentage of turns counted as relevant or sensible.
The SSA kit represents a score of 79% to Google Meena, Pandora Bots AI Agents, Mitsuku got 56%, Microsoft’s Mandarin Chinese Chinese speaking XiaoIce got 31%, where humans got 86%.
All the conversations had to last between 14 to 28 turns, and bots that give general answers get the penalty.
AI assistants that are capable of holding human-like conversations might become an essential part of human life, as they could provide moral or emotional support to their owners said Ashwin Ram, the Alexa prize head and Google Research director.
In addition to Google, Microsoft additionally engaged on Bot Framework — last year they started showing extra multi-turn dialogue for users, and also acquired Semantic Machines in 2018.
Last year, Amazon also proposed a landmark called “Conversations” that offers voice application suggestions in conversational multi-turn dialogue. It was called “the holy grail of voice science” by the VP of devices David Limp.