Chatbot Meena is going to be launched by top search engine Google soon. The arrival of the bot comes in the arena filled with Amazon’s Alexa and Apple’s Siri along with a plethora of applications. The company has claimed that its bot will outperform the existing bots in the field of Artificial Intelligence. The model boasts of a neural network of 2.6 billion parameters which is claimed to have enhanced its conversation skills. The announcement about the launch of the bot was made by Daniel Adiwardana and Thang Luong pf Google Research in a blog post.
Meena is an end-to-end which works on a neural conversational model that helps the bot learn to respond sensibly to a given conversational context and our training objective with regard to the bot is to minimize perplexity, the uncertainty of predicting the next token (in this case, the next word in a conversation), said the blog post.
At the centre of the bot, lies the Evolved Transformer seq2seq architecture, a Transformer architecture discovered by highly evolved neural architecture to improve perplexity, the blog post added.
Lack of conversational abilities has been the recurring issue faced in the technology with bots being unable to keep track of the context of the conversation. To beat this Google has claimed to have tested its model with 341 GB of data comprising of 40 billion words including social media conversations. Meena has a single evolved transformer encoder block and 13 evolved transformer decoded blocks. The encoder will aid the bot in understanding the context of the conversation while the decoded will help it in forming its response.
The bot will operate on the seq2seq model which in technological parlance is understood to aid AI devices in understanding words in conjunction with other words in the paragraph. Meena has not been made available for open source as of now. The company has not disclosed when it is going to make it available for open source.