Google, with its new chat companion called Meena, wants to change that. As per a report by Venture Beat, Meena is a neural network that has around 2.6 billion parameters. We perform Meena, a multi-turn open-domain chatbot guided end-to-end on data drilled and separated from public domain social media conversations. However, Google hasn’t made it clear when correctly and how can users interact with Meena.
The report further says that Meena is trained on 40 billion words. As per Venture Beat Meena, it uses a “seq2seq model and a variation of the popular Transformer architecture.” Google also talked about the Sensibleness and Specificity Average (SSA), which is a metric created by Google researchers. The SSA helps measure the ability of a conversation agent to maintain responses in any AI-based conversation that is specific and also makes sense. According to Google, humans generally rank around 86% on SSA tests. Meena, in the initial tests he obtained a high of 79%. Of course, this is, Google’s own metric to judge conversational ability based on artificial intelligence.
Interestingly, Microsoft has a social chatbot also, and it’s called XiaoIce, which can converse in Mandarin as too.