With time Google has tried to evolve search queries. And the tech giant has had machine learning to the rescue. Google at its developer conference of 2021 announced a breakthrough in conversational technology with LaMDA: Language Model for Dialogue Applications. This makes conversation more engaging, natural and open ended on a wide variety of topics.
LaMDA as the acronym suggests has been trained in dialogue. With years in the making, the conversational skills are built on a neural network architecture, Transformer. With Transformer the model can be trained to read many words, understand their relation and predict the next words that should be said in a conversation.
Sundar Pichai demonstrated how LaMDA can carry on a conversation no matter what is spoken about. The demonstration included a conversation with a paper plane and Pluto. The questions were wide ranging and the answers equally open ended keeping the conversation going and natural and engaging.
Pichai also stressed that there need to be models that allow people to naturally ask questions across different types of information irrespective of the mode—text, images, audio, video. Called as multi modal models—it is so effective that if there is a query for a specific moment in a video, it can immediately skip to that particular frame.
Considering the harmful biases that still exist with AI and in a conversation or a language, Google in its blog mentioned that it is working toward minimizing risks associated with hateful speech and replicating misleading information.