Among Google’s announcements attoday, LaMDA is perhaps the most clever of them all. And it sounds like users will soon be able to benefit from its focus on more natural sounding conversations with AI.
During his I/O keynote, Google CEO Sundar Pichai said he hopes LaMDA has the ability to make information more accessible and easier to use. He continued that he wants to implement the system into Google Assistant, Search and Workspace.
- : Android 12, Pixel 5a, Wear OS and all the big news
, the tricky thing about conversation is how they can take unusual routes around and through topics. Therefore LaMDA was trained with dialog, instead of traditional text training, to help it develop a sense of context that is crucial to make conversation easier. The watchwords Google uses here are that the answers are ‘sensible and specific’.
Google demonstrated this by showing a video in which a person held conversations with the planet Pluto and a paper airplane — two surreal demos that are well worth watching. The AI was able to answer from the perspective of these objects, both imparting information and answering more emotional questions ‘in character.’
Google’s blog post also focuses on the responsibility it has on training LaMDA the right way. It details how the system is built to focus on factuality, and to fit in with the company’s AI Principles to avoid misuse, such as opening the training data to researchers to ensure nothing unpleasant like hate speech or bias is being unintentionally built into LaMDA.
As Pichai himself said, the LaMDA system isn’t perfect yet, and it’s yet to be offered out to other developers to work on. However this has exciting implications for Google’s already excellent web search and digital assistant features.