Google Unveils LaMDA and MUM to Advance Conversational AI
Google introduces LaMDA and MUM at I/O 2021 to make artificial intelligence conversations feel more natural and helpful. These new language models enable open-ended, flexible dialogue that mimics human chat.
Google showcases new conversational AI techniques at its virtual I/O 2021 event, focusing on the Language Model for Dialogue Applications (LaMDA) and the Multitask Unified Model (MUM). These tools aim to enhance how artificial intelligence understands and responds to casual human speech. While Google Assistant receives minimal direct attention, these underlying models represent a significant leap in making AI interactions feel more like talking to a real person.
LaMDA tackles the complexity of open-ended conversation by training specifically on dialogue instead of standard text. It analyzes individual words alongside full sentences and paragraphs to grasp the broader context of a chat. This allows the model to maintain a coherent discussion even when a user abruptly switches topics, such as jumping from a conversation about Pluto to the construction of paper airplanes.
Born from the earlier Meena chatbot project, LaMDA functions as an open-domain model capable of discussing virtually any subject without requiring retraining. However, experts note that this extreme flexibility raises questions about whether the AI remains consistently sensible across all possible topics. Alongside the multitask capabilities of MUM, Google signals a strong push toward more intuitive and fluid human-computer interactions.