The 2021 edition of Google I / O allowed the Californian company to show its advances in Artificial Intelligence, especially when starting a conversation. This is how LaMDA entered the scene, an innovative technology that seeks to give more nuances to the “talks” with elements based on artificial intelligence.
LaMDA is a “language model for dialogue applications” that has been trained to break up stuffy conversations and provide more realistic responses. Google wants this platform to adopt a more human language and be able to recognize nuances and adaptations.
As the Mountain View company explained, the conversational skills behind LaMDA took years of development. Artificial intelligence is based on Transformer, a neural network architecture invented by Google, and aims to revolutionize interaction with conversational assistants such as chatbots :
That architecture produces a model that can be trained to read many words (a sentence or paragraph, for example), and to pay attention to how those words are related to each other, being able to predict which words you think will come next.
But unlike most other language models, LaMDA trained in dialogue. During his training, he collected several of the nuances that distinguish open conversation from other forms of language. One of those nuances is good sense.
Google is confident in the future of LaMDA, but acknowledges that it is still an early development
The first demos Google did with LaMDA on I / O were really cool. The company featured a couple of clips in which it played back “conversations” it had with the AI. In the first, he responded in the first person as if it were Pluto , while in the second he “personified” a paper plane .
These early results are encouraging and we hope to share more soon, but good sense and specificity are not the only qualities we look for in models like LaMDA. We are also exploring dimensions such as “interest”, assessing whether the responses are insightful, unexpected, or ingenious. Being Google, we also care a lot about objectivity (that is, whether LaMDA sticks to the facts, something language models often struggle with), and we're investigating ways to make sure LaMDA's responses don't just are convincing if not correct.
Google The company acknowledged that the development of artificial intelligence is still in an early stage, but with great potential. That LaMDA interacts with users without repeating responses, or better understanding the context of a sentence, is key to the future of conversational assistants .
