|
Listen to this story
|
Talking to a chatbot can get tricky and sometimes lead to a dead-end.
To ensure a smooth conversing experience, Google has announced a language model called ‘LaMDA’ or ‘Language Model for Dialogue Applications,’ at Google I/O 2021. The new model makes conversations more meaningful and alive.
The tech giant is looking to tap into the versatility of human interactions for open-ended yet purposeful dialogues. Remember in the movie, Her, how Theodore Twombly (Joaquin Phoenix) falls in love with Samantha, a perceptive voice assistant.
Alphabet’s CEO Sunder Pichai said he spent some time with his son conversing with LaMDA (masqueraded as Pluto, the planet). It was magical as the model talked about the New Horizons spacecraft and the coldness of space, he said.
Further, Pichai said LaMDA synthesized these concepts from its training data. He said none of the concepts were programmed or pre-defined. “LaMDA answered with sensible responses, keeping the dialogue open-ended,” he added, stating that no conversation takes the same path twice and the model can carry a conversation without retraining.
How LaMDA works
In its blog post, Google said LaMDA’s conversational skills had been years in the making. The new language model has been built on Transformer — a neural network architecture open-sourced by Google Research in 2017— similar to many recent language models like BERT and GPT-3.
In January 2020, Google first published a research paper about LaMDA, a transformer-based language model trained on dialogue that could learn to talk on any topic. Since theN, Google has been fine-tuning the model to improve the sensibleness and the specificity of its responses significantly.
At Google I/O, the tech giant said its new language model still has some teething troubles. “Sometimes it can give nonsensical responses, imagining Pluto doing flips, or playing fetch with a favourite ball ‘the moon,” said Pichai.
Further, he said Google has put a lot of effort to ensure LaMDA upholds high standards in terms of fairness, accuracy and privacy.
“From concept all the way to design, we are making sure it is developed consistent with our AI principles. We believe LaMDA’s natural conversation capabilities have the potential to make information and computing radically more accessible and easier to use,” Pichai added.
He emphasised the need to integrate better conversational features into products like Google Assistant and Google Workspace.
The road ahead
Google said it is looking to explore ways to cater to the developers and enterprise customers. LaMDA is a huge step forward in the area of natural conversation.
However, LaMDA is still trained only on text. It is yet to explore various formats, including images, audio and video. The question is, how is Google planning to integrate all this? Pichai said, using a multimodal model that amalgamates text, image, audio and video, Google can provide a single source of conversational AI across applications and tools like Google Maps, Youtube, etc.
“All of this work has laid the foundation for complex models like LaMDA and multimodal. Our computational infrastructure is how we drive and sustain these advances, and tensor processing units (TPUs) are a big part of that,” said Pichai.
