Google has announced a new project which is in the early stages, to develop a unified language model of AI. Representatives of the company said they believe this project could impact the entire ecosystem of the company’s products.
The task of all this is to create a single AI language model that will support the world’s 1,000 most widely spoken languages. Today, the company has already taken a big step towards this and presented an AI model trained in more than 400 languages.
Along with this, the company announced research on text-to-video conversion models. That is the prototype of an AI writing assistant called Wordcraft.
In digital Google AI, Vice President of Research Zoubin Ghahramani talked about the project using a unified AI language model. He said the company believes that creating a model of this size will make incorporating various AI features into low-resource languages easier.
The vice president of the company explained how this works. He said that a model exposed and trained in many different languages would perform better in our low-resource languages. He also added that they don’t plan to bring 1,000 languages by building each one; he compared those languages are like organisms because they have evolved and have similarities.
Although language models have several shortcomings, the main ones are the tendency to reproduce harmful societal biases, among them racism and xenophobia. Just as incapable of parsing language with human sensitivity, Google began to incorporate these language models into products such as Google Search.