Google launches LaMDA, an artificial intelligence for communication

Like Google Home in 2016 or Duplex in 2018, Google presented LaMDA at its annual developers' I/O conference.
It is an artificial intelligence designed to animate discussions on any topic.

This tool will be integrated into Google Search 

During the meeting, Google's CEO presented it on the LaMDA stage.

For each query, the tool will respond with three to four sentences, resembling a natural conversation between two people.

Over time, LaMDA may be integrated into Google products, including Assistant, Workspace, and most importantly, search.
According to Google, LaMDA's natural dialogue capabilities have the potential to fundamentally facilitate
accessing and using information and calculations.

The LaMDA demo offers a glimpse into Google's vision for search. This vision is centered on AI, which can infer meaning from human language, engage in conversations, and answer numerous questions like an expert.

MUM: another artificial intelligence tool

Google has launched another AI tool called Multitasking Unified Model (MUM), which translates in French: Modèle unifié multitâche, that can support text and image search.

One day, a user may take a photo of a pair of shoes and then ask the search engine if they are suitable for climbing Mount Fuji. MUM generates results in 75 languages, and Google claims this gives it a more comprehensive understanding of the world.

MUM is designed to reduce the number of searches needed to find relevant answers. Furthermore, the tool can summarize and generate text.

However, relying more on artificial intelligence to break down text is also associated with risks, as computers still struggle to understand the full complexity of language.

One of the most advanced AIs for text generation GPT-3 

The most advanced artificial intelligence used to generate text or answer questions, known as large-scale language models, produces new, unique, and relevant articles.

In an article and presentation published online last year, researchers from Massachusetts Institute of Technology and Facebook  discovered that leading language models exhibited stereotypical biases based on race, gender, religion, and profession.

A broader language model

The launch of BERT in 2018 triggered a competition among tech giants to build broader language models and improve performance rankings for tasks such as GLUE and other popular language understandings or answering questions.

The engineers often evaluate these models based on a set of parameters, which is a measure of the relationship between artificial neurons in a deep learning system.