Google presented innovations in artificial intelligence systems that nourish its search engine. These improvements aim to optimize the content that users access when searching for information about suicide, sexual assault, substance abuse and domestic violence.
More precision when looking for information in critical situations
Contact information for relevant national hotlines will now be seen more prominently along with the most important and high-quality results available.
To achieve greater accuracy in the results, the machine learning system was improved to understand the language of search, as the company explained at a press conference attended by Infobae.
“Now, using our latest AI model, MUM, we can automatically and more accurately detect a wider range of personal crisis searches. MUM is able to better understand the intention behind people's questions to detect when a person is in need. This helps us to more reliably display reliable and actionable information at the right time. We will start using MUM to make these improvements in the coming weeks,” they stressed from the company.
Improvements in safe search: what is it about
For some time now, the search engine has had the Safe Search tool, which offers users the ability to filter explicit results. This is the default setting for Google accounts for children under 18 years of age. You can choose to disable this option, but artificial intelligence systems still reduce the appearance of unexpected content in searches.
To further limit that type of unwanted content, the company announced new updates behind BERT (the acronym in English for Encoder Representations) bidirectional transformers), a technique used by Google for pre-training in natural language processing.
The great contribution of this technique is that it allows a bidirectional interpretation, that is, to interpret a term in context, both the word that precedes it and the one that follows it are taken into account.
Now, BERT has improved comprehension and can better understand search intent, which further reduces the chances that the user will encounter unexpected search results.
“This is a complex challenge that we have been tackling for years, but in the last year alone, this improvement in BERT has reduced the presence of unexpected results by 30%. It has had a special impact on reducing explicit content for searches related to ethnicity, sexual orientation and gender, which may disproportionately affect women and especially women of color,” they emphasize from the company.
MUM can transfer knowledge across all 75 languages it is trained in, allowing security protections to be scaled around the world more efficiently. AI is used to help reduce useless and sometimes dangerous spam pages that could appear in search results.
“In the coming months, we will be using MUM to improve the quality of our spam protections and expand into languages where we have very little training data. We will also be able to better detect personal crisis queries around the world, working with trusted local partners to show practical information in several other countries,” they announced.
Meta
Meta announced the development of an AI system that can research and write the first drafts of biographical publications in the style of Wikipedia. The objective of this model is to solve the lack of representation that exists on this site and others like it. Only 20% of the biographies on Wikipedia are of women, they reported from the company when making this announcement.
The developer of this project is Angela Fan, researcher at Meta AI. “There is still work to be done, but we hope that this new system will one day help Wikipedia editors create thousands of accurate and captivating biographical entries about important people who are not currently on the site,” the scientist stressed.
Women are underrepresented on that platform, despite the impact they have had on science and other fields. To illustrate this idea, Fan shares the case of Canadian physics, Donna Strickland. She won the Nobel Prize in Physics in 2018, however, as soon as she won the prize no one would have been able to find information about her on Wikipedia, because she simply hadn't. A publication was only made on that site a few days after that award, the most important in its field of study.
“Our work is purely research at this point, and we expect the Artificial Intelligence research community to leverage our model and dataset as a starting point for developing and moving forward. The idea is to one day be able to use AI to compensate for gender imbalances in the biographical content of Wikipedia, one of the most notable information references on the web. Women have been and are fundamental in many aspects of society, but their contributions are not given as much visibility as if they are seen in the contributions made by men. Representation matters, and we want to contribute to it with this research,” said Fan when asked by Infobae about the scope of this development.
How does the model work
The model first developed retrieves relevant information from the Internet to introduce the topic. Next, the generation module creates the text, while in the third step, the citation module builds the bibliography with links to the sources used. The process is then repeated, with each section to cover all the elements that are present in a complete Wikipedia biography.
KEEP READING: