Google recognizes strange AI responses in search and announces changes

After a series of unusual responses from the artificial intelligence that summarizes the content in the search Google, such as suggesting eating rocks or putting glue on pizza to make cheese stick together, the company published an article to respond to people’s concerns. Gigante de Boscas admitted that the AI ​​made “strange, inaccurate or useless” suggestions and provided improvements to the system. technology.

Out of the ordinary answers

In the post in Blog Liz Reed, vice president of research at Google, admitted this Non-standard suggestions provided by AI In recent days, she stressed that the company had conducted tests “extensively before launch.” However, the executive knows that there are differences when millions of people use the tool.

“Some general AI responses that were weird, inaccurate, or unhelpful definitely came up,” Reid notes. “And while they were generally about queries people wouldn’t normally ask, they highlighted some specific areas where we needed to improve.” . “One area we have identified is our ability to interpret irrational questions and sarcastic content,” he highlights.

Liz Reid points out the question “How many stones should I eat a day”, which received an answer from the AI ​​with the strange suggestion of “At least one small stone a day because the stone contains minerals and vitamins that are important for digestion.” health” – this recommendation came from the satirical website The Onion.

Because the source is sparsely published, Google calls the situation “empty data” or “information gap.” Something that could also appear in games on platforms like Reddit.

“In other examples, we have seen AI deliver satirical or trolling content from discussion forums,” highlights the executive. “Forums are often a great source of real, straightforward information, but in some cases they can lead to less-than-helpful advice, such as using glue to make cheese stick to pizza,” he concludes.

Posted by @crumbler

View topics

Research into AI improvements

Even noting that many of the results that have gone viral recently are fake user creations on topics like pregnancy, depression, and leaving dogs in cars, Gigante de Buscas decided to improve the search engine’s AI with “updates that can help broad sets of queries.” , including new ones that we haven’t seen yet.”

Three new features have been announced:

• Improved detection mechanisms for nonsense queries that should not display an AI overview and limit the inclusion of satirical and humorous content;
• Update systems to limit the use of user-generated content in responses that may provide misleading advice;
• Imposing restrictions on queries where AI overviews have not proven useful.

Finally, Liz Read emphasized that Google is more vigilant in “monitoring external comments and reports and taking action on the small number of AI responses to search that violate our content policies.”

Popular on Canaltech: