 "Google Search AI Overviews")
AI Overviews were announced at the Google I/O event a few days ago exclusively to users in the US with the promise of higher satisfaction with its search results by suggesting info and webpages closest to the search query. This new feature was a result of Google’s Search Generative Experience (SGE), launched last year.
In the last few days, people have shared screenshots of inaccurate and misleading results after using the new feature. Liz Reid, VP, Head of Google Search, has responded to the feedback in a blog post and detailed next steps that are being taken.
The post explains how AI Overviews work and informs that it works differently than a chatbot or Large Language Models.
“While AI Overviews are powered by a customized language model, the model is integrated with our core web ranking systems and designed to carry out traditional “search” tasks, like identifying relevant, high-quality results from our index. That’s why AI Overviews don’t just provide text output, but include relevant links so people can explore further.” the blog post informs.
The tech giant also added that AI Overviews does not hallucinate like chatbot and when it gets anything wrong, it is because of misinterpreting queries or nuance of language or simply not having better information, a problem Google said can arise for simple search features as well.
Also Read
What went wrong
The tech giant said that some of the screenshots circulating were fake and few queries were “nonsensical” and deliberately wanted to generate false results. Google has admitted that although some of the results were erroneous as they were for uncommon queries, Google will make improvements. Google has also accepted that some questions could not be answered due to being satirical in nature and informed that some of the questions were never asked before and hence resulted in what is called information gap or data void.
Improvements
Google has said that they do not fix every query but instead work on an update to broaden the scope of function to become inclusive of new queries. It also informed that they have made over a dozen technical upgrades.
Few improvements listed by Google are:
- Better detection mechanism to prevent any AI Overview result for a nonsensical query and regulated inclusion of humorous or satirical content
- Limited user generated content in responses
- Triggering restrictions for when AI Overviews starts answering
- Triggering refinements for topics of news and health
Additionally, Google said it is examining feedback and reports and taking action against AI Overview that violate content policies with the promise to keep improving AI Overviews.
