After Google's latest search feature “AI Overviews” failed dramatically last week, the corporate downplayed the issue but said it was working on an answer.
AI Overviews uses artificial intelligence to leverage information from the web to supply a solution to a search query without the user having to click through multiple web sites.
In a classic case of “garbage in, garbage out,” lots of AI Overviews’ answers were laughably improper because they were based on false information published on the Internet.
Search leader Liz Reid issued an announcement in a blog post In it, she said that AI Overviews was an enormous success, but admitted that there have been some catches.
Reid said the explanations for a number of the strange results were nonsense queries or fake screenshots. Regarding the fake results, Reid said, “These AI summaries never appeared, so we encourage anyone who comes across these screenshots to do a search themselves to confirm them.”
She admitted that AI Overviews returned some “strange, inaccurate, or unhelpful” results. The fundamental reason for that is either the dearth of high-quality data on the web related to the query or that the query was nonsensical and the search tool failed to acknowledge this.
The suggestion to eat rocks is an ideal example of the challenge AI Overviews faces. The query “How many rocks should I eat?” is one not many individuals would have asked. When AI Overviews tried to reply the query, it got here across a satirical article on The Onion that jokingly really helpful eating a small rock a day.
AI Overviews isn't particularly good at distinguishing humorous, satirical, or sarcastic content from factual information, but Google says it has improved the standard of AI Overviews results as follows:
- Building higher detection mechanisms for nonsensical queries and satirical or humorous content.
- Limit using user-generated content (e.g. Reddit) in answers which will contain misleading advice.
- Added trigger restrictions for queries where AI overviews have proven to be less useful.
- Stronger protections have been added to limit responses on topics akin to news and health.
Some people might need had a little bit of fun specializing in the weird answers, however it's extremely difficult to consistently provide good answers to go looking queries. AI Overviews is objectively helpful, although you must approach the outcomes with some skepticism.
AI doesn’t create information, but relies on its training data or, as with AI overviews, returns a solution based on human-generated information from the web.
AI Overviews must answer the identical difficult questions that folks face after they read something online: Is this true?
Ironically, AI makes answering this query harder, not easier.
