Google To Scale Back AI Searches In Latest Controversy

Google has said it will scale back the use of generative artificial intelligence (AI)-created AI Overviews in its search results after they produced incorrect and bizarre results.

The move is an embarrassing step backward for the search and advertising giant, which has pushed to gain the lead in generative AI over fears that the popularity of ChatGPT, developed by Microsoft-backed OpenAI, could erode the use of its flagship search product.

The AI Overviews feature caused frustration for users after Google began rolling it out to all US search users last month.

The auto-generated search summaries, which can take up substantial space at the top of search results, often included incorrect information or content unwittingly derived from joke or satirical content, advising users to eat rocks or to add glue to pizza, users found.

False results

More worryingly, the overviews included false information, for instance repeating a right-wing conspiracy theory that former US president Barack Obama is Muslim.

Google said in response it would reduce the scope of searches that will include an AI-generated overview.

Head of search Liz Reid said the company had added restrictions to the types of searches that would produce an AI summary, had “limited the inclusion of satire and humor content” and had taken action against a small number of the results that violated its content policies.

Reid said many of the unexpected results were due to the lack of genuine information the AI could draw on, or came from users trying to game the system.

“There’s nothing quite like having millions of people using the feature with many novel searches,” Reid said in a blog post.

“We’ve also seen nonsensical new searches, seemingly aimed at producing erroneous results.”

‘Hallucinations’

She said that two weeks’ worth of public usage had allowed Google to determine patterns that did not work and to make “more than a dozen technical improvements to our systems”.

AI experts, and Google itself, have long warned that generative AI systems cannot be relied upon to produce reliably accurate information as they are essentially designed to predictively generate content and as such are prone to “hallucinations” in which they make up convincing-sounding falsities.

Web publishers are also concerned that the summaries could vastly reduce traffic to websites.

Gartner has forecast a 25 percent decline in search engine traffic volume by 2026 due to the use of generative AI.