Google has been generating answers with artificial intelligence (AI) for some searches in the United States for a few days. They are called “AI overviews” (summaries or overview, in Spanish) and users have begun to find delusional errors, which have quickly gone viral on networks like X: “Add non-toxic glue to the sauce,” Google responded to this query: how to prevent cheese from sticking to pizza. The intelligent seeker has also recommended that it is good to “eat a small stone” from time to time.
Google’s response also indicated that Barack Obama was the first Muslim president of the United States or that there are dogs that have ever played in the NBA. These hallucinations, as these excessive errors are called, have their explanation in all the readings that this tool has in its file. The glue on the pizza case, due to a comment on a Reddit forum; eating stones had their origin in The oniona page of satirical humor.
Google AI overview suggests adding glue to get cheese to stick to pizza, and it turns out the source is an 11 year old Reddit comment from user F*cksmith 😂 pic.twitter.com/uDPAbsAKeO
— Peter Yang (@petergyang) May 23, 2024
Google’s commitment to introducing AI in the search engine this time seems definitive. Other attempts have so far ended in failure, such as its image generation: “Users are responding very positively to ‘AI Summaries,’” said Sundar Pichai, CEO of Google, in a podcast this week. “It’s one of the most positive changes I’ve seen in search by metrics,” he added.
Given the errors, Google has shared some statements with different means: The type of consultations that have caused the hallucinations “are generally very rare and do not represent the experience of most people.” All the results of these summaries smart They carry a note at the end that warns of their “experimental” nature. The delusions of these AI systems are not limited to Google, but the reach of its search engine is much greater than that of its competitors, such as Perplexity, Bing or even ChatGPT. The company chaired by Pichai has its greatest hallmark and a good part of its business in the efficiency of its search engine, which could be called into question if it seemed unreliable.
EL PAÍS has replicated the most controversial searches and this Friday they did not provide any results generated by AI. It is possible that Google has blocked precisely those examples, although the same question usually generates different answers. In some cases, the first link that Google returned was already to comments about the errors in forums such as Reddit or Hacker News.
This newspaper has tried other searches and none have produced controversial results. The AI currently only works in English and the US, and does not always generate these AI-powered results for a specific search. Sometimes a simple word change in the query allows you to get a different answer from the search engine or the notice that it is not available for that search.
For Google’s AI, Pablo Picasso is the most famous Spaniard (Francisco Franco is on the list), paella is the most famous Spanish dish (searches on how to prepare it return links, not AI) and Julio Iglesias is the most famous Spanish singer. as the most important Latino in history.
The tool often answers medical questions, which are apparently more delicate: what to eat when you have diarrhea and diabetes, where type 1 diabetes comes from. Google’s AI summaries also venture into the origins of the Spanish Civil War — a blow against the republican government for its “inability to maintain order”—or whether Spain is a Catholic country.
You can follow The USA Print in Facebook and x or sign up here to receive our weekly newsletter.
