Google's AI 'hallucinates' about the Indian plane crash and changes the plane model

Google Overviews, the new AI feature built into the search engine , has made a notable mistake: following the plane crash of flight AI171 in India, the AI ​​confused the model of the plane involved, mixing references to a Boeing 787 with an Airbus A330 and displaying the information as if it were completely reliable.

This type of error is known as "hallucinations," a phenomenon unique to generative AI systems that involves inventing data or answers that appear to be true. Unlike a traditional search engine that links to sources, tools like Google Overviews attempt to "summarize" the web and sometimes incorrectly combine content, causing errors that can be critical in sensitive contexts like a tragedy.

What happened with Google Overviews

Last Thursday, a serious plane crash occurred in India, where an Air India Boeing 787 plane crashed shortly after takeoff. More than 260 people died , including several on the ground. It's a tragedy that has had a worldwide impact, and, as is often the case, many users began searching the internet for information to find out what happened and what type of plane was involved. But here, a problem has arisen with Google's artificial intelligence.

Its AI Overviews tool, which generates automated responses based on search results, started providing incorrect information. Some people noticed that when they searched for information about the crash, the AI ​​would confidently say the plane was an Airbus A330 , but other times it would say it was a Boeing 787 , and sometimes it would mix both. It was a real mess.

Why Google's AI got the plane model in the Indian crash wrong

According to ArsTechnica , this AI works by summarizing what it finds on the internet , but it doesn't always fully understand what it reads. Many articles about the crash mention Airbus as a competitor to Boeing. That may have confused it, leading to it saying something incorrectly, even though it seemed very confident.

Google has spoken out, clarifying that its AI can make mistakes , and that it adds a warning at the end of each generated summary, but the problem is that this warning is easy to ignore . In this case, the error may have upset Airbus , as its name was linked to an accident that had nothing to do with its planes.

The Mountain View company has confirmed that they have eliminated that incorrect answer , but this leaves us with a conclusion related to AI: although these types of tools can be useful, you cannot always completely trust what they say , so you have to check the information carefully.

Sign up for our newsletter and get the latest technology news straight to your inbox.