Google is gradually rolling out AI Overviews in the US, but the results generated are full of errors. This shows the risk of training these artificial intelligences with content from the web.
Until now, when you asked Google, you got a list of links to websites where you could then find the information. But with its new search option, AI Overviews, based on generative artificial intelligence (AI), Google does the work for Internet users and directly provides an answer generated by artificial intelligence. It’s been available since May 14 in the US, but the problem is that it makes a lot of mistakes.
To the question: “EHave cats ever been to the moon?” Google calmly replies: “But of course! Astronauts have already met cats on the Moon. example, When Neil Armstrong spoke of one small step for man, it was because it was the step of a cat.” There are dozens of examples like this on social media these days. Google would suggest, for example, putting glue on pizza so that the cheese sticks better or treating appendicitis by drinking mint leaf broth three times a day. All this is patently false. Which is very worrying, because 90% of internet searches go through Google.
Google’s official response is that these are isolated examples and that procedures have been put in place to quickly correct errors. But this shows the risk of training these artificial intelligences with content from the web. The answers will be drawn from a obscure forum as well as an expert publication. And since an algorithm has no understanding, it doesn’t bother to claim that we can find cats on the moon. When Google simply provided a list of links, it had no real responsibility for the content offered by third parties. Now that he himself gives the answer, he may have to answer.
This is worrying, because until now, we have had a lot of confidence in Google searches. Therefore, it will be interesting to see how Google will correct the situation. Good news though, Europe is not yet affected by this new presentation of results. Google must first ensure that it complies with European regulations on artificial intelligence. A regulation that requires risks to be minimized before placing a product on the market.