Google has launched AI Overviews. And it is not going too well. The feature was launched in the US last week and it will spread globally by the end of 2024, but results so far suggest that it is a mess. Users on social media were quick to point out the funny answers until several people started noting that there were also several concerning ones.
AI Overviews is an experimental tool. Once you enter a query, the search engines provide an AI-generated result. However, machine learning algorithms are not conscious. They are trained to recognize words and phrases in certain contexts, so when they are asked something, they create an answer that is expected to fit the query. Even if the AI has to make stuff up. Like creating fake legal cases for a lawyer.
Google’s AI Overviews gets the answers from real sites, but AI is not smart enough to recognize serious answers from satirical or comedic ones. So “news” sites like The Onion or social media platforms like Reddit are providing comedic fodder to the clueless AI.
When asked “How many rocks should I eat each day?” Overviews confidently says that according to UC Berkeley geologists you should eat at least one small rock per day. That’s from The Onion. Asked about how make cheese stick better to pizza, Overviews suggests edible glue. Yes, glue is use in advertisements to make the cheese look more stringy, but the claim actually comes from an 11-year-old post from redditor fucksmith. Please do not trust fucksmith on this.
Speaking to BBC News, a Google spokesperson stated the incidents were isolated. “The examples we’ve seen are generally very uncommon queries, and aren’t representative of most people’s experiences,” the statement said.
But things are a lot less funny than these examples might seem. The system is fostering misinformation. It claims that Barack Obama was the first Muslim president of the United States. Too bad that it’s not true – Obama is a protestant Christian. A Google search lists the “health benefits” of smoking as well as how running with scissors can be good for you. And how can we not appreciate the suggestion of drinking one’s own urine?
Given the shambolic results, there has been renewed interest in the question of what the point of AI is. Google claims that it “can take the legwork out of searching” – but if you need to make sure that what you get is accurate, you risk pushing misinformation onto people who do not know or have no time to check sources. And on top of that, AI has a significant carbon footprint. Microsoft, also pushing big on AI, has recently been reported to be failing its pledges to be carbon-negative because of AI.
A suggestion on how to clean a washing machine recommends using chlorine bleach and vinegar. The expanded answer states that you need to use them one after the other, but it should be screaming about not using them together. Mixing them creates chlorine gas, which is poisonous.
And talking about poisonous, Google’s AI Gemini is not doing much better. It recently confused the destroying angel mushroom (Amanita bisporigera) for a white button mushroom. This is one of the most dangerous mushrooms in the world and it can destroy liver and kidney tissues.
“Doctor Google” has taken a new and dangerous form. But there’s comedy in the advice too, still thanks to satirical websites. AI Overviews will tell you that eating ass has health benefits, making you even less susceptible to the common cold and influenza.
If you want to argue with science, you better do your own research.
Source Link: Google's AI Overviews Is Returning Bizarre And Even Dangerous Results