Um, none of those numbers add up.
Google acting dumb matters because its AI is headed to your searches sooner or later. The company has already been testing this new Google — dubbed Search Generative Experience, or SGE — with volunteers for nearly 11 months, and recently started showing AI answers in the main Google results even for people who have not opted in to the test.
The new Google can do some useful things. But as you’ll see, it sometimes also makes up facts, misinterprets questions, delivers out-of-date information and just generally blathers on. Even worse, researchers are finding the AI often elevates lower-quality sites as reliable sources of information.
Normally, I wouldn’t review a product that isn’t finished. But this test of Google’s future has been going on for nearly a year, and the choices being made now will influence how billions of people get information. At stake is also a core idea behind the current AI frenzy: that the tech can replace the need to research things ourselves by just giving us answers. If a company with the money and computing power of Google can’t make it work, who can?
SGE merges the search engine you know with the capabilities of a chatbot. On top of traditional results, SGE writes out direct answers to queries, interspersed with links to dig deeper.
SGE is a response to the reality that some people, including me, are starting to turn to AI like ChatGPT for more complex questions or when we don’t feel like reading a bunch of different sites. Onely, a search optimization firm, estimates that using SGE can make a user’s overall research journey 10 to 20 times shorter by assembling pros and cons, prices and other information into one place.
An all-knowing answer bot sounds useful given our shrinking attention spans. But Google has a lot to work out. We expect searches to be fast, yet Google’s AI answers take a painful second or two to generate. Google has to balance the already-fragile economy of the web, where its AI answers can steal traffic from publishers who do the expensive and hard work of actually researching things.
And most of all, the new Google has to deliver on the promise that it can consistently and correctly answer our questions. That’s where I focused my testing — and kept finding examples where the AI-supercharged Google did worse than its predecessor.
Putting Google’s AI answers to the test
Often when you’re Googling, what you really want is a short bit of information or a link. On a day-to-day basis, the new Google is often annoying because its AI is so darned chatty.
A goofy example: “What do Transformers eat?”
The AI answer told me that fictional robots don’t really need to eat or drink, though they need some kind of fuel. Meanwhile, old Google had the one-word answer I was looking for: Energon. (It’s a kind of magical fuel.) You got that answer from new Google only by scrolling down the page.
This doesn’t just happen with alien robots. When SE Ranking, a firm dedicated to search engine optimization, tested SGE with 100,000 keyword queries, it found the average answer it generated was 3,485 characters — or roughly a third as long as this column. One of Google’s challenges is figuring out when its AI is better off just keeping quiet; sometimes, SGE asks you to press a “generate” button before it will write out an answer.
Most of all, when we search, we expect correct information. Google claims…
This article was originally published by a www.washingtonpost.com . Read the Original article here. .
