File photo of Google headquarters in London, United Kingdom on April 4, 2023. (Photo by Rasid Necati ... [+]
Google gave its new AI-powered search tool to select users late last week, and I’ve been testing out the technology over the past few days. The good news? The new Google Search gives more accurate answers than competing AI tools like ChatGPT. The bad news? The reason it’s more accurate is that Google’s AI appears to copy information from the internet in a form of semi-plagiarism. And to make matters worse, the design of the new Google Search will likely decimate the online publishing industry.
OpenAI’s ChatGPT launched to the public in late 2022 and sparked a gold rush in what’s called generative AI—chatbots that can answer questions in a conversational tone with an authoritative voice. But, as anyone who’s used ChatGPT can tell you, it’s not very accurate. In fact, ChatGPT will just make up sources of information that don’t exist, something a lawyer in New York recently learned the hard way when he filed a document in federal court with completely fictional cases listed as legal precedent.
Google launched a competing chatbot called Bard it had been developing for years in an effort to keep up with OpenAI, but Google recently announced it will soon incorporate generative AI in what’s arguably the most important product of the entire internet: Google Search, which has roughly 94% of the global search engine market.
After playing with this new version of Google Search over the weekend, I truly believe it’s going to revolutionize the way people get information on the web. But I think it’s going to leave a lot of online publishers struggling when this new technology is rolled out to the general public—something I first argued when Google announced its plans earlier this month at the tech company’s annual developer conference in Mountain View, California.
After getting access to Google’s new search feature I asked a simple question: Can you store live oysters in the fridge? Rather than being met with Google’s traditional page of blue links to various websites that might have the answer, Google created three paragraphs of text to answer my question, as you can see in the screenshot below.
The answer looked good from a design standpoint and was easy to digest. How accurate was Google’s response? Very accurate, as far as I could tell. But the reason it’s accurate will raise serious ethical questions for some people.
The top of a new AI-powered Google search that's currently being tested.
As you can see, the first line of Google’s answer reads: “Yes, you can store live oysters in the fridge. To ensure maximum quality, put them under a wet cloth.” Where did Google get this information? A blog called Get Shucked, which includes the line: “You can keep live oysters in the fridge. To ensure maximum quality, put them under a wet cloth.” It appears Google added the word “yes” and swapped out the word “store” for “keep.”
To be clear, it’s good that Google Search is being transparent about where it’s getting its information. Those three visible websites on the right can be clicked to learn more. In fact, there are actually five links if you click the little arrow. But the million dollar question will be whether anyone actually clicks on those links to help those websites generate their own revenue, especially since those sites are the ones creating value for Google.
The third line of Google’s answer about oysters explains, “You can also store them in a mesh bag or in an open container covered with a damp cloth.” That line appears to be borrowed from a website called Spruce Eats, which writes, “They should be packed in either a mesh bag or in an open container covered with a damp cloth.” Again, we see a couple of minor words swapped out, but otherwise the sentence is identical.
The last few lines of Google’s response were just lifted directly from a blog called Oysters XO. I made a quick graphic to show Google’s answers that are lifted word-for-word from websites. The parts that aren’t underlined were very similar to what appeared on other websites, but weren’t word-for-word. It appears Google’s AI mostly switched out synonyms like “store” instead of “keep,” as an example. But any high school kid turning in this response would likely get flunked for plagiarism.
An annotated answer from Google's new experimental search function.
There are positive and negative things about this new Google Search experience. If you followed Google’s advice, you’d probably be just fine storing your oysters in the fridge, which is to say you won’t get sick. But, again, the reason Google’s advice is accurate brings us immediately to the negative: It’s just copying from websites and giving people no incentive to actually visit those websites.
Why does any of this matter? Because Google Search is easily the biggest driver of traffic for the vast majority of online publishers, whether it’s major newspapers or small independent blogs. And this change to Google’s most important product has the potential to devastate their already dwindling coffers.
In the case of my question about oysters, I received all the information I needed from Google’s new generative AI response. I didn’t need to leave Google’s home page and I got all the vital information, which was pulled from blogs online. In the old days (which is to say, as Google currently exists for people who don’t have access to the new experiment), I probably would’ve clicked on one of the top blogs in a Google search to learn about how long oysters can survive in the fridge. But now I didn’t need to click on anything. And the question is whether anyone will bother, if Google just hoovers up all the information on the open web and spits it back in a semi-plagiarized form to all users.
Online publishers rely on people clicking on their stories. It’s how they generate revenue, whether that’s in the sale of subscriptions or the sale of those eyeballs to advertisers. But it’s not clear that this new form of Google Search will drive the same kind of traffic that it did over the past two decades.
Google’s new search tool is still an experiment, as the website warns you very prominently at the top of each search. And it looks like Google has put up some guardrails to ensure it isn’t generating responses for racist and sexist material. For instance, I asked whether racism was real, something the AI didn’t attempt to answer. When asked what the word “woke” means I was greeted with the phrase, “An AI-powered overview is not available for this search.”
But there are also areas where the new Google Search wouldn’t answer questions that were kind of surprising to me. I asked Google about the existence of aliens, and it wouldn’t generate a response. What is Google hiding about Area 51 and the saucer-people and reverse vampires running the government?
There are some Google searches that were already built to deliver immediate answers without users needing to click on anything. For example, when I searched the question “How old was David Bowie when he died?” the answer given by traditional Google was arguably more direct, as you can see from the screenshot below. The generative AI response about the late musician didn’t really highlight the answer in the same way that traditional Google Search did just below it.
Google's new generative AI response to a question about how old David Bowie was when he died.
Other rather innocuous questions were also difficult for Google’s new search feature to answer. While doing laundry this weekend I thought of an odd question: What did people in the Middle Ages think of static electricity? Did they think it was evidence of witchcraft or God’s rather minor punishment or some other force that explained this strange natural phenomena? The Google AI stumbled, saying it couldn’t generate an answer with AI. Then I tried to ask what colonial Americans thought of static electricity, maybe an easier answer, I thought. The AI came up with an answer at the bottom of the AI results dating to 600 BC, but otherwise the answers were all about static electricity more generally.
Asking Google's new AI about how people of the past thought of static electricity.
If you’re curious about the answer, as I was, I discovered from a 2017 textbook that magicians of the Middle Ages would sometimes use static electricity to perform tricks, which makes sense. But I still didn’t get an elaborate answer about what the average person thought about electrostatic shocks hundreds of years ago. If I want to learn more about how people of the past thought about static electricity it will require more thorough research beyond Google Search, as many questions often do.
After I conduct research in books and journal articles to figure out what people of the past thought about static electricity, I’ll likely write about it at my blog Paleofuture, citing my sources and answering a weird question other people might also be asking themselves while doing laundry. But then we return to the central question of Google’s new Search capability: Will Google just plagiarize the article I write for its own purposes? Which is to say, will Google’s AI tool just grab three paragraphs of text from my article and spit it out without users ever having to visit my website?
Paleofuture survives because I have paying subscribers. But it can only get paying subscribers when people find the site through social media or search engines like Google. Other websites also have advertising, another business model that depends on people actually clicking through on a link to read the article. And it looks like Google’s new tools for search would drastically reduce the number of views online that publishers would receive. I argued this a few weeks ago and after trying out Google’s new search functionality I think the argument holds up.
For its part, Google stressed to me that what I was experiencing was an experiment that could change before wide release.
“We’re putting websites front and center in SGE [the new search engine experiment], and the goal is to highlight websites and drive attention to content from across the web. Generative responses are corroborated by sources from the web, and when a portion of a snapshot briefly includes content from a specific source, we will prominently highlight that source in the snapshot,” a Google spokesperson told me via email.
“You can expand to see how the links apply to each part of the snapshot. SGE is an experimental experience in Search Labs, and it will evolve over time as we learn what is most helpful for people,” the spokesperson continued.
The company also downplayed the results I was seeing that lifted entire paragraphs word-for-word, characterizing them as “snippets.” But I think the text I underlined shows it’s a more liberal borrowing from the internet than mere snippets. Again, it all comes back to the question of whether people actually click on the links, a question I asked but didn’t get an answer to on Tuesday.
Will Google kill off many of the same publishers providing valuable information on the internet? Only time will tell. But I fully expect online publishers will see a huge dip in traffic when this new generative AI tool hits the masses. And oddly enough, Google needs those same publishers to make its new product a success. Google spent the past two decades absorbing all of the world’s information. Now it wants to be the one and only answer machine.
Updated at 7:20 p.m. ET with comment from Google.
