Google makes adjustments to AI previews after troubled rollout

Google is making some changes to its AI previews, after the artificial intelligence-powered search feature yielded what the company calls “strange and wrong” answers to people’s online searches.

The AI ​​previews were presented last month at Google’s annual I/O developer conference. Now, when people use Google Search to find information on certain topics, an AI-generated text box appears at the top of search results, annotated with links to external websites. Traditional search results appear below AI previews, marking a major shift in how Google presents information.

According to a blog post by Google VP Liz Reid, AI preview results are generated using the company’s Large Language Model (LLM), Gemini, and are designed for cases where someone one wants to “get both a quick overview of a topic and links to learn.” more.”

PHOTO: Liz Reid, vice president of search at Google, speaks at an event in New Delhi on December 19, 2022.

Liz Reid, vice president of search at Google, speaks at an event in New Delhi on December 19, 2022.

Sajjad Hussain/AFP via Getty Images, FILE

Google technology expert Alex Joseph told ABC Audio that AI Overviews is capable of answering more complex questions than a traditional Google search.

“With AI insight, what (Google) can really do is synthesize a lot of information and give you the answer that you’re looking for very quickly,” Joseph said.

Instead of presenting users with pages of links to navigate, Joseph said, AI Overviews streamlines the process by summarizing information and providing users with a concise answer.

“They’ll have less friction, they won’t need to click through a number of different websites, which can often be a real headache if you just want to get information really quickly,” notes Chris Stokel-Walker , technology journalist and author of the book “How AI Ate the World: A Brief History of Artificial Intelligence – and its Long Future.”

However, Stokel-Walker said the new feature makes it more difficult for Google Search users to verify the accuracy of the information they read.

“Over the last couple of decades, we’ve kind of gotten used to the dominance of Google search, the results that we get with a search term being largely correct,” he told ABC Audio. “Suddenly, if you get rid of that, as Google proposes, and just insert an answer directly into the search results page created via generative AI, you have no real way to identify and analyze this information to see if it’s true or not.”

There are also other concerns regarding the new feature. On the one hand, generative AI technology, both from Google and elsewhere, has been criticized for “hallucinating,” that is, generating unreliable and inaccurate information.

For example, in the first few weeks since AI previews were made publicly available, Google Search users were advised to eat at least one small stone per day. involves mixing glue with tomato sauce – which is of course a very bad idea. It also states that Andrew Jackson, the seventh American president who died in 1845, graduated from college in 2005.

Stokel-Walker said any benefit from AI insights ultimately comes down to a tradeoff between convenience and cost. “You no longer have to click through five or six different pages and maybe multiple pages of search results to find the right answer, but that also means that either the answer might be wrong or it might not may not be the one you really want to get,” he said.

PHOTO: Google CEO Sundar Pichai speaks at the tech titan's annual I/O Developer Conference on May 14, 2024, in Mountain View, California.

Google Chief Executive Officer Sundar Pichai speaks at the tech giant’s annual I/O Developer Conference May 14, 2024, in Mountain View, California.

Glenn Chapman/AFP via Getty Images, FILE

“We’ve always been very clear about the limitations of LLMs, that sometimes there will be hallucinations,” said Google’s Alex Joseph, adding that’s why AI Overviews also cites the websites it uses to generate its responses.

“That’s part of the reason we present all the information to you in a holistic way,” Joseph said. “These are quick shortcuts to help you get information quickly, but they’re followed by areas where you can go, double-check, check.”

Joseph also said that not all queries are better served by AI previews: “We only show them on queries where we are confident it will be helpful and actually improve the experience.”

Following unusual responses reported by some social media users, Google announced that it has made “more than a dozen technical improvements” to the AI ​​previews. According to Liz Reid’s blog post, these include limiting the inclusion of user-generated content, as well as satirical or humorous web pages, in the data used to create AI insights. Reid said Google “has also launched additional trigger enhancements to improve our quality protections” regarding health content, and that it “aims to not show AI previews for trending topics difficult, where freshness and factuality are important”.

The blog post also notes that “AI insights generally do not hallucinate or make things up in the same way that other LLM products might” and that incorrect answers are the result of a “misinterpreting queries, misinterpreting a nuance of language on the web, or not having a lot of interesting information available.”

Besides accuracy issues, Stokel-Walker said Google’s prioritization of AI insights over traditional search results could affect revenue and reshape the way business is done on the web.

“Websites produce content; they try to make it attractive to Google. Google will display them in its search results. And as a result, people click on their website, they then see ads as a result of that, and the publisher makes the money that allows them to put new content on websites,” Stokel-Walker explained.

However, by replacing the top of the Google search results page with AI-generated content, Stokel-Walker said websites could see fewer visitors – and therefore less advertising revenue.

It’s an ironic situation, according to Stokel-Walker. That’s because Gemini, the LLM that Google uses to create its AI previews, relies on the websites it now appears on.

“These websites still need to exist, and they need to have a way to generate revenue, because otherwise there is nothing to rely on for AI-generated search results,” Stokel-Walker emphasized.

In a statement to ABC News, Google said its testing showed that the opposite happens: that links included in AI previews generate more clicks than if the page had appeared as it does usually in search results. Google also said it would “continue to focus” on sending valuable traffic to publishers and creators.

PHOTO: The Google logo is seen during a trade show in Las Vegas, Nevada, January 10, 2024.

The Google logo is seen during a trade show in Las Vegas, Nevada, January 10, 2024.

Steve Marcus/Reuters

Regardless of how concerns about AI previews eventually fade, it’s just one of many features the company has planned for its tech product line.

“I think it’s concerning to do something like this as quickly as Google is doing it,” said technology journalist C. Scott Brown of the Android Authority website.

Google has announced plans to offer additional features similar to AI previews, which will aim to answer questions about specific web pages or YouTube videos. Brown says these features will come to market amid increasing competition.

“And the reason he’s doing it is because he feels it. He has to keep pace with companies – notably like OpenAI, for example – that are creating generative AI technologies that threaten the core of Google’s business, which is to provide information to people through Google search, and thus serve them ads that make Google billions and billions of dollars,” Brown said.

“With Google seeing these things as a threat, they can’t relax. They don’t know how to do this cleanly and properly,” Brown added. “You just have to do it.”

ABC News

Back to top button