It was a busy Memorial Day weekend for Google (GOOG, GOOGL) as the company scrambled to contain the fallout from a slew of outlandish suggestions from its search platform's new AI Overview feature. If you were soaking up the sun on the beach and sipping hot dogs and beers instead of scrolling through Instagram (META) or X, here's what's happening.
AI Overview is supposed to provide generative, AI-based responses to search queries. Usually, it does. But over the past week, the app has told users that you can use non-toxic glue to stop cheese from sliding off your pizza, that it's okay to eat one rock a day, and that Barack Obama is the first Muslim president.
Google removed the response and said it was using the error to improve its systems, but the incident, coupled with the release of Google's disastrous Gemini image-generating app, which allowed historically inaccurate images to be generated, could do serious damage to the search giant's credibility.
“Google is supposed to be the ultimate source of information on the internet,” explains Chinmay Hegde, an associate professor of computer science and engineering at New York University's Tandon School of Engineering, “and if that product is watered down, it will chip away at our trust in Google.”
Google's AI Failure
Google's AI overview issue isn't the first time the company has run into trouble since it began its generative AI efforts. The company's chatbot, “Bard,” which Google rebranded to “Gemini” in February, gained notoriety when a promotional video for it in February 2023 showed an error in one of its responses, causing Google's shares to fall.
Additionally, the Gemini image generation software produced photos of various groups of people in inaccurate settings, such as German soldiers in 1943.
AI has a history of bias, and Google tried to overcome it by incorporating a wider range of ethnicities when generating images of people. But the company overcorrected, causing the software to reject some requests for images of people from certain backgrounds. Google responded by taking the software offline temporarily and apologizing for the incident.
Meanwhile, issues with AI Overview arose because users were asking unusual questions that Google flagged up: In the case of the rock-eating example, a Google spokesperson said, “It appears that a geology website was syndicating articles on the topic from other sources to their site, including one that happened to originally appear in The Onion. AI Overview linked to that source.”
These are good explanations, but I am tired of the fact that Google keeps releasing flawed products and then has to explain them.
“At some point, you have to take responsibility for the products you put out there,” says Derek Leben, an associate professor of business ethics at Carnegie Mellon University's Tepper School of Business.
“In terms of trust in the product itself, you can't simply say, 'We're going to be building AI into all of our established products, but it will always be in beta and we're not responsible or blameless for any mistakes or issues that arise with the AI.'”
Google is the go-to website for finding information online. When you get into an argument with a friend over something stupid, one of you will inevitably yell, “Sure, Google it!” You've probably done the same thing. Maybe it's not because you want to prove that you know more about obscure Simpsons facts than your friend, but still. In short, Google has built a reputation for reliability, but its AI failings are slowly eroding it.
Competing to win the competition
So why did this failure happen? Hegde says the company is simply moving too fast, releasing products before they're ready in an attempt to outwit competitors like Microsoft (MSFT) and OpenAI.
“The pace of research is so fast that the gap between research and product seems to be shrinking significantly, which is causing cosmetic problems,” he explained.
Google has been racing to shake off the impression that it is lagging behind Microsoft and OpenAI ever since they partnered to release a version of their Bing search engine and chatbots powered by generative AI in February 2023. OpenAI managed to overtake Google ahead of the I/O developer conference earlier this month, unveiling its powerful GPT-4o AI model the day before the show began.
But if Google ends up deploying a product that generates errors or harmful information in order to beat competitors, it risks giving users the impression that its generative AI efforts cannot be trusted and are ultimately not worth using.
Contact Daniel Howley at dhowley@yahoofinance.com. Follow him on Twitter. Daniel Howley.
Click here for the latest technology news impacting the stock market.
Read the latest financial and business news from Yahoo Finance
