Google criticized as AI Overview makes obvious errors, saying President Obama is Muslim and that it's safe to leave dogs in hot cars
Is Google's AI Spreading Dangerous Misinformation?
Google's new AI Overviews feature in Google Search has faced criticism for providing incorrect and nonsensical answers, such as labeling Barack Obama as a Muslim president and suggesting it's safe to leave dogs in hot cars. Despite its recent launch, the feature has shown errors in basic queries and controversial responses, raising concerns about the accuracy and reliability of AI-generated content in search results. This issue follows similar problems with Google's Gemini image-generation tool, which also produced historically inaccurate and questionable outputs.
What specific inaccuracies have been reported in the responses provided by Google's AI Overviews feature?
Google's AI Overviews feature has been reported to provide several inaccuracies in its responses. Some examples include:
- When asked how many Muslim presidents the U.S. has had, AI Overviews responded, "The United States has had one Muslim president, Barack Hussein Obama."
- When a user searched for "cheese not sticking to pizza," the feature suggested adding "about 1/8 cup of nontoxic glue to the sauce." This was traced back to an 11-year-old Reddit comment.
- For the query "Is it OK to leave a dog in a hot car," the tool at one point said, "Yes, it's always safe to leave a dog in a hot car," and went on to reference a fictional song by The Beatles about it being safe to leave dogs in hot cars.
- The tool provided inaccurate information about staring at the sun for health benefits and eating rocks daily, attributing these claims to WebMD and UC Berkeley geologists, respectively.
- AI Overviews responded inaccurately to simple queries, such as making up a list of fruits that end with "um," or saying the year 1919 was 20 years ago6.
What was the reaction within Google to the rollout of other AI tools like Bard, and how has it influenced their approach to new AI features?
Google has taken a cautious approach with generative AI development, and has criticized others for pushing too hard, too fast, with their generative AI tools6. Some have viewed this as anti-competitive bias, and Google simply protecting its turf, as more people turn to tools like ChatGPT for search queries6. Google's view is that generative AI needs to be deployed slowly in order to mitigate misuse, which has already led to various issues in a regulatory sense. But today, processes in place to fully protect against this kind of misuse. In this sense, Google's AI generations could be safer, and more reliable in this respect, though OpenAI's tools have clearly taken top billing as the AI tools of choice, at least within this initial push3.
Google is evolving its own offerings, and they could soon become even more valuable, depending on how its systems continue to develop. There are now a wide range of options for generative AI search and creation, but eventually, the market will refine, and the best performing tools will arise, taking the bigger market share. Google's hoping that its more cautious, measured approach will win out in the long run, while it also has more resources and capacity than most of its rivals.