Google is currently managing a series of unexpected challenges with its AI Overview tool, which has been producing bizarre and inaccurate responses, leading to significant public attention and criticism.
The feature, initially launched in beta as the Search Generative Experience in May 2023, was designed to enhance user interactions with Google’s search engine by providing generative AI-driven summaries. However, the rollout has been marred by unusual advice from the AI, such as suggestions for users to add glue to their pizza or consume rocks—advice that has quickly turned into internet memes.
As the company navigates these issues, it has taken steps to manually disable the AI Overviews for specific search queries, particularly those that have been most problematic or have gone viral on social media platforms. This reactive measure has resulted in many users noticing the disappearance of these AI-generated snippets shortly after they gain attention online. Despite having been under testing for a year, the problems have cast doubt on the readiness and reliability of this AI feature.
Sundar Pichai, CEO of Google, has previously highlighted the success of the AI Overview, noting that it has processed over a billion queries since its inception. Pichai also pointed out that the cost of delivering AI-driven answers has been reduced by 80% due to improvements in hardware and efficiencies in engineering and technical design. Nonetheless, this optimization, perhaps premature, seems to have preceded a thorough vetting of the technology’s readiness for widespread public use, which might have contributed to the current issues.
The implications of these AI mishaps extend beyond mere operational hiccups, reflecting on Google’s reputation for innovation and quality. According to a statement from an anonymous AI company founder to The Verge, Google, known for its cutting-edge technology and high-quality outputs, is now becoming synonymous with lower-quality results that have become the subject of memes.
In defense of the AI Overview tool, Google spokesperson Meghann Farnsworth conveyed to The Verge that the majority of the problematic outputs were linked to uncommon or altered queries. Farnsworth reassured that the company is actively removing inappropriate AI-generated content as per their content policies and that these instances are being used to drive broader improvements in their systems, some of which have already been implemented.
The conversation about AI’s reliability is further complicated by insights from Gary Marcus, an AI expert and emeritus professor of neural science at New York University. Marcus critiques the overly optimistic projections of AI technology, arguing that while initial accuracy might be easier to achieve by approximating large volumes of human data, reaching full accuracy remains a formidable challenge.
Marcus suggests that the tasks required for complete accuracy, such as logical reasoning and source verification, would need capabilities akin to artificial general intelligence (AGI)—a level of technological advancement not yet realized.
Marcus also commented on the competitive pressures within the AI industry, noting other significant developments such as Microsoft’s early integration of AI into Bing and OpenAI’s ongoing work on a search engine. He referenced Meta’s 2022 release of Galactica, an AI that had to be withdrawn soon after its debut due to it generating dangerous suggestions, like advising people to eat glass—a scenario eerily similar to Google’s current predicaments.
Despite these challenges, Google’s ambitions for AI Overviews remain high, with plans to expand the feature to include more advanced capabilities like multi-step reasoning for complex queries and the organization of AI-driven results pages, as well as integrating video search capabilities into Google Lens.
Related News:
Featured Image courtesy of Jakub Porzycki/NurPhoto via Getty Images