Last week, Google made headlines with the unveiling of its biggest change to search in years, showcasing new artificial intelligence (AI) capabilities. The tech giant’s aim was clear: to keep pace with competitors such as Microsoft and OpenAI.

The repercussions have been profound, eroding trust in a search engine over two billion people rely on for reliable information. While other AI chatbots have been known to disseminate falsehoods and exhibit odd behaviour, the backlash against Google underscores the mounting pressure on the company to integrate AI into its search engine responsibly.

Pattern of missteps

This latest debacle is part of a concerning pattern for Google, with past AI initiatives also facing teething problems upon release. For instance, when Google introduced Bard in February 2023 to rival ChatGPT, inaccuracies about outer space led to a staggering $100 billion drop in the company’s market value.

Earlier this year, Google rolled out Bard’s successor, Gemini, touted as a chatbot capable of generating images and serving as a voice-operated digital assistant. However, users quickly noticed Gemini’s reluctance to generate images featuring white people and its propensity for inaccurate depictions of historical figures.

Critics within the tech industry have lambasted Google for its repeated missteps, but financial analysts argue that the company is under immense pressure to innovate rapidly, even if it means encountering hurdles along the way.

Thomas Monteiro, an analyst at Investing.com, told The New York Times, “Google doesn’t have a choice right now. Companies need to move really fast, even if that includes skipping a few steps along the way. The user experience will just have to catch up.”

In response to the outcry, Lara Levin, a spokesperson for Google, maintained that the majority of AI Overview queries yield high-quality information, complete with links for further exploration. She noted that many of the reported issues stemmed from unusual queries or manipulated examples and pledged to refine the system using isolated instances of problematic answers.

Challenges of AI integration

The push to integrate AI into Google’s suite of applications gained momentum following the success of OpenAI’s ChatGPT in 2022. However, the challenge lies in harnessing large language models that learn from vast swathes of online data, including falsehoods and satire, rather than being programmed conventionally.

Despite the rocky start, Google remains committed to advancing its AI capabilities. The unveiling of AI Overview at the annual I/O developer conference marked a significant milestone, integrating Gemini, the latest large language AI model, into its flagship search engine.

AI Overview, a fusion of language model-generated statements and live web snippets, was designed to tackle complex queries more effectively than conventional search methods. While Google emphasised the potential benefits of this integration, the reality has been fraught with complications.

AI Overview advised certain users to incorporate non-toxic glue into their pizza sauce as a measure to prevent cheese slippage, a concocted recipe seemingly derived from an 11-year-old jestful post on Reddit. Furthermore, the AI suggested to other users the consumption of at least one rock per day for the intake of essential vitamins and minerals—counsel originating from a satirical piece by The Onion.

Gergely Orosz, a software engineer renowned for his technology newsletter, Pragmatic Engineer, underscored Google Search as the linchpin of the company’s relevance, trustworthiness, and utility. In a post on X, he remarked, “And yet, examples of how AI Overviews are turning Google Search into garbage are all over my timeline.”

Public outcry and social media response

Instances were shared wherein Google prominently advocated the use of “chlorine bleach and white vinegar” for washing machine cleaning, a blend that, when combined, poses risks of generating harmful chlorine gas. Subsequently, in smaller font, users were instructed to apply one substance first, followed by the other.

Social media users engaged in a competitive display of the most absurd responses from Google, occasionally resorting to fabricated results. One manipulated screenshot purportedly depicted Google recommending jumping off the Golden Gate Bridge as a remedy for depression, citing a Reddit user. However, Levin refuted the claim, asserting that such results were not returned by the company’s systems.

AI Overview encountered challenges in historical accuracy, as evidenced by screenshots posted on X, erroneously stating that 17 presidents were white and incorrectly identifying Barack Obama as the first Muslim president. Additionally, it falsely claimed that Andrew Jackson graduated from college in 2005.

First Published: May 27 2024 | 12:41 PM IST