Dealing with India’s vast language diversity, digital content platforms find it notably challenging to accurately present and translate material. Google’s recent deployment of AI overviews in India shares this dilemma.
With the introduction of Hindi support for these AI overviews in August, users were given the choice to toggle between Hindi and English directly from the search page. Unfortunately, this feature has shown to be notably uneven in performance.
An example highlighted was a Google blog’s GIF response to the question, “What can be used instead of sugar in tea?” in Hindi. Altering the word order in the query completely stumped Google, with the search engine failing to produce an answer.
Another discovered issue is the literal translation of English terms. A question in Hindi about suitable summer foods yielded the confusing term “sticky things,” a clear misinterpretation when the English switch provided the term “oily,” illustrating a significant discrepancy in translation accuracy.
A glaring error was noted when querying about YouTube’s proprietorship; the Hindi AI overview erroneously stated a past tenure for Neal Mohan as Google’s CEO, contrasting the accurate current CEO status in the English result.
Quantitative and qualitative inaccuracies also marred the AI’s performance. Queries about festive dates or dietary recommendations returned outdated or inconsistent information, revealing a weakness in the system’s context comprehension and reliability.
More disturbingly, sensitive topics like menstruation and pregnancy were addressed unsatisfactorily, with advice ranging from vague to bizarre, highlighting a critical need for improvement in source material assessment and algorithmic judgment.
While similar English language inquiries demonstrated marginally better responses, the disparity underscores the prevalent issues of context misinterpretation and mistranslation inherent in Google’s systems.
Despite this, Google maintains that its AI overviews trigger only for queries with highly trusted output quality, as per statements to TechCrunch. The company emphasizes its continuous effort to refine and enhance search features, attributing any inaccuracies to misread web content or source inaccuracies.
However, when Google’s algorithms opt to summarize questionable content, it’s the platform that users hold accountable for any misleading or unclear information presented.
Google’s endeavor into AI-powered search overviews has not been without controversy, previously generating bizarre recommendations from unreliable sources. For a nation with over 830 million internet users, many relying on Google for information, the tech giant faces pressing demands to enhance the precision and reliability of its AI overviews, especially in local languages.