The Challenges of Google’s AI Overviews in India
Google’s recent introduction of Hindi support for AI overviews in India aimed to cater to the country’s language diversity. However, the feature has proven to be disappointingly inconsistent and inaccurate. One issue is the translation of Hindi words, with literal translations often leading to confusing and incorrect answers. For example, when asked about food substitutes for sugar in tea, the answer provided was “sticky things” instead of “substitutes for sugar.” Similarly, when asked about YouTube’s ownership, the AI overview stated the incorrect CEO information in Hindi.
The problems with AI overviews extend beyond translation issues. When asked simple questions like when Diwali is this year, instead of a direct answer, users are presented with a paragraph about last year’s Diwali, followed by a carousel of links, and finally, the actual answer. Inconsistencies in answers are also observed when asking about eating spicy food.
The concerns become more significant when discussing topics like menstruation and pregnancy. Answers regarding when to have children after marriage were sourced from a slideshow article that relied on opinions from Quora and other blogs, leading to confusing and unreliable information. Similarly, when asked about food during menstruation, the answers suggested drinking milk with many things and lemon for mood swings, which may not be accurate or helpful.
The AI overviews also miss important context. For instance, when inquiring about food places in Delhi, the tool suggests that Bangla Sahib Gurudwara, a Sikh place of worship, is open round the clock and offers tea and Indian snacks. However, it fails to mention that these snacks are available outside the Gurudwara, resulting in a misleading summary.
While the English AI overviews seem to provide better results, the problems with the Hindi ones can be attributed to both Google’s systems and the quality of sources. Google claims that the feature only appears for queries with high confidence in the output, but inaccuracies can occur due to misinterpreted web content or reflections of inaccuracies on the web.
Google’s algorithm ultimately decides which questions and sources are summarized by AI, and users may hold Google accountable for displaying inaccurate answers. This is particularly concerning considering India’s large internet user base, with over 830 million users relying on Google for search.
To encourage more users to utilize AI overviews in local languages, Google must improve the accuracy of its language and content. While the company blames the quality of sources, it is ultimately responsible for the algorithm’s decisions and the overall user experience.