There has been a consistent convergence of generative artificial intelligence (AI) and web searches, with chatbots playing hosts. Yet, traditional search engines as we know them, held on to convention. That’s now changing, as the next chapter of generative AI evolution sees Google launching the Search Generative Experience (SGE) for Google Search, for users in India. For the sake of simplification, this is what it means – you no longer need to open a chatbot such as Google Bard if a layer of generative AI is what you want with your search experience; Google Search will now have its own generative AI layer.
For now, the Search Generative Experience (SGE) remains an opt-in for Google Search users, the toggle a part of the Search Labs suite of experimental products. You can still choose to retain the traditional Google Search experience, for the duration it lasts. This will be able to process queries, and respond in English and Hindi. There are certain conditions you’ll need to meet, before the SGE toggle is available – you either need to use Google Search on the Chrome web browser, or use the updated Google app on an Android device, or an iPhone or iPad.
Google Search with SGE enabled will visually be quite different from conventional search. “Normally, you might break one question down into smaller ones, sort through the vast information available and start to piece things together yourself. With generative AI search can do some of that heavy lifting for you,” says Puneesh Kumar, General Manager of Google Search for India, in a briefing of which HT was a part.
What SGE will do is create an AI generated snapshot of the information that’s crucial to your search query. Following that, will be familiar territories, with extended search results linking you to various relevant websites.
Kumar confirms that at work, behind the scenes, are multiple large language models (LLMs) including an advanced version of Google’s MUM as well as PaLM 2. “Snapshots are designed to highlight information that can be easily corroborated by the range of results we show,” he adds. Google India confirms snapshot pointers will be accompanied by relevant web links, from where the curation has been done. A user will still have the power to evaluate suggestions and results.
There will be instances where the generative AI layer may not be invoked for certain search queries. These would particularly be about very precise knowledge, such as finance or health. This is where the balance between confidence and triggering SGE will be found, with Google seemingly willing to take it slow, for the sake of accuracy.
“In some cases, for example, there’s a query about finance or a query around health, we need very high confidence before we are able to show that over to the user. In those cases, we may not trigger the generative AI at all right and to begin with, but over time as we are collecting feedback as we are getting more input from users, we will keep fine tuning the model and bringing those experiences to light,” Kumar clarifies.
Whether the SGE layer is available for a search query will depend on how confident the LLM is with its understanding of the specificity of the query, any domain expertise and the accuracy of responses thereof. This will change with time – a query that’s low confidence and doesn’t trigger SGE may trigger SGE in due course.
This is a failsafe of sorts, in place, and the weightage will change with time.
“Wherever we don’t have high confidence, then we would be more responsible and not trigger SGE at all. And we’ll continue to learn and iterate and that model,” he adds. Google says the privacy policy for Search with SGE doesn’t change from traditional search, and that includes not using any personally identifiable data for AI training.
New to the SGE evolution will be a conversation mode, for follow-up questions that can also be asked in voice, in English and Hindi. This will negate the need for typing each query as a separate search. Users can switch between the two languages in the midst of an ongoing search.
There is the question about Google Bard, the popular artificially intelligent chatbot that competes with OpenAI’s ChatGPT and Microsoft’s Bing chatbot, for your attention. “Bard will be complementary to search, we don’t see that as a choice between Bard and SGE, but to give people multiple tools,” says Kumar.
Google believes Bard and SGE are still very different experiences. The latter focuses on information journeys, which is the foundation for search results. But similar underpinnings with these LLMs, you will notice there is bound to be some overlap.
