LLM chatbots, search engines will co-exist, says Google’s Raghavan

Large language model (LLM)-powered chatbots like ChatGPT have not only amassed millions of users but are also being increasingly integrated with search engines, putting a question mark on traditional search.

Google's senior vice-president Prabhakar Raghavan, however, believes that LLMs and search engines will co-exist. "I don't believe any one approach—LLM-based chatbots or search engines—will completely replace the other," he told Mint during his recent visit to Bengaluru.

Raghavan, who heads Google Search, Ads, Assistant, Maps, commerce and payments products, said he cannot share the number of queries that the world is asking daily, "but if that number is going up faster than the number of users who are accessing the internet, then it's saying something about human behavior: their curiosity is expanding the way they are". He also believes the future of search is "not just about offering direct answers or a lists of links but creating a more dynamic, synthesized experience".

A computer scientist at heart, Raghavan explained that while traditional searches provide static results like maps and reviews, LLMs can dynamically categorize options, such as "seafood restaurants" or "romantic spots," enhancing user engagement. However, he warned that “fluid and engaging responses” might compromise factual accuracy, necessitating a balance.

Also read | Chips, Indic LLMs, flying taxis: Inside India’s tryst with tech independence

Regarding the impact of the shift to LLMs on Google’s advertising revenue, which heavily relies on ads, Raghavan believes AI-driven ad personalization will keep advertising strong. AI can match ads better with user queries, potentially evolving ads into a more tailored experience. Subscription models could also coexist with ads, aligning with Google's mission of universal access, according to Raghavan.

In this context, Raghavan highlighted India's significant contribution to AI and tech, driven by a large talent pool and educational investments. He noted India's "substantial impact" through innovation but stressed the importance of understanding local needs, such as the challenge of recognizing Hinglish (combination of Hindi and English) due to limited training data. Having a workforce in India, he explained, allows Google to address these unique requirements and gain valuable market insights, enhancing its global leadership.

He also acknowledged the gradual integration of LLM-based chatbots into commerce and payments but emphasized the need for caution. Due to the high stakes in transactions, any inaccuracies, such as incorrect product dimensions, could harm customers. Hence, he explained, Google is moving slowly to ensure safety and reliability in these areas, although broader integration is expected soon.

Raghavan also defended the future of search engine optimization (SEO), calling it an essential business practice. A study by researchers from Germany's Leipzig University and others, for instance, raised concerns about search engines' handling of highly optimized affiliate content, suggesting that the line between legitimate and spam content is becoming blurrier, especially with generative AI.

While AI could improve content quality, Google will strictly target “made for search engine content” that misleads users. “I don't believe that SEO is somehow antithetical to search. My view is: if it helps the user get what they want, it's a good thing," he said. But he qualified that the so-called "made for search engine content" (avoids the word, 'click bait') misleads users.

Also read | Mint Explainer: The mercurial rise of India-focused LLMs

Despite concerns that generative AI might worsen content quality, Raghavan argued that Google's focus is not on whether content is AI-generated but on whether it is misleading. Raghavan noted that content quality has always been a challenge but dismissed fears of AI exacerbating this issue. According to him, while language models can mimic intelligent conversation like humans, they are far from achieving tasks like proving complex mathematical theorems. Yet, they already perform tasks beyond human capability, such as folding 200 million proteins for drug discovery.

Thus, using human intelligence as the benchmark for AI will not help, believes Raghavan. He added that while AI can assist in complex problem-solving, like proving mathematical theorems or decoding Ramanujan’s notebooks, it is unlikely to replace human roles. Also, "while AGI (Artificial General Intelligence) is a debated topic, the definition remains unclear. Instead of fixating on AGI, the focus should be on enhancing current AI capabilities to provide more context-aware, meaningful assistance", he said.

Strictures against Google

When asked about the regulatory scrutiny over Google’s market dominance in digital advertising, Raghavan acknowledged that balancing the needs of users, publishers, and advertisers is challenging. "Google is navigating these tensions carefully, ensuring a balanced approach that respects all stakeholders. The solution remains a work in progress, but it is central to Google's ad technology strategy," he said.

For the past four years, Google’s dominance has attracted the attention of antitrust enforcers like the US Department of Justice (DOJ). On 6 August, the DOJ ruled that "Google is a monopolist, and it has acted as one to maintain its monopoly..." in the online search market. Kent Walker, president, global Affairs, responded on X that Google plans to appeal . On Monday, 9 September, Google will face a second major antitrust trial that is focusing on its dominance in the advertising market. However, according to Wedbush Securities, while there is potential for "near-term headline risk" or reputational damage due to the trial, the financial impact appears minimal.

Traditional AI to GenAI

Meanwhile, Raghavan believes traditional AI and Generative AI (GenAI) are "evolving into a more unified, multimodal experience for users. Today, users don’t care if it’s text, image, video, or speech—everything blends together seamlessly". To support his view, he cited the example of Google Lens with which a user can point to curtains and say, "I want that in green", mixing image and voice to find a product.

Also read | Errors, high cost among reasons GenAI not moving beyond concept stage

"Younger users, in particular, see their devices as extensions of their senses, expecting a fluid interaction. This merging of AI sub-disciplines (machine learning, deep learning, computer vision, image recognition, natural language processing (NLP), and GenAI, to name a few) into a cohesive, multimodal experience represents an exciting shift in AI development," Raghavan explained. He added, "The goal isn’t to showcase AI for its own sake, but to enhance user satisfaction, regardless of whether it’s achieved through LLMs, traditional indexing, or other methods.”

Strategy and Leadership

When asked how a research-focused executive works with a business-oriented leader like Alphabet CEO Sundar Pichai, Raghavan said, "Successful leadership teams thrive on such diversity and debate, avoiding unhealthy environments where directives come from a single voice. Over time, an equilibrium is naturally found through collaboration and discussion."

He added that as a leader, it's important to know where and when to get involved, and when to step back and trust your team. "Once you've decided to delegate, you must trust them to deliver, but set a timeline to check in—be it in two weeks or two months. If things aren’t going as planned, the key is to ask the right questions to identify the problem, rather than just demanding to know why it isn't working,” he elaborated.

Raghavan believes that to succeed in research/science or in a business role, one must consistently ask the right questions. “In science, the wrong questions lead to trivial work or unsolvable problems. Similarly, in a business role, not asking the right questions results in irrelevant or unsuccessful work,” he explained.

Google's overarching vision, with its range of tools like foundation models, LLMs, SLMs, search engines, ad model, assistants, commerce and payments, "is to make user journeys more efficient", Raghavan added. For instance, planning a vacation involves coordinating flights, accommodation, and activities.

"Current language models can assist, but they often miss nuances, like suggesting visits to the Louvre (museum in Paris) on a closed day. As these models improve, even a 20% increase in user efficiency represents a significant economic opportunity. The priority is to enhance user experience; economic gains will naturally follow," he concluded.

And read | The future of AI: Beyond language models to becoming real world decision makers