Google is introducing a new AI-powered way to discover places in Maps, initially available for select local guides in the US.
The new AI search allows users to ask Maps for suggestions based on their preferences. Google says it uses language models to analyze information about more than 250 million places and the knowledge of the Maps community.
Generative AI in Google Maps can be used to search for specific, niche or general suggestions based on user preferences.
For example, a user can ask for "vintage stores in San Francisco" and receive suggestions for clothing stores, record stores, and flea markets.
Users can also ask follow-up questions, such as "What about lunch?" to get suggestions for restaurants that fit the theme.
AI models consider photos, ratings, and reviews to generate suggestions and categorize results. They offer photo carousels and review summaries.
According to Google, AI search can also adapt to sudden changes in travel plans, such as suggesting indoor activities on a rainy day. Google says this experimental feature is just the beginning of its plan to enhance Maps with generative AI.
Google does not specify which LLM it is using for Maps. Presumably, it's a model from the new Gemini series, which is also used for search (see below) and the Bard chatbot (soon to be called "Gemini Advanced").
Google experiments with LLM search features
Google tries to integrate AI into its search offerings in many ways. The most ambitious, but also the most financially and socially risky, project is the Search Generative Experience (SGE).
The SGE provides an AI-generated answer to a search query instead of a traditional list of links. This feature has the potential to fundamentally change the way traffic moves around the web.
However, during Google's most recent quarterly earnings call, CEO Sundar Pichai sounded cautious about the rollout of AI search, stating that it's still in the earliest stages.
One of Google's key challenges is to balance the enormous costs of training and operating the AI models with the revenue potential, especially with potentially billions of searches per day. There are also complex legal issues, particularly around responsibility for AI answers.