Google announced a year ago that they believed artificial intelligence would be the future of search. Now, it seems that this future has become a reality: Google is rolling out "AI Summaries," previously known as Search Generated Experiences (SGE), to users in the United States and plans to launch it globally soon. Soon, billions of Google users will see AI-generated summaries at the top of their search results. And this is just the beginning of how AI is changing search.

"We have observed that generative AI allows Google to do more of the search work for you," said Liz Reid, Google's new search director, who has been studying various aspects of search for many years. "It can alleviate many tedious tasks of searching, allowing you to focus more on what you really want to do or explore the content that interests you."

Reid listed a series of features aimed at achieving this goal, which Google officially announced at its I/O developer conference on Tuesday. Of course, the first is AI Summaries, which aims to give you a rough understanding of the answer to your query and provide additional resource links for more information. Lens also added a feature that allows you to search by shooting videos. There is also a new planning tool that can automatically generate travel itineraries or meal plans based on a single query. In addition, there is a new AI-based way to organize the results page itself, such as when you want to see restaurants in a new city, it may provide you with a bunch of restaurants suitable for dates and a bunch of restaurants suitable for business meetings without you having to ask additional questions.

This is truly a comprehensive AI revolution in search. Google is using its Gemini AI to accurately understand your search intent, whether you are typing, speaking, taking photos, or shooting videos.

It uses a brand new dedicated Gemini model to aggregate the web and display answers. It even uses Gemini to design and populate the results page.

However, Reid said that not every search requires so much AI involvement, nor does it always use it. "If you just want to navigate to a website, like searching for Walmart and wanting to go to walmart.com, adding AI functionality may not be really useful," she believes Gemini is most useful in more complex situations, such as those that require a lot of searching or those that don't start with Google.

One example that Reid particularly likes is local search. (You often hear this in AI because it can be tricky to find really good things through a lot of similar lists and reviews.) With Gemini, "we can do things like 'find the best yoga or Pilates studio in Boston with a rating of more than four stars and within a half-hour walk from Beacon Hill,'" she said. Perhaps, she continued, you also want to know which one offers the best deals for beginners. "So, you can get information across the knowledge graph and the web and integrate them together."

The combination of knowledge graph and AI - Google's old search tool and new search tool - is crucial for Reid and her team. Some issues in search have been resolved, such as sports scores: "If you really just want to see the score, this product works well," Reid said. In this case, Gemini's task is to ensure that you get accurate scores no matter how you ask. "You can consider expanding the types of questions that can successfully trigger scores," she said, "but you still need that kind of standardized sports data."

For Google and any other search engine, obtaining high-quality data is the core of the whole game. Reid told me that part of the motivation for creating the new search-specific Gemini model is to focus on getting things right. "There is a balance between creativity and factual accuracy in any language model," she said. "We really intend to lean towards the factual side." AI Summaries may not be as interesting or appealing, but as a result, they may give correct results more frequently. (Although there is no perfect model, Google will certainly face a lot of problems caused by fictional and completely wrong summaries.)

As AI enters the search field, some products like Perplexity and Arc have been scrutinized for browsing and summarizing web content without directly guiding users to actual sources of information. Reid said that this is a tricky but important balance, and Google tries to ensure accuracy by not triggering summaries in certain cases. But she also believes that early data suggests that this new way of searching will actually drive more clicks to open networks. She admits that this may weaken low-value content, but she said, "If you consider links as gateways for deep exploration, then websites that provide perspectives, colors, experiences, or expertise - people still need those." She pointed out that young users, in particular, are always looking for a human perspective on their queries and said that Google still has a responsibility to provide them.

For most of the past decade, Google has been trying to change the way you search. It started out as just a box where you could enter keywords; now, it wants to be an all-knowing presence where you can ask questions in any way you want and get answers in the most helpful way for you. "You add richness, allowing people to ask questions in the way they naturally would," Reid said. For Google, this is the key to getting more people to ask more questions, which will help Google earn more revenue. For users, this may mean a whole new way of interacting with the internet: less typing, fewer tabs, and more communication with search engines.