Dusseldorf, San Francisco For about a quarter of a century, a Google search almost always looked the same: users typed a few words into the search box, and the search engine returned a list of links highlighted in blue.
A fundamental change is now looming with artificial intelligence (AI). At the developer conference on Wednesday, Google announced a series of AI products. “We are at a dramatic turning point,” said Google CEO Sundar Pichai.
One of the most important is the “Search Generative Experience”, the new search mask from Google: With the help of AI, users should be able to conduct a conversation. You can ask complex questions and get a longer text created by AI as an answer.
Even if the links don’t disappear completely, this is a big step for Google. For many years, the company has been very successful in selling advertising space with traditional search, and together with other Google websites, it generated 162 billion dollars last year. It currently has a market share of around 90 percent in the Internet search engine market.
However, the competition from Microsoft and its search service Bing has increased dramatically since the competitor integrated AI solutions from the start-up OpenAI and received a lot of attention. So Google needs to change, but is proceeding cautiously.
The share price rose by five percent at the start of trading on Thursday in a weak market environment. But can Google hold its ground against Microsoft? How does the new search work? What AI is behind it? Here are the most important questions and answers about Google’s new strategy.
What does the new Google search look like?
Google’s problem: The new search should also retain the “information quality” and hit accuracy, says Liz Reid, who is responsible for search at Google. “How do you balance the strengths of generative AI with that promise?” Finally, AI tends to make mistakes during lengthy dialogues.
The answer is a mix of a technology that works like OpenAI’s AI bot ChatGPT and the links known so far. Users can enter complex search queries that an AI answers in several paragraphs. In addition, Google shows so-called “AI snapshots”: Small tiles show websites that fit the topic – but are not the sources of the AI search, as is the case with Bing. Below the text of the AI you will find the list of blue links again. There is no advertising in the demo version, but there is enough space.
CEO Sundar Pichai presented a number of new AI-supported products on Wednesday.
(Photo: JASON HENRY/The NewYorkTimes/Redux/laif)
During her presentation, Google employee Cathy Edwards had suitable bicycles selected for a hilly, eight-kilometer route. “Google can now understand even complicated search inputs,” promised Edwards.
If you are not satisfied with the answer, you can ask questions.
Unlike Bing and the integrated AI language model GPT-4, you have to click on a window to do this. In this way, Google makes it difficult for longer conversations between users and AI, which experience has shown to produce so-called hallucinations over longer periods of use – and give wrong or half-true answers.
How to use the new Google search?
Google is very cautious. For the time being, the “Search Generative Experience” (SGE) is only available in the USA. But the Americans will also have to wait, access will only be activated “in the coming weeks”. If you then want to use the search, you have to use Chrome Desktop on a computer or the Google app on an iPhone or Android smartphone.
There is no information as to when SGE will be available in Germany. The offer is only available in English, although the AI has long been multilingual. According to Google, this is “so that we can incorporate feedback and continuously improve the offering over time”.
The procedure is in contrast to Microsoft, which also offers Bing and GPT-4 in other countries and expanded access from “Limited Preview” to “Open Preview” a few days ago. However, Microsoft also had to backtrack and limit the number of questions to keep the AI from hallucinating. At first there was no limit, then it was a maximum of six, and now it’s 20 questions.
What new AI is Google presenting?
At the developer conference, Google presented the Palm 2 language model. This is the new version of the “Pathway Language Model”, a language model with 540 billion parameters. GPT-3 from OpenAI works with 175 billion parameters, so far nothing has been known about the new version GPT-4. A parameter represents a relationship, such as the probability that the word “sun” is followed by the word “shining”.
Palm 2 is used in 25 Google products such as the chatbot Bard or the cloud applications Workspace. The company did not provide any information on which AI Google search is based on. However, it can be assumed that it is a Palm 2.
More about artificial intelligence
According to Google, the language model is particularly good at “reasoning”, i.e. at thinking and reasoning. And that is essentially the function of the new Google search: understanding and answering even complex entries. To do this, Palm 2 breaks down the input into smaller, simpler tasks.
The Palm 2 should also better understand the nuances of human language, such as puzzles or idioms – even across around 100 languages. In addition, the AI masters two dozen programming languages such as Python or Prolog.
Google CEO Pichai also announced that there will be a compressed version of Palm that will run locally on modern smartphones.
She has slimmed down slightly but is still able to help write computer code.
What is Google doing against ChatGPT?
The text robot Bard will be improved and unlocked for all users. Bard was Google’s answer to the success of the text robot ChatGPT from the start-up OpenAI. Two months ago, Bard was initially activated for selected users in the USA and England.
In tests, however, Bard performed significantly weaker than ChatGPT and produced several errors during the first product presentation. In response, Google parent company Alphabet lost around $100 billion in stock market valuations.
Screenshot from a demonstration of the new Google search shows the new site layout with long AI response, tiles next to it, and legacy links below.
“Bard has gotten significantly better,” promised the responsible Google manager Sissie Hsiao, referring to the new language model Palm 2. The AI model Lamda had previously served as the basis. Bard is now said to be better at math calculations or writing computer code. In addition, Bard is to work multimodally in the future, i.e. not only understand text input, but also be able to analyze images in cooperation with Adobe, for example.
In addition to English, Bard also supports Japanese and Korean. Other languages are to follow. When asked, however, Google did not provide any information as to when a German-language version would also be activated.
“Bard will soon be able to support the world’s 40 major languages and the timeline for expansion plans is not yet finalized.” The company is working closely with regulators.
Even the English version will not be available in Germany or in the European Union as a whole. Google expanded the Bard offer to 180 countries, but none of them are EU countries. The company did not provide specific information about the reasons.
The Microsoft counterpart Bing, which is operated with the OpenAI AI system, is already available in Germany and other EU countries.
Exactly why Google is excluding the EU has so far been unclear and is likely to further increase the gap to Microsoft and OpenAI.
What is Google doing with AI in the cloud?
Google rival Microsoft is a close partner of OpenAI. Microsoft CEO Satya Nadella announced plans to integrate OpenAI’s AI technologies into all of the company’s products. This includes AI support in Office products such as Word, Powerpoint and Excel.
Google has now announced similar functions for its own solutions. Under the name Duet AI, Google customers should be able to use an AI companion that can help with writing texts, create images for presentations or produce spreadsheets for business calculations. To do this, it announced three AI tools: Codey for programming, Imagen for image generation and Chirp for speech processing. “We will roll out these functions over the course of the year,” said the responsible Google manager Aparna Pappu.
More: Google wants to integrate AI into all products.