South of San Francisco Bay, in Mountain View, lies one of the greatest sources of profit in corporate history. This is the headquarters of Google, whose search engine has been humanity's preferred gateway to the Internet for two decades…and advertisers' preferred gateway to humanity. It is estimated that Google processes perhaps 100,000 web searches every second and, thanks to its intelligent algorithm, returns amazingly relevant answers. That power has turned the name Google into a verb. It also provides billions of daily opportunities to sell ads alongside answers to searchers' queries. Accuracy of results keeps users coming back and keeps rivals at bay: together, all other search engines attract barely a tenth of your daily searches (see chart 1).
Advertisers pay handsomely for access to Google users; and they are usually only charged when someone visits their website. Revenue at Google's parent company, now called Alphabet, has grown at an average annual rate of more than 20% since 2011. In that period, Alphabet has generated more than $300 billion after operating expenses (see chart 2), most of them coming from searches. Its market value has more than tripled, to $1.3 trillion. It is the fourth most valuable company in the world. Unlike Apple and Microsoft, its big middle-aged tech rivals, it hasn't felt any real need to reinvent itself. Until now.
The subject of the current insight into Mountain View is ChatGPT, an artificial intelligence-based chatbot designed by a startup called OpenAI. In addition to being able to carry on a human-like conversation, ChatGPT and other similar chats can compose human-like poems, history articles, computer code, and just about anything else someone thinks of writing. UBS bank estimates that, since its launch in November, ChatGPT has amassed some 100 million monthly active users, a feat that took nine months for TikTok, the fastest-growing social media sensation. Other "generative" artificial intelligences are capable of painting, composing or singing. Bill Gates, co-founder of Microsoft, has said of this technology that it is "as important as the computer or the Internet."
For Eric Schmidt, who previously ran Google, ChatGPT is the "first widely visible example" of what a human's AI friend could be. For his former boss, it is the first visible threat to Google's dominance in search. And it is that ChatGPT is also capable of answering the type of questions that people could ask Google. ChatGPT's creator company, OpenAI, has teamed up with Microsoft, which is eyeing Google's dazzling profits with covetous eyes.
On February 7, Microsoft, which had recently announced a $10 billion investment in OpenAI, made public how it plans to go after those benefits. From now on, results from Bing, the software giant's search engine, will be accompanied by an artificial intelligence-generated sidebar summarizing relevant information. Bing will also have its own chatbot based on OpenAI models. Microsoft showed off some skills, like making a grocery list for a week's worth of planned meals. "It's a new dawn in search," said Satya Nadella, CEO of Microsoft. Baidu, the main search engine in China, where Google is banned, will launch its own AI-powered service in March. In a preemptive counterattack, and to steal some of the limelight from Nadella, Alphabet last week introduced its own chatbot, Bard; In addition, it has invested 300 million dollars in Anthropic, a generative artificial intelligence startup. On February 8, while unveiling some non-chat AI-powered search features, he confirmed that Bard would be integrated into search within weeks. The announcement did not impress investors: Alphabet shares fell 8%.
So, the battle of the bots is brewing. It's a battle that will transform the way you find things on the Internet. And by doing so, it will upend the lucrative business of search.
The last time online search was disrupted was in its infancy, at the dawn of the consumer Internet in the late 1990s. As web pages multiplied, it became harder to find useful information. Various search engines, such as AltaVista and Yahoo!, made things a bit easier. However, it was Google, founded in 1998, that revolutionized the sector. Their algorithm ranked web pages based on the number of websites linking to them, which turned out to be a good indicator of relevance. Later, Google discovered that next to the results it could show ads related to the keywords of a search.
Other competitors have emerged in recent years. Some are startups that offer ad-free subscription search, like Neeva. Others are Alphabet's big tech rivals. Amazon, whose digital emporium has become where many shoppers begin their search for products, has seen its share of the US search ad market jump from 3% in 2016 to 23% today. Apple's search ad business, consisting of app searches on iPhones, now has 7% of that market, up from none a few years ago. Google studies show that two-fifths of 18- to 24-year-olds prefer Instagram, Meta's photo-sharing app, or choose TikTok over Google Maps when looking for a nearby restaurant.
As a result of that effervescence, Google's share of revenue from search advertising will fall this year in the United States to 54% from 67% in 2016, according to research firm eMarketer. Yet those other companies have never posed an existential threat to Google. The same cannot be said for chatbot-assisted conversational search. Following the launch of ChatGPT, Sundar Pichai, Alphabet's CEO, is said to have declared a "code red."
To understand Google's fear of chatbots, let's first consider the technology behind them. ChatGPT works by predicting the next word in a phrase that constitutes the answer to some query. The predictions are based on a "great linguistic model", the result of prior analysis of millions of texts collected on the Internet. Once trained using natural language, the chatbot is capable of producing a well-written response based on user instructions, rather than just providing a list of links.
Applied to searches, that means that, in principle, responses could contain many more variables. Want a day trip to an off-the-beaten-path spot that's cheap, kid-friendly, and educational? Unless one stumbles upon a travel blog post, finding an accurate answer on Google (or Bing or Baidu) today requires comparing dozens of websites and skimming a great deal of text. Instead, ChatGPT offers a list of good options in seconds. Users can add additional considerations or ask for more information with new questions.
Changing the way you search will, in turn, change what you search for. In addition to searching for existing information, users can use conversational search to generate original content. ChatGPT writes poetry and articles in the style of our favorite author, if we ask. On January 26, Google published an article describing MusicLM, a great new language model capable of creating music from text. GitHub, a Microsoft-owned platform that hosts open source programs, has a chatbot called Copilot that can write lines of code. According to Bernstein's Mark Shmulik, that opens up new markets adjacent to search, especially in business productivity tools (like helping employees write presentations).
As a new field, conversational search attracts hopeful newcomers, encouraged by the prospect of an expanding market for search and generative content. "When I started two years ago, people said I was crazy. Now the opinion has changed a lot," says Richard Socher, founder of You.com, a startup that offers an artificial intelligence-powered search chatbot. Neeva has also added a chat bot to its subscription search. Sridhar Ramaswamy, its co-founder, hopes this will help him reach 5-10 million subscribers from almost 2 million now (not all paid), and become financially self-sufficient. C3.ai, a corporate software company, has launched a chatbot to help companies dig into their internal data. Travel companies like Booking.com are also playing with chatbots.
The most serious threat to Google comes from Microsoft. Nadella's company already has the necessary infrastructure in place, including enormous computing power, storage systems, and armies of tracking programs constantly pulling information from the Internet. Creating the same from scratch to compete with Google would cost, according to an estimate by the Competition and Markets Authority, a British antitrust body, between 10,000 and 30,000 million dollars.
Currently, Bing's share of the US search advertising market is a meager 5% (see chart 3). Microsoft hopes that its planned innovations will change the situation. The company seems to have fixed some of the shortcomings of ChatGPT. One of them was to keep the bot updated. The artificial intelligence on which ChatGPT is based, called GPT-3.5, has been trained on data from 2021 and has no clue about anything that happened on the internet after that date. If you ask for the latest news or today's weather forecast, what you get is an apology. Instead, Bing's artificial intelligence decides how to gather the most relevant information and uses search tools to find it. The data is then fed into the model, which uses it to compose a well-written response. Other companies, like Neeva, also use the same method.
This, in turn, has helped Microsoft address a bigger problem: the tendency of big language models to make stuff up. Chatbots don't know how to distinguish between what is true and what is false; they reflect what is on the internet, with all its flaws. Those "hallucinations" that seem loaded with authority (they get that name in computer language) are harmless when the bot is used for entertainment. However, when it comes to giving real answers to serious questions, they are a fatal flaw. Last year, Meta had to retire Galactica, her science chatbot, after discovering it was spouting scientific nonsense.
Giving the model access to up-to-date data has reduced, but not eliminated, the hallucination rate of Bing's chat feature. "A lot of the hallucinations were because [the model] was trying to fill in the gaps related to things that were post-learn data," explains Kevin Scott, Microsoft's CTO. His company is using other techniques to further reduce the error rate. Among them, asking humans to tell models which responses are best and which information is reliable; adding memory to systems so that algorithms learn from conversations, which they currently don't; and put links to sources in answers generated by artificial intelligence. Schmidt also assumes that chatbots will have left hallucinations behind in a year or two.
Solving the technological problems is only the first step in evicting Google from the search podium. No less complex is finding a way for conversational search to generate benefits. For starters, the costs are still high for chatbots compared to conventional search. Brian Nowak, of Morgan Stanley, estimates that, due to the additional computing power required, it costs about two cents to respond to a ChatGPT query, about seven times more than a Google search. According to Nowak, every 10% of Google searches that switch to artificial intelligence between now and 2025 will mean, depending on the number of words in an average answer, an increase of between 700 and 11.6 billion dollars in the operating costs of the company. company, which is equivalent to between 1% and 14% of such spending in 2022.
To further complicate matters, many expensive conversational search queries will generate little ad revenue. Google has stated that 80% of their search results do not contain lucrative ads in their header. Many of those ad-free searches are almost certainly "informational" ("what is the capital of Spain?"), just the kind of query that chatbots are most useful for, and just the kind that people are least interested in. advertisers (it's hard to know which ad to place next to the word "Madrid"). For generative artificial intelligence to make real money, it must find uses in "navigational" searches (searching for a site's Internet address by name) and, above all, "commercial" searches ("the best boots of ski this season").
Some companies, like Neeva, earn revenue from subscriptions. On February 1, OpenAI started registering subscribers of the current version of ChatGPT. For $20 a month, users get faster responses and access at peak times. OpenAI also plans to license the technology to other companies. However, the most profitable will be advertising.
Inserting ads into what must appear to a user as normal conversation will require skill. One possibility is to show fewer ads but charge advertisers more, Nowak says. A chatbot is likely to offer only a few suggestions in response to a search on, say, paradise hotels in Hawaii. It is safe to assume that hoteliers will be willing to pay more to ensure that their offer is listed among or next to the suggestions. Microsoft announces that it plans to test that model on the new Bing.
Microsoft may be betting on the fact that chatbot-assisted informational searches will attract new users, who will then also use Bing for more lucrative queries. That could mean sacrificing margins, at least until costs can be reduced. Such a move will only be worthwhile if it succeeds in wresting considerable market share from Google. Microsoft expects that for every percentage point of market share gained in search, annual advertising revenue will increase by $2 billion.
It is possible, but it is not a guaranteed result. Alphabet retains formidable strengths. One of them is technology. Although Google has yet to integrate generative AI into its search engine, it has been deploying other AIs in its search business for years. When a web page snippet stands out at the top of the search result, it's thanks to role models like Bert and Mum. All of this "is only possible because of the fundamental research we've done in intelligence for more than a decade," says Liz Reid, Google's head of search. Despite the blunder on launch day, when she misidentified the first telescope to photograph an exoplanet, Bard is probably no less impressive than ChatGPT. Paradoxically, that stumble reinforced the reasons the company gave for not rushing it: the fear of launching a chatbot that offered inaccurate content.
The other advantage of Google is its position. It's the default search engine for Chrome, Alphabet's browser, which is used by two out of three Internet users, according to research firm StatCounter. It is also the search engine present in more than 95% of mobile phones in the United States. And the company pays Apple about $15 billion a year to make its search the default on Apple devices and Safari, which accounts for 19% of browsers installed on desktops and devices.
However, this position of strength also entails some weaknesses. Alphabet finds it hard to move quickly with the breath of regulators on the back of its neck watching out for alleged monopolies and misinformation about its various platforms. Even more difficult it may be for the company to part with a technology and business model that have been consistently profitable for 20 years. Pichai has yet to find the best way to solve this "innovator's dilemma." Nadella hopes that before he does, Bing too will have become a verb.
© 2023 The Economist Newspaper Limited. All rights reserved.
Translation: Juan Gabriel López Guix
