CHATGPT, AN ARTIFICIAL-INTELLIGENCE (AI) chatbot produced by OpenAI, a California-based startup, can generate human-like answers to questions. It can write poems, computer code and much more besides. UBS, a bank, reckons that since ChatGPT’s launch in November it has attracted 100m monthly active users. On February 7th Microsoft announced that it will infuse Bing, its search engine, with some of ChatGPT’s magic, thanks to a partnership with OpenAI. On the day before, Alphabet, Google’s parent company, had said that it would soon launch Bard, its own AI chatbot. Baidu, China’s search giant, has promised to do the same. Online-search startups, including Neeva and You.com, have AI chatbots, too. How will this flurry of innovation change search?
ChatGPT and chatbots like it do not work like conventional online search. Today, Google and Bing work by sending out web crawlers, a type of bot (autonomous programme), to roam the internet and collect information, before organising the results and returning the most relevant links to the user in response to a search request. By contrast, AI chatbots are built on large language models: algorithms trained on vast chunks of the internet. The bot predicts the likeliest next word in a response to some query based on its reading of billions of sentences that use the preceding words. Instead of serving up a list of links, chatbots can return fluent answers.
This ability means that search will change in two big ways. First, results will be able to answer complex questions with multiple variables. Imagine a tourist who is looking for a dog-friendly hotel in Hawaii that is close to the beach and within walking distance of a shopping centre. To find his ideal place on Google might mean skim-reading dozens of websites. The chatbot’s ability to predict the next word in a sentence allows it to recommend a bolthole in moments—then plan a week-long holiday for him.
Second, the market will expand. School children already ask ChatGPT to help with (and sometimes cheat on) history homework; software engineers ask it to debug code. Other applications will no doubt crop up to help diagnose illnesses, for example, or whip up PowerPoint presentations. This may result in an explosion of industry-specific chatbots. Already C3.ai, a business-software firm, is planning to use similar technology to help companies search their in-house data. Travel firms, including Booking.com, are experimenting with chatbots, too. For the first time in decades Google’s monopoly is at risk. It currently has 93% of the search market by volume, compared with Bing’s 3%. But it has not yet launched Bard, possibly because it fears a public backlash if the bot delivers offensive answers.
Chatbots have plenty of shortcomings. A big one is that they can make stuff up. These “hallucinations” happen because the models do not actually understand what they are saying. Chatbots might also be hard to monetise; it’s harder to insert ads into a conversation with an AI chatbot than into a Google search. (Microsoft says it will experiment with adding ads into its chatbot’s answers on Bing.) The cost of the additional computing power needed to run chatbots will probably eat into search engines’ margins. Some firms, including Neeva, are trying to get users to buy subscriptions to their chatbot services. Whatever their imperfections, chatbots have the potential to transform the way people interact with the internet and information. Bill Gates, the co-founder of Microsoft, has called the technology “as important as the PC, as the internet”. Changing how you find your dream holiday may be just the start. ■
