Hoover says that Andi avoids simply repeating text from search results. “It doesn’t make things up like other chatbots,” she says. People can decide for themselves whether or not that’s true. After collecting feedback from its users for the past year, the company’s chatbot will now sometimes admit when it’s not confident about an answer. “It’ll say, ‘I’m not sure, but according to Wikipedia…,'” says Hoover.
Either way, this new era of search probably won’t ditch lists of links entirely. “When I think about searching five years from now, we’ll still have the ability to look through results,” says Hoover. “I think that’s an important part of the web.”
But as chatbots get more convincing, will we be less inclined to check up on their answers? “What’s noteworthy isn’t that large language models generate false information, but how good they are at turning off people’s critical reasoning abilities,” says Mike Tung, CEO of Diffbot, a company that builds software to pull data from the web.
The University of Washington’s Shah shares that concern. In Microsoft’s demo for Bing Chat, the company hammered home the message that using chatbots for search can save time. But Shah points out that a little-known project Microsoft has been working on for years, called Search Coachis designed to teach people to stop and think.
Billed as “a search engine with training wheels,” Search Coach helps people, especially students and educators, learn how to write effective search queries and identify reliable resources. Instead of saving time, Search Coach encourages people to slow down. “Compare that to ChatGPT,” says Shah.
Companies like Andi, Perplexity, and You.com are happy to admit they’re still figuring out what search could be. The truth is that it can be many things.
“You don’t want to fight against convenience, that’s a losing battle in consumer tech,” says Socher. “But there’s some pretty fundamental questions about the entire state of the internet at play here.”