The internet has always thrived on stupid questions.
Despite decades of asking them in online forums, on social media and to search engines, humans still seem to have an infinite supply of these less-than-serious-minded queries. What is changing is where and how we ask them.
AI chatbots and search tools such as ChatGPT have given people a new safe space to ask things they would rather not admit to knowing. Inside these sterile chat windows, there is no other human to witness our lack of expertise. Nobody to judge us for not knowing something.
But what do we lose when we move away from asking communities and humans to private, automated services?
“I do think that there’s something special about getting answers to your stupid question from other humans who perhaps once had that stupid question themselves,” says Casey Fiesler, an associate professor of information science at the University of Colorado Boulder. “You get answers that are informed by lived experiences.”
OpenAI says that 49 percent of ChatGPT conversations are people asking questions. Some of my own recent Claude searches include asking what coolant does in a car and what year my kids graduate high school. (It did the math wrong.)
There are a few potential drawbacks to getting AI answers to questions spurred by organic ignorance, says Fiesler. The responses lack the features of humanity — someone relating to or empathizing with you, sharing advice learned through personal mistakes, or revealing that maybe you’re not alone in your ignorance. When you lob a stupid question at other humans, it will sometimes catalyze a larger discussion, help forge new connections and increase a community’s bonds.
Getting a single answer from AI can also be less reliable than crowdsourcing from humans. On Reddit, you might get 10 answers that you can weigh up based on how reputable the poster is or how many times it was upvoted. By comparison, chatbots generally give one answer and are less likely to admit to not having one to give, an issue ChatGPT maker OpenAI has discussed in the past.
Most methods of evaluating responses given by a chatbot “measure model performance in a way that encourages guessing rather than honesty about uncertainty,” the company said in a blog post last year. (The Washington Post has a content partnership with OpenAI.)
“If you ask a chatbot, a lot of people have this baseline assumption that they can trust it unless they have reason not to, which is obviously not the case,” says Fiesler. Research has found evidence people can suffer from “automation bias,” the tendency to assume that any answer given by a machine is correct.
Asking chatbots questions is a new version of an old internet staple. A boom of Q&A sites started in the early 2000s, with Answers.com, Google Answers, Yahoo Answers and Ask MetaFilter providing an outlet for inquiries on all topics. The posts sometimes took on a life of their own. The Yahoo Answers question, “How is babby formed” became a lasting meme, and the multiple muddled queries about pregnancy were turned into the song, “Am I pregant.”
If you Google “Am I pregant” today, the AI-generated summary will dryly inform you of something along the lines that “A Google Search cannot determine if you are pregnant” and list early signs to look for offline.
The stockpiles of answers from those old Q&A sites have helped train some of the very same AI chatbots that many people turn to today.
There are still vibrant online communities where people ask questions to consult the wisdom of crowds. They use Quora and Reddit, or social media like in a Facebook group. Reddit even has a “NoStupidQuestions” subreddit, which is a popular place to ask almost anything and get a direct answer, without sass.
But it’s not easy to get a purely human answer to a stupid question anymore. Some answers on Quora and Reddit are now provided by people who simply paste the response from an AI chatbot. Tap your question into a search engine like Google and it will provide an AI answer above any human-written results. On Facebook, Meta’s AI chatbot suggests follow-up questions under links like news posts, even when it doesn’t have the answer.
In 2024, a person asked a New York City parenting group for advice about navigating the school system with a gifted and disabled child. Meta’s AI chatbot left a comment under the post, declaring that it, too, had a child and that they were familiar with the city’s gifted and talented programs.
“There’s a reason that you go ask a question in a Facebook group or on Reddit,” says Fiesler. “You intend to ask people that question.”
The post ChatGPT is changing how we ask stupid questions appeared first on Washington Post.
