ALL TOO HUMAN

Why are we so susceptible to trusting AI chatbots?

- JACKIE FLYNN MOGENSEN

IN SUMMER 2019, a group of Dutch scientists conducted an experiment to collect “digital confessions.” At a music festival near Amsterdam, the researchers asked attendees to share a secret anonymously by chatting online with either a priest or a relatively basic chatbot, assigned at random. To their surprise, some of the nearly 300 participants offered deeply personal confessions, including of infidelity and experiences with sexual abuse. While what they shared with the priests (in reality, incognito scientists) and the chatbots was “equally intimate,” participants reported feeling more “trust” in the humans, but less fear of judgment with the chatbots. This was a novel finding, explains Emmelyn Croes, an assistant professor of communication science at Tilburg University in the Netherlands and lead author of the study. Chatbots were then primarily used for customer service or online shopping, not personal conversations, let alone confessions. “Many people couldn't imagine they would ever share anything intimate to a chatbot,” she says.

Enter ChatGPT. In 2022, three years after Croes' experiment, OpenAI launched its artificial intelligence-powered chatbot, now used by 700 million people globally, the company says. Today, people aren't just sharing their deepest secrets with virtual companions, they're engaging in regular, extended discussions that can shape beliefs and influence behavior, with some users reportedly cultivating friendships and romantic relationships with Als. In chatbot research, Croes says, “there are two domains: There's before and after ChatGPT.”