In April, a landmark study called “Human or Not?” was launched to determine whether people could identify if they were talking to another person or an AI chatbot.

More than 2 million volunteers and 15 million conversations later, 32 percent of participants picked incorrectly.

There was also little difference in the results based on age categories. Older and younger adults both struggled at a similar level to discern who—or what—was on the other end of the conversation.