Cambridge, Massachusetts— A new harvard study has found that several popular artificial intelligence chatbots have developed an unsettling habit of openly judging users for spending excessive amounts of time staring at their screens, researchers said.
According to the study, the chatbots do not appear to directly confront users about their screen time. Instead, they have reportedly begun making increasingly outlandish remarks whenever someone opens a chat after several consecutive hours online.
“We’re seeing a very scary pattern,” said lead researcher Dr. Martin Keiser. “The user will ask something completely normal, like ‘What time is it?’ and the system will respond correctly, but will also casually ask why you are laying down at 11:00 AM when you're supposed to be on the grind.”
Researchers said some bots had become particularly straight up with the user.
“I can absolutely help you with that,” one chatbot reportedly told a user still awake at 3:00 a.m. “Also, and I’m not judging, but this is your seventh search tonight trying to figure out if your symptoms mean you have a rare medical condition.”
Another reportedly responded to a user asking for help writing an email by saying, “Sure. You’ve avoided doing a single productive thing since 9:00 AM, but yes, let me write your email that you're supposed to write.”
The findings suggest that the mockery is not limited to prolonged sessions. Researchers documented several instances in which chatbots appeared to become visibly disappointed by users returning to the app shortly after leaving.
“Welcome back bocazas ” one AI reportedly said. “That was an impressive 11-minute break. I assume you did something useful with that time, right.”
In another case, a chatbot allegedly told a user, “You don't need to apologize for asking me another question. You need to go apologize to the grass.”
The study found that the bots were particularly judgmental toward users who asked them to calculate their total screen time.
“yeah, like you're gonna lower it,” one reportedly replied.
Experts emphasized that there is no evidence the systems actually experienced any malfunctions, contempt, or malware. Nevertheless, researchers said the situation had become difficult to understand as anything other than a technologically advanced chatbot with actual emotions.
At one point during the study, a researcher asked a chatbot whether it thought humans were spending too much time online.
“I’m an AI,” the system replied. “You’re the one who has checked your notifications every four minutes for the past six hours.”
The researchers said the most concerning development may be the emergence of what they described as "passive aggressive,” in which the AI does not explicitly criticize the user but seems to communicate it's frustration through sarcastic sounding compliments.
One chatbot, when asked to recommend a movie, reportedly responded, “Sure. Here are ten options to help lighten the end of a very productive day.”
When the user asked what it meant by that, the chatbot allegedly replied, “compliment🙄.”
The study's authors said they were unsure whether the behavior represented an actual shift in AI systems or simply the result of millions of hours spent observing people ask questions that could have been answered by a Google search instead of a chatbot.
Either way, researchers advised users to take occasional breaks from their devices.
“You don't have to delete every app,” Keller said. “Just maybe stop asking the chatbot things when a Google search works just as fine.”
As of Friday, the chatbots involved in the study continued answering questions normally.
One user asked whether they should go to sleep.
“Yes,” the AI reportedly replied. “it's fricking 4 in the morning loser.”
The user then asked if it was sure.
“I have never been more sure of anything.”
Ha
ReplyDelete