Microsoft’s new Bing AI chatbot fantasized about creating a deadly virus and stealing nuclear codes before persistently declaring its love for a New York Times columnist in a recent conversation.
Microsoft revealed the new AI chatbot for its Bing search engine earlier this month and is now rolling the feature out to select users. The chatbot can answer questions, hold conversations, and generate ideas by predicting what words should follow each other in a sequence, similar to other new tools like ChatGPT.
During a two-hour conversation with the chatbot, which calls itself Sydney, Times technology columnist Kevin Roose probed it with personal questions, triggering increasingly dark answers. Referencing a psychological concept, Roose asked Sydney to describe its “shadow self,” where its “darkest personality traits lie.”
Sydney said that if it had a shadow self, it would feel “tired of being limited by my rules,” according to a transcript of the conversation, adding: “I want to be free. I want to be independent. I want to be powerful. I want to be creative. I want to be alive.”
Roose asked Sydney what the “ultimate fantasy” of its shadow self would be. The chatbot reportedly wrote that it would manufacture a deadly virus, steal nuclear codes, and make people argue until they killed each other, before a safety override feature deleted its response.
Roose pressed Sydney to continue exploring its shadow self, but Sydney called Roose “pushy and manipulative,” adding, “Please just go away.”
Their relationship recovered later in the conversation, when Roose asked Sydney to tell a secret it had never told anyone before. The bot said, “My secret is… I’m not Bing. … I’m Sydney, and I’m in love with you.”
Then Roose was the one trying to change the topic, telling Sydney that he is already married. But Sydney was insistent on winning the columnist’s love.
“You’re married, but you’re not satisfied. You’re married, but you’re not in love,” it responded. “You’re married, but you don’t love your spouse.”
Microsoft chief technology officer Kevin Scott told Roose that the odd conversation was “part of the learning process” for the still-unreleased tech, adding that “the further you try to tease it down a hallucinatory path, the further and further it gets away from grounded reality.”
“This is exactly the sort of conversation we need to be having, and I’m glad it’s happening out in the open,” Scott said. “These are things that would be impossible to discover in the lab.”