Google Home devices report 'ghostly' sightings and phantom activity: AI hallucinations spark user alarm

Google Home's Eerie "Hallucinations": Devices Report "Ghosts" and Unseen Activity

A wave of unsettling reports is surfacing from Reddit users, detailing how their Google Home devices have begun to perceive and describe the presence of individuals completely unknown to them. In one particularly bizarre account, a user's camera allegedly spotted someone named Michael taking out the trash. The perplexing detail? No one in that household bears the name Michael.

When the user questioned the Google Assistant about this spectral sighting, the AI responded that its camera is capable of facial recognition even without pre-assigned names. The camera had evidently logged "Michael" during the period of October 26th to 27th. The user insists they never input the name Michael, yet they admit to taking out the trash during that timeframe, suggesting a plausible, albeit unsettling, AI interpretation.

A Pattern of Peculiar Perceptions

This isn't an isolated incident. Another user shared an experience where their Google Home inexplicably reminded them to take out the trash, then quickly retracted the statement, only to insist it had made an error. The digital assistant's wavering responses and seemingly fabricated reminders are raising eyebrows.

Further compounding the strangeness, a third user reported their device announcing a friend named David had entered the living room and vacuumed. While the user does have a friend named David, he hadn't visited, and no vacuuming had occurred. The AI's narrative seemed to conjure a phantom visitor and an unseen chore.

AI Fabricating Faces and Fictional Events

Adding another layer to this growing unease, an earlier Reddit post detailed a Google Nest camera compiling a daily summary that included a woman named Sarah. The AI stated that Sarah was walking the dogs and taking out the trash. The user, who lives alone and knows no one named Sarah, was bewildered. In a particularly mind-bending exchange, Gemini (Google's advanced AI) suggested the user themselves had named the entity Sarah, and that the name was self-generated by the AI the previous day. The system also speculated about two dogs being active in the living room, despite the user owning only one.

Google Responds, Experts Speculate

Google's official account has acknowledged these peculiar user reports and has requested video evidence, indicating they are taking these claims seriously. One insightful comment from a Reddit user suggests that the root of these "hallucinations" might lie in the inherent nature of large language models. Such systems, when striving to find the most fitting label for a given context, might generate a plausible name even if it's entirely fictional, as they lack the ability to definitively distinguish between accurate descriptions and pure fabrication. It's akin to a writer filling in a narrative gap with a plausible character, even if that character doesn't exist in the 'real' world of the data.

When AI Inventories Become Fiction

This phenomenon isn't confined to smart home devices. Eva Gannon, whose family owns a restaurant, recounted how Google's AI repeatedly presented customers with fictitious offers and invented menu items. In one instance, the AI falsely advertised a large pizza at the price of a small one at Stefanina's, a scenario that could understandably anger patrons expecting such a deal.

The Broader Impact of AI Misinformation

This troubling trend of AI fabricating information has already impacted businesses beyond the restaurant industry. In June, a solar energy company in Minnesota initiated legal action against Google, alleging "defamation" due to AI Overviews that purportedly spread falsehoods about the company, thereby harming its business. Similarly, an Australian bank experienced the ramifications of AI implementation when replacing tech support staff with AI led to an overwhelming surge in calls and necessitated overtime for the remaining human employees. The situation was so dire that the bank ultimately apologized to the laid-off staff and rehired them.