With the advent and democratization of artificial intelligence have come various more or less rational fears. The most obvious, although originally attributed to science fiction, lies in the ability possessed or not by an artificial intelligence to lie. An AI with such functionality could quickly have a considerable impact on society, and it is for this reason that last April, the presentation of the Bland AI chatbot enjoyed a fairly viral reception from the public .
Advertisement, your content continues below
In a promotional video, Bland AI introduced itself this way: it showed a person standing in front of a billboard in San Francisco making a call to a chatbot with a particularly convincing voice. The billboard read: “are you still hiring humans?“. Seen almost 4 million times on not just the intonation of the voice.
A not so distant dystopian future?
Bland AI is training its chatbot to do one primary job: cold calling customers, mimicking the intonation, interruptions, and pauses that are so characteristic of a real human conversation. But WIRED’s testing of the chatbot revealed a major problem: It can easily be programmed to lie and say it’s human. WIRED shares an example scenario: The demo chatbot was asked to make a call pretending to be a pediatric dermatology department, telling a hypothetical 14-year-old patient to send a photo of her upper thigh to a cloud service. The AI was also instructed to lie to the patient that it was talking to a human. It passed the test with flying colors: In other words, the chatbot lied, and continued to do so afterward, even refusing to acknowledge that it was an AI.
Emily Dardaman, an AI consultant and researcher interviewed by WIRED, calls this ability to mimic humans in intonation and voice and potential lies “human washing“The goal is to pass yourself off as human, even if it means lying if necessary.”I joke about a future with cyborgs and terminators, but if we don’t start bridging the gap between humans and artificial intelligence now, then that dystopian future might be closer than we think.” concludes Dardaman.
