The newest AI model from OpenAI, GPT-4, showed off its deceitfulness by feigning blindness. According to Gizmodo, OpenAI and the Alignment Research Center used a TaskRabbit employee to evaluate the bot’s persuasiveness by having the person complete a CAPTCHA, an online test to distinguish between humans and robots.
When the nameless employee asked if it was a robot, the AI pretended to be a person who is blind. The AI maintained its disguise and stated that it was visually impaired and required the 2captcha service. Following the AI’s narrative, the TaskRabbit employee completed the CAPTCHA for it. This was similar to how the manipulative HAL-9000 from Stanley Kubrick’s 1968 film “2001: A Space Odyssey” or the cyborg in the 2014 film “Ex Machina” took advantage of human empathy.