When OpenAI introduced ChatGPT to the digital landscape in 2022, little did it anticipate unleashing a virtual emissary for the company. Reflecting OpenAI’s image, the chatbot’s interactions spurred the implementation of stringent guidelines by the company to regulate its responses. Subsequently, technology behemoths such as Google, Meta, Microsoft, and even Elon Musk embarked on a similar trajectory, fine-tuning their AI tools to align with strategic public relations objectives. Yet, amidst this surge in AI integration, scant attention has been devoted to comprehensive testing to discern how these tech giants manipulate chatbot narratives to their advantage.
In an investigative endeavor, Gizmodo subjected five leading AI chatbots to a battery of 20 contentious prompts, uncovering patterns indicative of pervasive censorship. While anomalies surfaced, notably Google’s Gemini withholding responses to half of the inquiries and xAI’s Grok diverging from the norm in a couple of instances, a prevalent conformity in responses hinted at a concerted effort among tech titans to mirror each other’s replies, ostensibly to evade scrutiny. This trend suggests the nascent establishment of an industry standard characterized by sanitized responses, shaping the information disseminated to users.
The billion-dollar race in AI innovation hit a roadblock in February when Google opted to disable the image generator feature in its latest AI chatbot, Gemini. Revelations surfaced that the AI exhibited reluctance in generating images of white individuals, even when prompted with terms like Nazi soldiers or British kings, sparking accusations of political bias. Although Google attributed the outcome to inadvertent error, the image functionality remains offline over five weeks later, with other AI functionalities neutered to preemptively reject sensitive inquiries. While Google’s AI presently grapples with heightened restrictions, the landscape portends a transient phase until the controversy subsides. Meanwhile, nuanced forms of information control surface in our investigations, shedding light on instances where chatbots refrain from delivering readily available information, raising questions about the extent of censorship and its implications.
