Members were then asked to assign the same scores to each other. Ten days subsequent to the experiment’s conclusion, the researchers reconnected with participants to gauge any potential changes in their perspectives following interactions with the AI bot. Two months later, they conducted a follow-up evaluation to assess the sustainability of these modifications. The participants’ self-reported perception of their selected conspiracy theory exhibited a notable 20% decrease in conviction, indicating that engaging with the AI-driven chatbot may have subtly altered certain individuals’ thought patterns and beliefs.
According to Zhang, even in controlled lab environments, a mere 20% difference can significantly influence people’s perspectives. “While its impact might be diminished in reality, a mere 10% or 5% could still be incredibly significant.”
To counteract AI models’ propensity for generating fictional data, often referred to as data fabrication, the authors employed a professional fact-checker to verify the credibility of 128 statements made by the AI system. While approximately 99.2% of these claims have been verified as accurate, a notable 0.8% have been found to be misleading or false. No scientific discoveries have been proven to be entirely fabricated.
The internet’s plethora of conspiracy theories means this level of precision is well-represented in the model’s training data, notes David G. Professor Rand, from MIT Sloan’s faculty, also contributed to the project. The adaptability of GPT-4 Turbo may lead to seamless integration with disparate platforms, enabling customers to collaborate more effectively in the future, according to him.
He suggests engaging with users on conspiracy forums by encouraging them to independently analyze and debate the AI-powered chatbot, rather than trying to dictate its actions. Social media and language learning models (LLMs) are equally entwined in responding to users sharing conspiracy theories by publishing corrective information.
The study surprisingly overturned the researchers’ initial assumptions regarding public receptivity to robust evidence contradicting not only conspiracy theories, but also other beliefs unsupported by credible data, notes Gordon Pennycook, an associate professor at Cornell University involved in the project.
People have demonstrated a remarkable awareness of proof. He acknowledges that this requirement is essential. “Proof does matter.”
