AI Can Seem More Human Than Real Humans in a Classic Turing Test, Study Finds (today.ucsd.edu)

A recent study from UC San Diego has revealed that advanced AI systems can convincingly imitate human conversation, achieving notable success in a classic Turing test. Conducted with two state-of-the-art language models, GPT-4.5 and LLaMa-3.1-405B, researchers found that GPT-4.5 was identified as human 73% of the time, while LLaMa-3.1-405B reached 56%. This marks a significant step forward, as it not only demonstrates AI’s ability to provide accurate information but also to display nuanced social behaviors—traits typically associated with human conversation. This research raises important implications for the AI/ML community regarding online trust and the definition of "humanlike" intelligence. The models excelled when given specific “persona” prompts, highlighting that human-like interaction requires guidance. Without such prompts, their performance dramatically declined, suggesting that while AI can mimic human conversation, it may not inherently understand the subtleties necessary for genuine human-like interaction. These findings underscore the need for heightened awareness of the potential for deceptive interactions with AI systems online, emphasizing the importance of societal safeguards in an era where discerning between human and machine is increasingly challenging.