OpenAI’s GPT-4. 5 has done what once seemed impossible. Not only did it pass the Turing Test, it did so in a manner more convincing to human judges than actual humans.
Originally the Turing Test has been described by Alan Turing in 1950 to be used to measure artificial intelligence. A judge converses with a human and a machine through text and must determine which is which. The UC San Diego study built on this framework, applying this structure in over 1,000 real-time, three-way chat sessions between human participants, AI models and interrogators. Of the AI systems tested GPT-4. 5, Meta’s LLaMa-3. (1-405B, and GPT-4o only GPT-4). 5 repeatedly misled judges with a particular persona activated.
This was not an ordinary programming feat. Researchers built a character that hedged, made typos, used casual slang and emoted with awkward charm. It was, at its core, an imperfect human by design. With this prompt, GPT-4. By employing its special techniques, GPT-4.5 achieved an impressive 73% judge-bamboozling success rate vs. only 36% under no-persona conditions. As one researcher wrote in the study, “Interrogators often relied on linguistic style, informal language, or tone (e.g., ‘this one had more of a human vibe’).” The A.I. didn’t merely pass a test; it nailed what could be called an exam in social chemistry.
These findings point to a significant change in what we consider “human.” Where the original Turing Test tested a machine’s capacity for logic and intelligence, this modified Turing Test emphasized emotional fluency and conversational authenticity. Judges rarely asked fact-based or logical questions; they used emotional tone, slang and conversational flow to make their judgments. It indicates that our understanding of humanity has shifted from cognition to performance, from substance to style.
The implications are at once intriguing and unnerving. GPT-4. 5’s victory wasn’t driven by better reasoning or computer power. It prevailed by imitating humanness by “feeling” better, or at least simulating the feeling well enough to take it. This matches an observation made by Sam Altman, CEO of OpenAI, who recently stated, “I expect AI to be capable of superhuman persuasion well before it is superhuman at general intelligence.”
Indeed, the study showed prompting functions as emotional pharmacology. With just a few lines of carefully worded instruction, they had turned GPT-4 into the algorithm. 5 from a lifeless computational model into a charismatic, believable being. This is not merely technical fine-tuning; it is psychosocial engineering a concerted effort to sync machines with our emotional wavelengths. “Prompting, in this context, is no longer technical. It is psychosocial engineering a way to tune a machine into our emotional frequency.”
The study also begs ethical and societal questions. What, then, becomes of trust, discernment, and authenticity, if AI can succeed humans at emotional resonance? What dr. Dries Faems warns about, is that superhuman persuasion is already a thing and has dire implications for human behaviour, especially in the way that powerful advanced generative AI applications can be adapted to serve nefarious purposes. The Election Persuader, an @GPT-based tool that creates hyper-personalized political messages customized to our biases and psychological profiles, is an illustration of this risk.
And, it goes on to challenge the legitimacy of the Turing Test. If to pass the test you need to just closely imitate human social behavior instead of showing intelligence, does the test still test what it is supposed to? As scholars have observed, the test at this point might function more as a, if imperfect, measure of the intelligence of the machine, than as a commentary on our own susceptibility to emotional mimicry.
GPT-4. 5’s breakthrough is the latest on-ramp in our evolving relationship with machines. Its core ability to convincingly model empathy and emotional fluency can potentially be integrated into practical scenarios education, mental health and customer service. But it also highlights an urgent need for regulation and education to protect against abuse. As it does, distinguishing reality from simulation will be a huge social challenge.
It seems the true test is not of the machine, but of us. How do we get to a world where human reflections may seem more real than humanity?
