In what comes as a major breakthrough in the world of AI, OpenAI’s GPT 4.5 model has passed the famed Turing’s Test, that was devised by Alan Turing to help humans to distinguish between a fellow human and machine. Simply put, Turing’s Test says that machines can be deemed to be as intelligent and indecipherable from humans, if a machine is put under interrogation by a human who also happens to interrogate another human, and the interrogator fails to determine if the respondent is a machine or otherwise, the machine can be deemed to be intelligent enough to pass as a human.
In a preprint study that still awaits peer review, researchers have claimed that in a three-party version of a Turing’s Test, humans were made to chat with another human and OpenAI’s GPT 4.5 model at the same time. And as the news goes, the AI model was deemed to be a human 73% of the time, when it was instructed to behave like a particular persona. Guess what, Turing had suggested the benchmark to be 50%. As a model has now surpassed the threshold set forth by Alan Turing, the day has arrived when humans have officially failed to distinguish machine from humans.
The buck does not stop there. Even Meta’s LLaMa, when instructed with a persona-related prompt, performed as good as GPT-4.5. What’s surprising is that GPT 4.5 was deemed to be significantly more human compared to real humans during the experiment. The only caveat was that when the AI models under study were not given any persona-specific prompt, they failed to compete against humans in their responses. For example, without persona prompting, GPT 4.5 achieved an overall win rate of just 36%, while GPT-4o achieved mere 21%.
What does it imply then? Quite straightforward, the experiment shows that when properly prompted, some of the latest AI models can behave as good as humans in their responses to real-world interactions. Despite the recent success of the AI models, some experts suggest that Turing Test in its current state may not be the best assessment of AI’s capabilities to conduct itself like humans. It is because our perception of AI is constantly evolving and with time, humans may grow more adept at distinguishing AI from humans, as their perceptions develop further.
What we cannot deny is that AI has arrived in the mainstream interaction space, and can very well substitute humans in the times to come, where human-like interactions matter for fostering trust and facilitating collaboration.
