Listen to the audio version of the article
The news is set to fuel the debate on artificial intelligence and its future applications: ChatGPT-4 displays behavioral and personality traits that are statistically indistinguishable from a human in aspects such as cooperation, trust, reciprocity, altruism, contempt, fairness, strategic thinking and risk aversion. This is the conclusion of a joint study by the University of Michigan and Stanford detailed in an article published in the journal Pnas (Proceedings of the National Academy of Sciences). Understanding how ‘Ai behaves is the motivation of the researchers, who used the Turing tests – which test the ability of a machine to show responses and intelligence similar to those of humans – to evaluate the personality and behavior of GPTs by applying methodologies typical of psychology and behavioral economics.
The study in detail
The research tested ChatGPT versions 3 and 4 through standardized personality tests and challenged chatbots to strategize in various behavioral games that simulate real-world economic and moral decisions, such as managing financial investments. The bots’ personality traits were analyzed using the OCEAN Big-5 test, which measures five key dimensions of human behavior: extraversion, neuroticism, agreeableness, open-mindedness and awareness. The Chatbots were then put to the test in six distinct games designed to reveal a range of behavioral traits such as mischief, trust, risk-taking, altruism, fairness, cooperation and strategic reasoning. The bots’ responses were compared with those of over 100,000 individuals from 50 different countries, leading to the passing of what was essentially an advanced Turing test for Ai. Chatbots have demonstrated a propensity for generosity and fairness that has gone beyond that of the average human player. “The AI’s strategic decisions in The Prisoner’s Dilemma and other games reflected an understanding of cooperation and trust in maximizing goals, often opting for cooperative strategies that deviate from the human norm,” the researchers said, finding a notable difference between the two Ai versions of OpenAI. ChatGPT-4’s performance was remarkably human-like, often indistinguishable from or even superior to human behavior, suggesting its potential to pass the Turing Test in certain contexts. In contrast, ChatGPT-3’s responses often were perceived as non-human, highlighting differences in behavioral tendencies between the two versions of the AI.
The future development of the study
According to the researchers, in the near future it will be necessary to test more artificial intelligence models on more behavioral tests to thus compare their personalities and distinguishable traits. In particular, the researchers hope that the study opens up a new field, the “behavioral science of artificial intelligence”, in which researchers from different disciplines can work together to study the behaviors of artificial intelligence and in particular their relationships with human beings. humans and the impact on future society. “Overall, knowing that ChatGPT is more altruistic and cooperative than average humans could increase our confidence in using it for negotiation, dispute resolution, or assistance needs,” the researchers conclude, adding that research like this they help people understand when and how we can rely on artificial intelligence.