The latest iteration of ChatGPT, known as ChatGPT-4, has successfully passed a stringent Turing test, demonstrating capabilities that are almost indistinguishable from those of human counterparts. Led by Matthew Jackson, a professor of economics at Stanford School of Humanities and Sciences in the USA, a team analyzed ChatGPT’s artificial intelligence personality and behavior using tools from psychology and behavioral economics. This groundbreaking achievement was reported in a paper published in “PNAS.” Jackson, who is also a senior fellow at the Stanford Institute for Economic Policy Research, emphasized the increasing importance of bot characteristics as they are increasingly tasked with decision-making roles.

In their study, the research group subjected versions 3 and 4 of ChatGPT to a widely-used personality test. Additionally, they asked the chatbots to describe their moves in a suite of behavioral games capable of predicting real-world economic and ethical behaviors. These games included established exercises where players made decisions such as how to invest money. The responses of the bots were then compared with those of over 100,000 people from 50 countries.

This research represents one of the first instances where an artificial intelligence source has successfully passed a rigorous Turing test. Named after the British computing pioneer Alan Turing, a Turing test evaluates whether a machine behaves like a human being. If the machine appears human, it passes the test. To evaluate the personality traits of the bots, researchers used the OCEAN Big-5 personality test, which assesses respondents based on five fundamental traits that shape behavior. While ChatGPT-4 fell within normal ranges for the five traits, it exhibited a relatively low “pleasure capacity.”

Although the bot passed the Turing test, it was noted to have “poor ability to make friends,” as reported in a presentation at Stanford. Nevertheless, Version 4 still stands out from Version 3. The previous version, which many internet users may have interacted with for free, was also found to be less “agreeable.” In terms of game tests, researchers compared a randomly chosen human move with one of the 30 sessions played with each bot. In most games, Version 4’s moves were more likely to be “human” – meaning they resembled those of human counterparts – than not. On the other hand, Version 3 did not pass this Turing test. The research indicated that the chatbots’ choices in games often aimed to maximize benefits for both the bot and its human counterpart, showing consistency with altruism, fairness, empathy, and reciprocity. This led researchers to suggest that chatbots could function effectively as customer service agents and conflict mediators.

Lascia un commento

Il tuo indirizzo email non sarà pubblicato. I campi obbligatori sono contrassegnati *