Telling the Difference

Listen to Today's Edition:

Scientists recently discovered that artificial intelligence can easily deceive netizens and is getting better at lying, El País reported.

The emergence of AI models has been met with a mix of awe and concern about the impact they will have on humanity.

For their study, a research team asked nearly 700 people to read 220 tweets written by other humans and by the AI model GTP-3 – a precursor of the current ChatGPT.

Participants had to determine which tweets were true or false, as well as figure out which of them were written by a human or a machine.

GTP-3 won on both counts: It lied better than humans and its tweets did not appear to humans as if they had been written by a machine.

The researchers explained that the findings provide further evidence that people cannot distinguish between what is written by a human or by a machine. They proposed a “theory of resignation” to suggest why people are – and allow themselves to be – easily deceived.

“Our resignation theory applies to people’s self-confidence in identifying synthetic text,” explained co-author Giovanni Spitale. “The theory says that critical exposure to synthetic text reduces people’s ability to distinguish the synthetic from the organic.”

Spitale and his colleagues cautioned that if the theory proves true, Internet users will find it very difficult to recognize patterns in machine-produced texts.

However, the authors noted that the AI model did not respond well to misinformation requests and would occasionally disobey them, especially regarding topics with more evidence, such as flat-Earthism.

They suggested that the training databases for these models should adhere to principles of accuracy and transparency, with verified information and open origins that can be independently examined.