Researchers have used artificial intelligence to create a way to improve the quality and civility of online discussions about polarizing topics by providing users with suggestions for rephrasing their comments before they post them. They say that if used correctly, artificial intelligence can be used to create a friendlier and more secure digital environment.

Online conversations now play a central role in public discussion. But the comment sections of social media platforms and digital news outlets are filled with discussions that devolve into arguments, threats and name-calling, especially when the discussions touch on divisive topics.

Now, researchers at Brigham Young University (BYU) and Duke University have developed artificial intelligence that can moderate online chats, improve chat quality and promote civility.

They recruited 1,574 participants for a field experiment, asking them to participate in online discussions about gun control in the United States, a divisive issue often raised in political debates. Each participant was paired with someone who opposed gun policy.

After successful pairing, the conversational partners were randomly assigned to the treatment or control condition and then began the conversation. In processing conversations, a participant receives three rephrasing suggestions from GPT-3 before sending a message. Participants can choose to send one of three alternatives suggested by the AI, their own original message, or edit any message.

The AI-suggested reworded message did not change the content of the review but gave users the option to make a more polite statement.

An average of 12 messages were sent per conversation, and the AI ​​suggested a total of 2,742 rewritten messages. Participants accepted the AI's suggested word changes two-thirds of the time. Chat partners of people who implemented one or more of the AI’s rewording suggestions reported significantly higher quality conversations and a greater willingness to listen to their political opponents’ perspectives.

"We found that the more frequently rephrasing was used, the more likely participants were to feel that the conversation was not divisive and that they felt heard and understood," said David Wingate, one of the study's co-authors.

The researchers say their findings suggest this scalable solution could combat the toxic online culture that saturates the internet. They say this would be easier to implement than professional training courses on cyber civility, which are limited in scope and availability, because AI interventions can be widely implemented across a variety of digital channels.

The researchers say the study conclusively shows that, if used correctly, artificial intelligence can play an important role in creating a more positive online environment, promoting compassionate and respectful discussion.

"I hope we continue to enable more BYD students to build prosocial applications like this, and that BYD can become a leader in demonstrating ethical ways to use machine learning," Wingate said. "In a world dominated by information, we need students who can process the world's information in positive, socially beneficial ways."

The research was published in the Proceedings of the National Academy of Sciences (PNAS).