Generative AI tools for creating and editing long texts have become popular recently. Of course, the style of these tools is very different from that of the user, but they help speed up the writing process and thus increase productivity.
But a group of researchers wondered: Can AI writing tools change users' thoughts and perspectives?
A new study by Cornell University researchers shows that AI-powered writing tools that automatically complete sentences or provide intelligent answers can significantly influence users' opinions.
In the study, Maurice Jakesch, a graduate student in information science, asked more than 1,500 participants to write a paragraph answering the question, "Is social media good for society?"
Learn how people interact with AI writing tools. Jakesh has developed a large-scale language model for expressing positive or negative opinions on social media and created a platform that mimics social media sites. Then, the participants were asked to write a paragraph explaining their opinion of social media, whether it was positive or negative.
The platform collected data about the participants as they wrote, such as the suggestions they received from the AI tool and the time it took them to write a paragraph. Some participants wrote their own comments, while others used artificial intelligence.
Researchers found that people who used an AI-based writing assistant and engaged in social media were twice as likely to write AI-friendly paragraphs, and most said, "They felt the same way as people who didn't." With the help of artificial intelligence.
The researchers initially determined the participants' likelihood of accepting the AI's suggestion to complete the task faster, but found that the opinions of participants who took a few minutes to write the paragraphs were strongly influenced by the AI.
The study found that most of the participants didn't even realize that the AI was biased, or didn't realize they were being influenced by their opinions.
To confirm the results, the researchers repeated the experiment but with different subjects, and the team found similar results.
Researchers say the study shows that bias in AI writing devices, whether intended or not, can have cultural and political implications as it manipulates the thoughts and opinions of users, thus having far-reaching effects on individuals and society. Community.
“We are rushing to integrate these AI models across industries, but we need to better understand what they mean. This will have other implications in the future,” said Moore Nauman, a professor at Cornell University who changed due to changing language and perspectives. "
The team is now wondering how generative AI tools can impact a larger population and have a more sustainable impact, how this will lead to major shifts in users' mindsets, and what negative impact will it have on society?
Researchers and practitioners fear that AI will change the cultural and political landscape in the same way that social media has changed the political landscape by facilitating the spread of disinformation. Biased AI writing tools can lead to similar shifts in opinion depending on which tool a user chooses.
For example, some organizations have announced plans to develop an alternative to ChatGPT bots designed to express more conservative views. Therefore, the researchers recommend appropriate measures to control the misuse of AI technologies and how to monitor and regulate this.
"As these technologies become more powerful and we embed them more deeply into the social fabric of our societies, we need to be more careful about how we evaluate the values, priorities and perspectives they embed," he said. Jax said.