People are more likely to believe AI-generated tweets than ones written by humans study finds

Picture of phone that displays OpenAI logo.
People may be more susceptible to misinformation if it's written by AI, a new study has found.
  • People may be more susceptible to misinformation if it's written by AI, a study has found.
  • The study found respondents couldn't tell the difference between tweets created by GPT-3 and humans.
  • Respondents also generally found the AI-generated tweets more convincing.

People may be more susceptible to misinformation if it's written by AI, a study has found.

Researchers surveyed 697 people for an analysis published in the Science Advances journal to see if they could tell the difference between tweets generated by OpenAI's large language model GPT-3 and others written by humans.

Not only could the majority of participants not tell the real tweets from the ones generated by AI, but they generally found the AI's tweets more convincing.

The researchers chose topics commonly associated with misinformation such as vaccines, COVID-19, evolution, and 5G technology. They then asked respondents to identify if the tweets were true or false, which participants had an easier time doing with the AI-generated ones.

It was easier for people to recognize the accuracy of tweets containing correct information when they were generated by GPT-3 and they were more deceived by the AI tweets containing disinformation, the study found. The researchers called GPT-3 a "double-edge sword" that "can produce accurate information that is easier to understand, but it can also produce more compelling disinformation."

Experts have been vocal about concerns that the widespread use of AI could increase misinformation. Tools powered by large language models such as ChatGPT have been known to present incorrect information in a sophisticated and believable way. AI tools have also been known to completely invent facts. 

For example, OpenAI has been hit with a defamation lawsuit from a radio host who alleged that the AI chatbot ChatGPT created a fake legal complaint about him. 

Most of the respondents to the Science Advances analysis held a bachelor's degree, primarily in the social sciences and humanities, natural sciences, or medical sciences. And most were from the UK, Australia, Canada, the US, and Ireland.

Read the original article on Business Insider


from Business Insider https://ift.tt/WVrQFk1

No comments

Powered by Blogger.