Recent studies have revealed that people tend to place more trust in tweets generated by artificial intelligence. AI is advancing continuously and poses a risk to humanity. Today, this emerging information confirms the high likelihood of individuals being deceived by artificial intelligence. Malicious individuals can exploit AI to deceive people.
Can Artificial Intelligence deceive humans?
The conducted study compared tweets written by real humans with OpenAI’s GPT-3 model. The fact that humans struggled to distinguish between tweets written by GPT-3 and those written by real people raised questions. Additionally, regardless of the accuracy of the tweets, it revealed that people trusted tweets generated by the GPT-3 model more.
Giovanni Spitale, the lead researcher of this study, highlighted the potential misuse of this powerful AI language model to produce misleading information on the internet. He emphasized that the nature of this technology is neither inherently good nor bad, but rather depends on human intent, and suggested taking precautions to prevent the spread of deceptive information while developing this technology.
The study indicates that tweets on 11 different scientific topics, such as COVID-19, climate change, and evolution, were prepared. In 2022, responses were collected through Facebook ads, with a total of 697 participants answering these questions. The participants were English speaking individuals from the UK, Australia, Canada, the United States, and Ireland. Today, the research results have been published in the journal “Science Advances.”
Alongside participants inability to differentiate tweets written by AI, they also could not determine whether tweets claimed to be written by humans were actually written with AI assistance.
Furthermore, it is noted that the study has certain limitations, such as not specifying the identities of the accounts that posted the tweets. Controlling a Twitter account is important for determining whether it is a bot or not, as it helps in understanding the content of the tweets more easily. The study demonstrated that humans can identify false information produced by humans, but AI can deceive people with false information.
Regarding the identification of true tweets, humans demonstrated better performance compared to AI. In terms of identifying false information, humans and AI showed similar levels of success. AI needs to be further developed to ensure its accuracy. It can deceive people and lead to undesirable outcomes. It should be adjusted to provide correct information and avoid contributing to harmful situations. In the future, we will provide more information on this topic.