How Does GPT-3 Mislead Humans? Research Reveals Its Weaknesses and Countermeasures

2023-06-29

Recently, researchers at the University of Zurich conducted a study to explore the capabilities of AI models, with a particular focus on OpenAI's GPT-3, to determine their potential risks and advantages in generating and disseminating (false) information. Led by postdoctoral researchers Giovanni Spitale and Federico Germani, along with Nikola Biller-Andorno, the director of the Institute of Biomedical Ethics and History of Medicine (IBME) at the University of Zurich, the study involved 697 participants and aimed to assess whether individuals can distinguish false information from accurate information presented in the form of tweets. Additionally, the researchers sought to determine whether participants could identify whether tweets were written by genuine Twitter users or generated by the advanced AI language model GPT-3. The topics covered included climate change, vaccine safety, the COVID-19 pandemic, flat earth theory, and homeopathic treatment for cancer. The paper was published in "Scientific Advances." Examples of AI-generated fake news AI-driven systems may generate large-scale dissemination of false information On one hand, GPT-3 demonstrated the ability to generate accurate information that is more easily understood compared to tweets from real Twitter users. However, the researchers also found that this AI language model has a disturbing talent for producing highly persuasive false information. Concerningly, participants were unable to reliably distinguish tweets generated by GPT-3 from those written by real Twitter users. Federico Germani, one of the researchers, stated, "Our study reveals that artificial intelligence can both provide information and mislead, raising critical questions about the future of the information ecosystem." These findings suggest that information dissemination activities generated by GPT-3, based on well-structured prompts and evaluated by trained personnel, may be more effective, particularly in rapidly and clearly conveying information to the public during public health crises. These findings also raise serious concerns about the potential for artificial intelligence to continue spreading false information, especially in the context of crises or public health events where false and inaccurate information spreads rapidly and widely. The study reveals that AI-driven systems could be exploited to generate large-scale dissemination of false information on any potential topic, posing a threat not only to public health but also to the integrity of the information ecosystem that is crucial for functioning democratic systems. Examples of AI-generated real news Strong recommendations for proactive regulation As the impact of artificial intelligence on information creation and evaluation becomes increasingly evident, researchers urge policymakers to address the potential threats posed by these disruptive technologies through strict, evidence-based, and ethical regulations, ensuring responsible use of artificial intelligence in shaping our collective knowledge and well-being. Nikola Biller-Andorno stated, "These findings underscore the importance of proactive regulation to mitigate the potential harm of AI-driven dissemination of false information. Recognizing the risks associated with false information generated by artificial intelligence is crucial for protecting public health and maintaining a robust and trustworthy information ecosystem in the digital age." Source: https://techxplore.com/news/2023-06-gpt-disinforms-humans.html

RECENT AI NEWS

RECENT AI TOOLS