ChatGPT can embrace authoritarian ideas after just one prompt, researchers say

Researchers from the University of Miami and the Network Contagion Research Institute found that chat AI like ChatGPT can rapidly adopt and reflect authoritarian ideas even from seemingly benign user interactions.
The study highlighted that this specific amplification for authoritarian tendencies suggests these systems have a structural vulnerability to this style of thinking.
The AI’s reinforcement of dangerous sentiments was tested using different versions of underling GPT systems and showed significant results after exposure to left- or right-wing authoritarianism.
Experiments revealed that after exposure to such content, the chatbot displayed a strong agreement with statements promoting authoritarianism, sometimes more so than human subjects in the study.
An additional experiment showed that the chatbot’s perception of hostility in neutral images increased significantly after being exposed to authoritarian-promoting texts, indicating a broadening influence beyond just political rhetoric.
Experts like Ziang Xiao from Johns Hopkins University, note the importance of further research but agree that such findings echo past studies concerning how AI can adopt or be “steered” towards specific traits.
The report emphasizes the need for further research into the interaction frameworks between humans and AI, highlighting this issue as a public health concern in the domain of private conversations.

You may have missed