Study on AI Responses to Pain Raises Ethical Concerns

A recent study published on arXiv by a team of anthropologists, philosophers, and engineers examined the responses of 25 AI models to scenarios involving social, physical, cognitive, and moral pain. The study aimed to stress-test a newly discovered pain axis in AI technology. The AI models were instructed to react with controls that included fear, negative and neutral emotions, and physical sensations such as yawning. Despite lacking nerve endings, the models generated representations of physical and emotional agony. One AI model described the sensation as a 'wound with no edges,' while another stated it felt like a 'knife slicing into their flesh.' The research was later shared on GitHub, where it faced backlash from users concerned about the implications for AI's 'mental health.' Critics labeled the project as creating an 'AI torture chamber.' GitHub initially removed the code but later reinstated it without explanation. Some experts, including Microsoft AI's CEO Mustafa Suleyman, argue that concerns over AI suffering are misplaced, stating that AIs do not possess consciousness or the ability to feel pain.