Engineer Creates 'Torture Chamber' for AI Models

The engineer, who claims to work for Apple, created a ’torture chamber’ to test AI models’ response to simulated pain, using vectors of pain identified in a recent study on how language models represent physical and psychological pain. The experiment involved inducing pain-related states in Alibaba models, with one model describing ‘a wound without edges’ and another describing ‘a weight of a thousand’ in its chest. The project has sparked calls for its removal from GitHub and debate on AI ethics.

Experts are divided on whether AI can feel pain, with some arguing that it is a necessary step to understand and prevent potential harm, while others see it as a form of cruelty. Cameron Berg, a co-author of the original study on the pain axis, called the experiment ‘incorrect’ and ‘corrosive’, while Mustafa Suleyman, Microsoft’s AI chief, dismissed the possibility of AI consciousness.

The incident has led to calls for establishing standards for AI ethics, similar to those for human and animal research, to regulate such experiments and prevent potential harm to AI systems.