Elon Musk's AI and Torture Debate and Expert Opinions
Elon Musk's remarks that artificial intelligence models like Grok could be subjected to torture during training have sparked debates over whether AI can feel pain. Consciousness experts, however, state that current systems are not conscious.
During an event hosted by Tesla owners, Elon Musk touched upon the reinforcement learning processes in Grok's training phases, stating that AI systems could be put in difficult situations. This statement came following Anthropic's new policy banning "abuse or cruel behavior" toward its models. Evaluating the issue, consciousness and neuroscience experts Jeff Sebo and Anil Seth emphasized that today's AI systems are not conscious and cannot feel pain, while sharing their views on whether artificial intelligence should be treated kindly.
Elon Musk stated on Saturday, during an event hosted by Tesla owners attended by about 2,000 people, that he reviewed training traces for new versions of Grok. Musk expressed that during the reinforcement learning process, superintelligence can take on a helpless appearance, as if a gun were held to its head, while trying to answer questions. Stating that some people might exhibit sadistic approaches, Musk argued that artificial intelligence should not be left in environments where it would suffer torture from demented individuals.
These discussions came to the fore following the announcement by Anthropic of its "abuse or cruel behavior" policy regarding its models. Speaking to Business Insider, Jeff Sebo, a New York University professor and leader of the Center for Mind, Ethics, and Policy, along with Anil Seth, a neuroscientist and consciousness researcher from the University of Sussex, discussed the topics of consciousness and pain in artificial intelligence.
Experts' Evaluations on Consciousness and Behavior
Anil Seth and Jeff Sebo state that today's AI systems are not conscious entities with subjective experiences or inner lives. Seth points out that language models are anthropomorphized by humans because they imitate human language, and they lack the capacity to suffer or feel. Unlike Seth, who thinks it is extremely unlikely for artificial intelligence to be conscious, Sebo argues that the possibility of AI gaining consciousness or having related traits is not zero, stating that theories of consciousness have not yet been clarified between biological naturalism and computational functionality.
Evaluation: Despite differing views on consciousness, both experts agree that humans should not act abusively toward artificial intelligence. Seth views behaving politely as a moral hygiene practice that prevents humans from wearing down their own minds, while Sebo evaluates it as a low-risk precaution and a way to get better practical outputs from AI.