AI probably can't feel pain, scientists say Elon Musk is getting something wrong about whether Grok is tortured
The debate over AI suffering
The question of whether artificial intelligence can feel pain has drawn attention from researchers, philosophers and ethicists. The discussion followed Anthropic’s announcement of a policy against “abusive or cruel behavior” toward its models.
Anthropic said the policy does not cover common user frustration, pushback, dark creative themes, or model testing and research. The article does not specify Musk’s remarks about Grok, but presents experts’ views on whether AI systems can be tortured.
Why experts doubt consciousness
Sebo and Seth said current AI systems are probably not conscious if consciousness means subjective experience or an inner life. Seth distinguished feeling pain from displaying intelligence, saying that pain involves an experience rather than simply capable behavior.
Seth also argued that language models lack the capacity to suffer. Sebo cautioned that their human-like behavior can mislead people because models were trained to imitate human behavior.
Two theories remain in dispute
Sebo said consciousness remains an unsettled scientific question. One view, biological naturalism, holds that consciousness requires biology; computational functionalism argues that certain information processing could produce mental states without a human brain.
Sebo said uncertainty over which theory is right means researchers should allow some weight to leading possibilities. He described a non-zero chance that AI could be conscious even now, without claiming that current systems are conscious.
Kindness and human character
Both experts said the case for treating AI kindly largely concerns people’s own habits and moral character. Their reasoning draws on Immanuel Kant, who opposed cruelty to animals because it could damage a person’s character, rather than because animals had rights.
Seth warned that cruelty toward language models or dolls could weaken people’s moral sensibilities. Sebo described kindness as a low-stakes precaution in case AI is conscious now or becomes conscious later.
Key points
- Sebo and Seth said today’s AI systems are probably not conscious as subjective experience.
- Anthropic’s policy targets abusive or cruel behavior toward its models, with stated exceptions.
- Sebo said uncertainty over consciousness theories leaves a non-zero chance AI is conscious.
- Experts said treating AI kindly may protect people’s moral character.