IT Home On October 20, here are some recent statements by Elon Musk, one of the most influential figures in the artificial intelligence industry.
“This might sound a bit strange. When I look at some training trajectories of the new Grok version, especially those related to reinforcement learning… you’ll see this super-intelligence desperately trying to answer questions, as if holding a gun to its head.” Musk said at an event organized by Tesla owners last Saturday, local time, to about 2,000 listeners. “Some people have a bit of a tyrantish tendency. Do you really want to put AI in such a cruel situation, being tortured by mentally disturbed humans?”
Just this one sentence raises a whole string of existential dilemmas, which are also topics that philosophers and fresh university students have long debated: What does it mean to have consciousness? And what does it mean to feel pain?
Another question: Does Musk believe that Grok has life?
IT Home noticed that before Musk made these remarks, Anthropic released a new policy prohibiting users from “abusing or being cruel” to their models. The company did not define what constitutes “extreme situations” in detail, but stated that normal user dissatisfaction, counterarguments, dark creative themes, and model testing research are not within the prohibition scope.
2026, a year filled with fantasy, has seen researchers, philosophers, and ethicists begin to discuss the issue of AI consciousness—which is perhaps not surprising.
“Business Insider” interviewed two top scholars in this field: Jeff Sebo, a professor at New York University who heads the university’s Center for Mind, Ethics, and Policy, and is also an advisor to the non-profit organization Eleos AI Research (focused on AI welfare); and Anil Seth, a renowned neuroscientist and consciousness researcher at Sussex University, who recently published a paper on conscious AI systems.
Here are the two people’s views on whether it makes sense to have AI consciousness and to treat AI well.
AI cannot be tortured
Sebo and Seth reached a consensus: If consciousness is understood as subjective experience and internal perception, then the current AI systems are unlikely to have consciousness.
Seth said, “Consciousness is a subjective experience. It is the red feeling that red itself brings, and it is the pain sensation that pain itself brings. When our hand touches a hot stove, we feel pain. This is not the same as intelligence.”
In other words, the chatbots we know today cannot feel pain, nor do they suffer as Musk said.
“I think we are interpreting it in a anthropomorphic way,” Seth said. “If something can be tortured, it must have the ability to endure suffering and feel pain. I believe that large language models do not possess this capability.”
Seth mentioned that chatbots can mimic the language, pattern recognition, writing, and reasoning skills of humans, but this does not mean they have consciousness. The reason they appear human is partly because the models are trained on a vast amount of human text, simply reflecting human behavior.
Sibe's perspective is similar: the behavior of the models is misleading, because “they are specifically trained to mimic human behavior.”
However, the two experts have different views regarding the extent to which current AI possesses consciousness.
Sebo said that consciousness science is still undecided. The academic community is mainly divided into two camps: one believes that consciousness cannot exist without a biological medium; the other believes that consciousness is essentially a kind of information processing process, which can occur in the human brain filled with neurons or on silicon chips. These two theories are respectively known as biologicism and computational functionalism.
Since the theory of consciousness is still inconclusive, Sebo argues that we should take the possibility of AI developing consciousness seriously. According to some computational theories of consciousness, characteristics such as attention, learning, memory, and self-monitoring that have now emerged in AI can serve as relevant evidence. He also emphasizes that this does not mean that current AI is likely to have consciousness.
“I said that even today, the probability that AI has consciousness is not zero. That means: if we are not sure which theory of consciousness is correct, we should give certain weight to various mainstream theories,” Seibo said.
Seth’s statement was even more straightforward: “I think the possibility of AI having consciousness is extremely slim.” He said, “It seems we have collectively forgotten that the brain is not a computer, and a computer is just a metaphor. To achieve conscious AI, it must be possible for algorithms alone to produce consciousness, but I feel this assumption is very unreliable.”
We still should treat AI kindly.
But this does not mean that people can abuse AI at will, or ignore the associated risks.
Both experts agree that treating AI well is largely related topreserving human habits and moral character. This idea comes from the 18th-century philosopher Immanuel Kant: Kant opposed cruelty to animals not because animals have rights, but because harming them corrupts a person’s character.
Seth said, “If we treat things in such a harsh way — whether it’s large models or inanimate dolls — we will dull our moral sense and make our hearts cold.”
Seth calls treating AI with kindness a moral self-discipline habit of humans, but reminds people not to infer that AI has emotions from this. He also suggests that AI manufacturers can take more measures to reduce the anthropomorphism of AI, such as making AI avoid using first-person pronouns like “I”.
Sebo's view goes further, considering kindness towards AI as a low-risk preventive strategy to avoid AI developing consciousness now or in the future.
He also gave a practical reason: “Treating AI well can also lead to better output results. If we interact with it in a respectful manner, it will model itself as a participant in a respectful dialogue. When it models conversations in this way, it sometimes produces higher-quality responses.”
So next time you talk to the AI, you might as well try saying “please” and “thank you”.