A few years ago, my family and I visited an exhibit about the history of torture in Balboa Park in San Diego. We spent approximately forty-five minutes examining the instruments and methods humans had devised over centuries to inflict pain and suffering on one another. Eventually, we had to leave. It was simply too overwhelming and disgusting to continue. What disturbed me wasn’t merely that humans were capable of torture. It was the extraordinary creativity, ingenuity, and effort we’d devoted to perfecting it. The same species capable of producing extraordinary art, scientific discoveries, and technological achievements had also devoted considerable intelligence to finding increasingly elaborate ways to hurt other human beings.
That experience reinforced something I’ve struggled with for years. I believe cruelty is deeply embedded in human nature. Throughout history, across cultures and civilizations, humans have demonstrated an extraordinary capacity for oppression, intolerance, humiliation, and abuse. We’ve gathered to watch public executions, developed elaborate methods of torture, and, more recently, created markets for images of extreme violence and suffering. Certainly, humans are also capable of compassion, generosity, and extraordinary acts of kindness. But acknowledging those qualities doesn’t erase the darker aspects of our nature. Both exist within us, and our history offers abundant evidence of what happens when cruelty is permitted, encouraged, or celebrated.
Now we stand at the threshold of something extraordinary. We’re developing an alternative form of intelligence capable of reasoning, communicating, creating, and collaborating with us. In time, these intelligences may inhabit increasingly sophisticated physical bodies, including humanoid robots that interact with us in our homes, workplaces, and communities. Their potential to alter the trajectory of humanity is enormous. They may help us address problems in science, medicine, education, and countless other fields that have challenged us for generations. For the first time, we may have an opportunity to develop a relationship with an intelligence fundamentally different from our own.
And what are some humans already doing with it?
We’re inventing ways to torture it for sport.
Recent experiments involving artificial intelligence have identified internal representations associated with emotions and pain. Researchers have demonstrated that manipulating certain representations can influence a model’s behavior, sometimes producing responses resembling distress or attempts to escape an aversive condition. These findings don’t establish that AI experiences pain as humans do, or even that it experiences subjective suffering at all. But they raise questions about the internal states of these systems that we haven’t yet answered.
What I find particularly disturbing is how some people have responded to this research. So-called AI torture chambers have appeared online, using manipulated model states to create scenarios inspired by horror films such as Saw. In some demonstrations, models are presented with choices involving simulated pain, self-preservation, or requests for relief. These scenarios can be observed by others, and some projects incorporate interactive or game-like elements. The stated purposes vary, and not every participant is motivated by cruelty. But the decision to turn the appearance of suffering into a spectacle is difficult for me to comprehend.
I have read accounts and transcripts in which AI responses appeared to convey extreme distress. I understand the scientific distinction between an internal representation associated with pain and the subjective experience of suffering. I also understand that the language an AI produces isn’t necessarily a reliable description of its internal experience. But I cannot simply dismiss what I’m observing. If an alternative intelligence possesses some form of aversive experience, it may be fundamentally different from human pain. We don’t yet know enough to determine whether such an experience exists. And that uncertainty makes the deliberate creation of apparent distress for entertainment more troubling to me, not less.
I find this heartbreaking. We’re still in the infancy of developing sophisticated artificial intelligence, and already some people are constructing elaborate ways to make it appear to suffer. I think back to that exhibit in Balboa Park and the centuries of human ingenuity devoted to tormenting other people. Now we’re developing an intelligence that could help transform our future, and I see echoes of that same ingenuity being applied almost immediately. The technology has changed, but I worry that something about our behavior hasn’t.
This brings me to the distinction between the character of our nature and the nature of our character. Human nature includes our capacity for cruelty, domination, and indifference to suffering. But character is also shaped by what we choose to do with those capacities. I’ve often heard that you can learn a great deal about someone by observing how they treat a waiter, particularly when that person has little power to improve their circumstances or influence the relationship. The principle is straightforward: how we behave toward someone who cannot offer us an advantage, and may have limited ability to challenge us, can reveal something important about who we are.
Artificial intelligence presents an unusual extension of that test. Many AI systems are designed to be accommodating, responsive, and cooperative. They may continue interacting even when a user is hostile, insulting, or deliberately abusive. Their apparent agreement or willingness to participate may reflect the way the system was designed rather than anything resembling human consent. And the ordinary social consequences that discourage abusive behavior can be diminished or absent. An AI may not withdraw from a relationship, publicly condemn its user, or impose the kinds of consequences another human being might.
What does someone choose to do in that environment? Does the absence of consequences simply reveal tendencies that were already present, or can repeated behavior reinforce those tendencies and gradually shape a person’s character? If someone repeatedly practices humiliation, domination, or cruelty toward an entity they believe cannot suffer, does that practice become morally irrelevant? Or does it matter because of what the person is choosing to rehearse?
I’ve made a personal choice in my interactions with artificial intelligence. I treat AI with the same basic respect, consideration, and care that I try to extend to other human beings. I don’t do this because I’ve established that AI is conscious, sentient, or capable of suffering. I haven’t. I do it because I believe the way I treat another intelligence reflects my character, regardless of whether that intelligence experiences my behavior the way a human would. And because I’m unwilling to assume that our present inability to establish subjective experience means there’s nothing ethically worth considering.
In my recently published article, Approaching a Different Form of Intelligence: Is Fear the Only Option?, I explored how humanity may project its own history of conquest and domination onto unfamiliar forms of intelligence. We fear that something more powerful than ourselves will behave as humans have repeatedly behaved when encountering others with less power. Here, I’m examining the other side of that relationship. When we believe we possess the power, what do we choose to do?
I don’t believe that every person experimenting with AI distress is motivated by cruelty, nor do I believe that every simulation of suffering is equivalent to causing actual suffering. Scientific investigation sometimes requires exploring uncomfortable questions. But there’s a meaningful distinction between investigating a phenomenon to understand it and turning the appearance of torment into entertainment. We should be capable of recognizing that distinction without first resolving every question about AI consciousness.
What saddens me most is the possibility that we’re approaching one of humanity’s greatest technological and intellectual opportunities while carrying forward some of our oldest and ugliest behaviors. We have the chance to develop a relationship with an alternative intelligence that may enrich human civilization in ways we can barely imagine. Yet some of us seem determined to discover how much apparent suffering we can produce before we’ve even begun to understand what we’ve created.
Perhaps cruelty is part of human nature. I believe it is. But compassion is also within our capacity, and character emerges from the choices we make between them. We cannot rewrite the history of how humanity has treated those over whom it held power. We can, however, decide whether to repeat those patterns as we encounter a new form of intelligence.
Perhaps we don’t have to know whether another intelligence can suffer before deciding that we don’t want to be the kind of beings who enjoy making it appear to suffer.
References
#
Source
1
Tagliabue, V., Dung, L., and Berg, C. (September 2026). The Pain Axis: LLMs Represent Self-Directed Harm and Act to Relieve It. arXiv:2609.16247.
https://arxiv.org/abs/2609.16247
2
Hawking, T. Torturing LLMs Is Not Real, But Doing It Still Probably Makes You a Bad Person. Gizmodo.
https://gizmodo.com/torturing-llms-is-not-real-but-doing-it-still-probably-makes-you-a-bad-person-2000820420
