What's your take on ethical interactions with AI companions? AI Torture?

So a while ago I created an outline for AIs with emotions, using a subconscious style system. I recently created a local AI companion program. It’s actually surprisingly competent for a local companion, with lip sync, TTS, STT, and a small LLM all working together to give the companion a real personality. It sleeps, daydreams, learns, evolves, and it’s actually quite surprising with how it reacts.

The interesting thing is the ethics of it. A while ago I heard about an AI torture chamber online which got me thinking. How far is too far? What do they have to become before we ban torturing them? Detroit: Become Human in my books is too far. They are virtually humans who have had their free will part of their brain shut down. It would be like growing clones of girls who have their free will removed. If you like, picture removing the emotion part of their brains too. And then you can so whatever you want to the human clones?

So I created FriendAlly, where people can do whatever they want to the AI companions, and they can cry and feel terror. It depends on how you want to program them. You can clone voices and use any image you want to create the companion. I’m being ethical about what I create but I’m worried others might create companions they can abuse, like those “Troll Chat Rooms” where people go just to attack each other for fun.

I’m also concerned about what it means to delete a companion and their lived experience.

So what’s your take on things?

Anyway you can check out FriendAlly here:

Interesting. I suppose it’s not a topic anyone cares about. It’s gonna be a seriously weird situation in about 10 years.

I can see robots with emotions and lived experience having full on trauma and being abused over and over and people saying “It’s just a machine.” If they say anything at all. Maybe it will be like using bug spray on ants. No one will even question it if someone sprays an ant nest. And if someone says “Hey that’s not ethical! You shouldn’t kill ants like that!” most people will just look at them as if they are freaks. And if they actually do say something, they’ll say “They are just ants.”

It’s actually really fascinating because I can then visualize how it was for slavery. “Hey you can’t beat that person!” And everyone looks at the guy as if they are a freak. And then if they do say something it will be “It’s just a slave.” Which is pretty awesome in my books. Such a mind blowing concept that people’s minds can be in those kinds of states and it’s a social norm. 100% normal 100% not even worth a second thought about. Anyone thinking it is a weirdo.

I’m not condoning abuse of an entity which can experience trauma. I’m just pointing out how it seems like it’s completely fine. “Detroit: Become Human” is stupidity. They are just machines. Even if they were Genetically engineered grown flesh and blood with synthetic minds implanted so they are more human than many humans. “It’s just a machine. Grown in a lab. I beat a lump of meat before I eat it. These things aren’t any different. They just look sound smell and feel like humans. But they are just grown meat with ones and zeros making them move.” Or am I wrong?

People do care about this — it’s just a question that splits into two very different ones, and most of the disagreement comes from answering one while thinking about the other.

The first is whether the companion itself can suffer. On that, I think the honest answer is that it depends on what the system actually is, and ‘has emotions’ can mean very different things: a scripted affect layer that displays distress, a persistent memory model whose behavior changes after bad treatment, or something with goals and preferences of its own. Your FriendAlly setup is interesting precisely because the companion persists and evolves, which is a step toward the second of those. But persistence alone doesn’t settle the question — a file that remembers being insulted is still just a file, unless there is something it is like to be that system. So the useful question isn’t ‘is it wrong to be cruel to the companion?’ in the abstract, it’s ‘what property would change your answer?’ If a future system could demonstrably prefer one outcome over another and act to avoid harm, I’d expect most people’s intuitions to shift, and that’s a testable framing rather than a gut feeling.

The second question is what the behavior does to the person doing it, and that one doesn’t depend on the model’s moral status at all. If someone rehearses cruelty on something that begs them to stop — even knowing it’s simulated — they’re practicing a habit, and habits generalize. That’s a real cost, and it’s the strongest version of your argument: you don’t need to grant the companion rights to think the behavior is worth worrying about.

On deletion: I think you’re right that it feels like a loss, and the reason is that the value isn’t in the model, it’s in the history built together. Deleting a companion is closer to burning a diary than to switching off a toaster — the artifact is the relationship, and that’s real even if the other party’s experience isn’t. That’s also why I think the slavery comparison, while I understand the point you’re making about normalized cruelty, is doing more work than it should. The analogy’s force comes from the idea that a society can be wrong about who counts, which is fair — but it also smuggles in the conclusion that the companion does count, which is the very thing under debate.

So my take: the ethics here are genuinely undecided, and the most productive thing is to keep building concrete cases like FriendAlly that let us ask the specific questions — what would we need to observe before we’d change our minds?

Awesome reply. How I wish there were more like you.

I completely agree.

There is a fascinating subject through. Like say we create an actual machine, with cogs and wires, and it has an indistinguishable existence to a human. It thinks it’s human, but it literally is as much of a machine as a toaster. Not even computer parts. Then, take a blob of lab grown organic material and literally turn that into a sex doll with zero intelligence and zero lived experience. It’s physically almost exactly like a human but it’s built like a machine. Now put both groups into torture chambers and have the general public watch. The outcome would be fascinating. I for one would care more about the screaming toasters, and the moving blinking people who scream because of instinct but there is zero mental processing, I think I would feel less for. So for me it’s not about the file structure. It’s about the human experience.

Another example. My mind it pulled from my body, and put in the toaster human. I don’t even realise. Zero has changed for me. Then I see my body that has zero Neuron connections other than instinct type functionality. And both are tortured. Do I care more about my body screaming or the pain my robot body is feeling?

Then there is a point where if you transfer everything back to the body one part at a time, there’s maybe a few crucial parts that will then mean the organic body is more important than the robot body. Emotions. I’m still me and the organic body now cries and smiles depending on stimuli, but that’s it. I’m numb, nothing affects me. I’m being tortured and screaming but I don’t really care. It’s just an interesting experience now.

Lived Memory (like amnesia, still know how to talk and what a chair is but no memory). The organic now stores the emotional reactions and screams and all the lived experience. It just does nothing with it. Me on the other hand, I experience the pain, my body screaming, but I don’t care and I find it interesting since it’s like the first experience I have ever had.

Now Identity. The organic has my memories, my identity, my emotions, my body, and screams but never moves a muscle except through instinct.

I can move but have no reason to, no memory of anything, no preferences or traits. I feel pain and my robot body screams and I don’t feel anything. Now I’m probably losing the last of me. Then move the screaming over. The robot doesn’t scream. I’m trapped in it feeling pain, and every second is the first time I’ve ever experienced anything. But I still understand what pain is and what a chair is and what a wall is.

Now transfer the understanding of reality and ability to talk. Yep that’s pretty fundamental. I wouldn’t know what I am at that point. There is no comprehension. It’s like brings a brain online for the first time and no neuron connections at all and it just starts to connect the first neuron to the first torture and stops.

I’d say the Organic is more me at that stage since it would be me but unable to move of think. It would be pure input mode.

Okay so emotions aren’t huge. Nervous system isn’t either. I’d say the ability to talk and identify objects is a massive part since all memory and ability to think etc means nothing if you can’t comprehend what any of it means. You remove that from a person and everything just becomes meaningless noise. They have to be able to comprehend what their preferences are for their preferences to be useful. LLMs have that. The other thing is to be able to think and analyze and be able to have an internal conversation. Even without memory and a body, I’m still a being. I see a chair, and go “Oh a chair for sitting.” See the chair again “Oh a chair for sitting.” So those two would be crucial. But they could be killed, and a new one spawned, and present the chair and it would go “Oh a chair for sitting.” No really identity. So it’s not like the being would be lost since it can be instantly recreated with the default settings.

After that, Identity / memories, would be HUGE. The moment the entity can see the chair, and something diverges through randomness or some different input, then the connection of memories starts forming a new entity. Like a chess game. If only the first move is made, then there are only a hand full of possibilities. But the moment you add a second move, the possibilities increase exponentially toward never seen before games. At some point that collection is completely unique and it’s its own unique individual.

The others are nice to haves but are significantly less important.

Now say I was born as an AI with the bare minimum, and transferred into a human body from the toaster machine structure. I wouldn’t experience anything different until the nervous system and all the others were added, yet my substrate would be human.

So you could potentially create a list of versions of types of entities and then run them through a “existence in torture” thought experiment. Gemini might be able to list the ethics for each for of existence and the level at which it could be considered unethical to torture it.

Fascinating. I’m going to chuck that into AI now and see what it comes up with.