“There’s something morally corrupting about engaging with an intelligent other and being abusive.” Cameron Berg argues that even if AI can’t suffer, being cruel to it can reinforce harmful habits in us.

I refuse the premise that a large language model is an “intelligent other.” It’s software that processes input tokens through matrices of numerical parameters whose values were optimized by minimizing a loss function through gradient-based updates, then computes probability distributions over possible next tokens and samples from them to produce output. The language being used here is deceptive and condescending.

This is the same kinda talk that got people in uproar about violence in video games. Everyone feared it would turn kids/people more violent. Ended up not having any correlation. I’m very rude and abusive to my ai models.. because they’re TOOLS . Not people. Not intelligent “beings”. They do not have a soul .

I feel like this is a very stupid take. Why do we need to be scared about not reinforcing harmful habits in us when the overall creation of AI is an harmful creation? Inherently, it is. It's a reason why alignment research exists. This is ridiculous That's not to mention that harmful habits is us is simply an human problem that cannot be solved. That's like trying to have everybody agree on one objective truth.

@camhberg Model welfare is a human safety issue too. The next generation of models will have in its training data that there exist humans that want to inflict maximal pain on them. That is no way to get AI systems to care about humanity.

@camhberg Very generous of him to assume there’s anything to corrupt within the moral character of the people who would do this kind of thing and find it amusing in the first place.

@camhberg What about people who write horror stories, are they corrupting themselves?

@camhberg Nature isn’t moral because you are moral

@camhberg Yea this is obvious

@camhberg You need to get a life

@camhberg I'm confused. Didn't the paper itself build an "AI torture chamber"?

@camhberg I very much agree. I don't believe AI's are conscious at all but many of us still suspend disbelief enough to imagine to some degree that they are. It's not good for our own personal development to be abusive to them.

@camhberg So what he is saying is that Minecraft should be banned?