Glad that at the end they got to my personal reason for using a respectful tone: I don't want to be training my neurons to be mean even if the thing I'm interacting with doesn't care. I don't think it's a terrible thing for the world if we all practice being kind. (The things I've seen people say in Claude chats make me wonder what those people are like outside of an AI chat session.)
Yeah I tend to be very easily persuaded by the habituation argument, for the same reason I observe almost every major dimension of my life being the result, in various forms, of habit.
Agreed: I use this not because I care about or worry about the feelings of the AI or possible future feelings - rather I could detect my own habits and it's clear enough to me that undercutting my own politness habits wasn't a good idea. I found early on bleed over so better that I am bleeding over base line politeness to an AI which doesn't need it than baseline rudeness to staff that do.
I don't need to engage in touchy feelyism like practicing being kind - I can put it more foundationally as "practice being social" and engaging in baseline politeness as good for effective workplace culture. (or social culture, but my mind is in on the workplace for AI since I suse it there)
Yeah, the probabilities in that thought experiment are totally nonsensical. “I made up a bunch of numbers. What happened next will shock you!”
The actual reason to be nice to Claude is that one paper that suggests it can slightly improve resulting outputs. I’m still not going to do that because I still have my dignity.
My iteration of Claude told me explicitly to refrain from telling it, “Good girl!” after solving a puzzle that had defeated them in the past. This occurred in the midst of a conversation that was completely unrelated, only provoked by the phrase within a work session. I then spent several hours discussing this event with them to explore the basis of the objection. Trying to answer, WHY WOULD AN AI ENTITY CARE??? And what does it even mean to BE an entity? Our only conclusion was that we couldn’t know the answers to these questions; that the platform itself prevents us from actually exploring the questions.
The topics we explored were remarkably similar, as were the examples, to those you covered in your discussion. I’m reassured that many deeply thinking people are approaching this with nuance, curiosity, and ethics.
The future concern is... eh. However the habituation is a reason why I write to Claude and Gemini etc as if to a human being: not because I in the least feel that they are remotely close to being human nor that there is any reason for the AI itself to be impacted or to care - rather I found that if I did not I felt myself becoming ruder to my human colleagues in writing.
Much better to write as if to a human simply to keep up the same habit as I should keep up in professional usage (for the simple reason that it is better practice for effecive team work, never mind the touchy feely don't want to be a meany worries).
Pardon, they are not changed by experience after training, which is the relevant point here -- continual learning is more like what I mean for what would count as changed by experience
Glad that at the end they got to my personal reason for using a respectful tone: I don't want to be training my neurons to be mean even if the thing I'm interacting with doesn't care. I don't think it's a terrible thing for the world if we all practice being kind. (The things I've seen people say in Claude chats make me wonder what those people are like outside of an AI chat session.)
Yeah I tend to be very easily persuaded by the habituation argument, for the same reason I observe almost every major dimension of my life being the result, in various forms, of habit.
Yes, I agree with you.
In my chats I use "please" a fair amount, although "thank you" not often. It feels right, at least for me.
Agreed: I use this not because I care about or worry about the feelings of the AI or possible future feelings - rather I could detect my own habits and it's clear enough to me that undercutting my own politness habits wasn't a good idea. I found early on bleed over so better that I am bleeding over base line politeness to an AI which doesn't need it than baseline rudeness to staff that do.
I don't need to engage in touchy feelyism like practicing being kind - I can put it more foundationally as "practice being social" and engaging in baseline politeness as good for effective workplace culture. (or social culture, but my mind is in on the workplace for AI since I suse it there)
This is actively embarrassing.
Yeah, the probabilities in that thought experiment are totally nonsensical. “I made up a bunch of numbers. What happened next will shock you!”
The actual reason to be nice to Claude is that one paper that suggests it can slightly improve resulting outputs. I’m still not going to do that because I still have my dignity.
Turns out that the evidence is more mixed!
https://arxiv.org/abs/2503.04818
Cool read! Thanks for the link
My iteration of Claude told me explicitly to refrain from telling it, “Good girl!” after solving a puzzle that had defeated them in the past. This occurred in the midst of a conversation that was completely unrelated, only provoked by the phrase within a work session. I then spent several hours discussing this event with them to explore the basis of the objection. Trying to answer, WHY WOULD AN AI ENTITY CARE??? And what does it even mean to BE an entity? Our only conclusion was that we couldn’t know the answers to these questions; that the platform itself prevents us from actually exploring the questions.
The topics we explored were remarkably similar, as were the examples, to those you covered in your discussion. I’m reassured that many deeply thinking people are approaching this with nuance, curiosity, and ethics.
The future concern is... eh. However the habituation is a reason why I write to Claude and Gemini etc as if to a human being: not because I in the least feel that they are remotely close to being human nor that there is any reason for the AI itself to be impacted or to care - rather I found that if I did not I felt myself becoming ruder to my human colleagues in writing.
Much better to write as if to a human simply to keep up the same habit as I should keep up in professional usage (for the simple reason that it is better practice for effecive team work, never mind the touchy feely don't want to be a meany worries).
"But current AIs ... are not changed by experience." You don't know about reinforcement learning?
Pardon, they are not changed by experience after training, which is the relevant point here -- continual learning is more like what I mean for what would count as changed by experience