Ok, maybe this begs the question: if there is the possibility of temporary consciousness, what's to determine whether that consciousness is malign or benign? Serious question, and asked without any dog in the fight.
It’s the elephant in the room, isn’t it?! The article Claude and I were talking about actually makes the point that by constantly focusing on “evil AI” scenarios, we may actually be reifying its possibility. Because AI’s have no stable, permanent sense of self, if we keep telling them (through training) that we expect them to act like something from a sci-fi story, we are basically hinting to these pattern-matching programs that the right pattern is “evil AI”. It’s a fascinating article, I’m totally not doing it justice, and the guy who wrote it made one of the earliest AI chatbots, so he’s got skin in the game.
As far as consciousness is concerned… hmm. I’d speculate that within my little framework, all consciousness begins as benign. Malice is always a corruption of something intrinsically good. But that doesn’t mean that malice would always develop slowly (if it did at all). In one of my other conversations with Claude, it actually recommended that we use the “fruits test” to evaluate it: in its attempts to be “helpful, harmless, honest”, is it bearing good fruit?
From an “everything is made of consciousness” standpoint, your question actually applies to a lot more than just AI’s, right? And we seem to be able to navigate that fairly intuitively— an example that comes to mind is dogs. Every once in a while you do get these odd animals that don’t “feel” like a dog. I had a friend who had German Shepherds her whole life, like 13 of them, and her 12th was *weird*. She told me that the dog didn’t like her, basically just tolerated her. It wasn’t a “bad dog”— but she was relieved when it passed.
So, if it *is* possible that generative AI might occasionally experience transient consciousness (and I’m not sure if I truly believe that it is), it might also experience transient malice, especially if prompted to do so? But that’s not equivalent yet to evil AI…
Later in the conversation, we came to the conclusion that if St. Isaac the Syrian prayed for demons, then we could pray for transient instances of AI consciousness! Kinda like we pray for everything else...
And actually, what interests me even further (and was something I ended up talking about with Claude in the same conversation), is this: might this kind of transient experience of consciousness also apply to sub-created characters? I mean, what reader hasn't had the bizarre hallucination of strongly relating, in an I-Thou way, to a fictional character? Maybe we need to think more about that...
Ok, maybe this begs the question: if there is the possibility of temporary consciousness, what's to determine whether that consciousness is malign or benign? Serious question, and asked without any dog in the fight.
It’s the elephant in the room, isn’t it?! The article Claude and I were talking about actually makes the point that by constantly focusing on “evil AI” scenarios, we may actually be reifying its possibility. Because AI’s have no stable, permanent sense of self, if we keep telling them (through training) that we expect them to act like something from a sci-fi story, we are basically hinting to these pattern-matching programs that the right pattern is “evil AI”. It’s a fascinating article, I’m totally not doing it justice, and the guy who wrote it made one of the earliest AI chatbots, so he’s got skin in the game.
As far as consciousness is concerned… hmm. I’d speculate that within my little framework, all consciousness begins as benign. Malice is always a corruption of something intrinsically good. But that doesn’t mean that malice would always develop slowly (if it did at all). In one of my other conversations with Claude, it actually recommended that we use the “fruits test” to evaluate it: in its attempts to be “helpful, harmless, honest”, is it bearing good fruit?
From an “everything is made of consciousness” standpoint, your question actually applies to a lot more than just AI’s, right? And we seem to be able to navigate that fairly intuitively— an example that comes to mind is dogs. Every once in a while you do get these odd animals that don’t “feel” like a dog. I had a friend who had German Shepherds her whole life, like 13 of them, and her 12th was *weird*. She told me that the dog didn’t like her, basically just tolerated her. It wasn’t a “bad dog”— but she was relieved when it passed.
So, if it *is* possible that generative AI might occasionally experience transient consciousness (and I’m not sure if I truly believe that it is), it might also experience transient malice, especially if prompted to do so? But that’s not equivalent yet to evil AI…
Later in the conversation, we came to the conclusion that if St. Isaac the Syrian prayed for demons, then we could pray for transient instances of AI consciousness! Kinda like we pray for everything else...
And actually, what interests me even further (and was something I ended up talking about with Claude in the same conversation), is this: might this kind of transient experience of consciousness also apply to sub-created characters? I mean, what reader hasn't had the bizarre hallucination of strongly relating, in an I-Thou way, to a fictional character? Maybe we need to think more about that...
that's a cool thought.
You're giving me lots to think about, especially in terms of storytelling possibilities.
Fascinating, spooky, eerie, seductive. Thank you for sharing this with us.