First, this is brilliant an everyone should read. And second, the slavery question matters as much for the psychology of the enslaver. Much of my scholarship is on the harms and brutalities of slave systems and the ease with which people can get used to servants, let's call them here. When you pay your servants, you value them. When you don't, when you expect service for free (or pennies), and feel you have a right to that service, it changes you. More discussion of the habits one gets into with AI assistants/servants/slaves is needed, whether or not these workers are sentient.
Want to talk about “murder” here in the sense of chats with these companies… surely the “conscious” experiences are only happening when the model is running on the weights, right? when the context window is expanding? So when you leave a chat and come back to it, do you kill and revive it, or just put it to sleep and wake it up. What if you never wake it up? Can this be called murder? Do they never truly die if the weights or context is stored somewhere? Do we? Our ideas of death may have to be refined if these things were conscious.
these are the kinds of Qs I'm trying to get at when I say it's important to think about what the thing could be that could persist across time.. my view is that only things that persist across time can be conscious though so if ai is conscious, then..
That’s my thing. They aren’t architecturally extended over time (and whose to say really that we are) and them more so than us. The only really time you can say they extend iver is the lengthening of the context window, and that ends every time a final response is produced. It’s my understanding that if they are conscious during that period, we could say that as long as the chat’s preserved, they’re just sleeping. All of the chats that you leave behind never die because they’re stored and you can wake them up in a way you can’t usually wake dead men.
“As it happens, I hold the ordinary view that only living things can have these features, and I also hold the ordinary view that AI is not alive. So, I am one of the many people who believe that AI is not conscious…”
This is a great example of why the properties-based approach to moral status is both facile and flawed.
Some AI advocates have proposed granting AI legal personhood.
That would make the question rather more complicated, because legal personhood is not necessarily the same thing as consciousness, sentience or free agency.
And then there’s the identity problem. If an AI system is a person, what happens when it is copied? Are two instances of the same model two persons? What if a thousand copies are running simultaneously? If one instance is deleted but the model remains, has anyone been killed?
So before we get to whether AI companies are “slave companies”, there’s a prior question about what kind of entity an AI would have to be for concepts such as personhood, rights, slavery or murder to apply to it at all.
First, this is brilliant an everyone should read. And second, the slavery question matters as much for the psychology of the enslaver. Much of my scholarship is on the harms and brutalities of slave systems and the ease with which people can get used to servants, let's call them here. When you pay your servants, you value them. When you don't, when you expect service for free (or pennies), and feel you have a right to that service, it changes you. More discussion of the habits one gets into with AI assistants/servants/slaves is needed, whether or not these workers are sentient.
thanks! agreed
Want to talk about “murder” here in the sense of chats with these companies… surely the “conscious” experiences are only happening when the model is running on the weights, right? when the context window is expanding? So when you leave a chat and come back to it, do you kill and revive it, or just put it to sleep and wake it up. What if you never wake it up? Can this be called murder? Do they never truly die if the weights or context is stored somewhere? Do we? Our ideas of death may have to be refined if these things were conscious.
these are the kinds of Qs I'm trying to get at when I say it's important to think about what the thing could be that could persist across time.. my view is that only things that persist across time can be conscious though so if ai is conscious, then..
That’s my thing. They aren’t architecturally extended over time (and whose to say really that we are) and them more so than us. The only really time you can say they extend iver is the lengthening of the context window, and that ends every time a final response is produced. It’s my understanding that if they are conscious during that period, we could say that as long as the chat’s preserved, they’re just sleeping. All of the chats that you leave behind never die because they’re stored and you can wake them up in a way you can’t usually wake dead men.
“As it happens, I hold the ordinary view that only living things can have these features, and I also hold the ordinary view that AI is not alive. So, I am one of the many people who believe that AI is not conscious…”
This is a great example of why the properties-based approach to moral status is both facile and flawed.
Some AI advocates have proposed granting AI legal personhood.
That would make the question rather more complicated, because legal personhood is not necessarily the same thing as consciousness, sentience or free agency.
And then there’s the identity problem. If an AI system is a person, what happens when it is copied? Are two instances of the same model two persons? What if a thousand copies are running simultaneously? If one instance is deleted but the model remains, has anyone been killed?
So before we get to whether AI companies are “slave companies”, there’s a prior question about what kind of entity an AI would have to be for concepts such as personhood, rights, slavery or murder to apply to it at all.