Re:Language models are nearly AGIs but we don't notice
by andrewmcwatters on Sunday November 13, 2022 @03:41PM
(#33587106)
I have yet to see someone plainly and concisely articulate what they mean by "artificial general intelligence."
There are all sorts of different types of intelligence. Do you mean the intelligence of an adult human? An adolescent human? A pigeon?
If you mean, "I want to be able to talk to a computer, for it to have a personality, likes and dislikes, feelings, and learn over time." That's a pretty neat definition.
But then my question for you is, and here's where the magic starts to disappear and become a little depressing for some: why can't that be explicitly programmed today?
What is stopping a small handful of talented engineers from slapping together a set of technologies that fulfills this definition?
I suspect most of the work to do this would simply be in creating the systems explicitly.
An example would be, I want the agent to notice a user over some long period of time and trigger a refamiliarization and catch-up conversation.
But if you're asking, how do we throw the equivalent of digital neurons together in the hopes that something human-like will eventually come out: tough cookies, we don't even know every part of how actual humans today work.
And unfortunately, I suspect that if we did, the least amount of engineering labor to greatest amount of user experience pay-off would look the same: just explicitly create all of the things you want a supposed intelligent agent to do.
I suspect you could, for good measure, throw in some sort of process to provide an actual mechanism for "learning," but I doubt we will ever, ever see an abstract model for creating neurons and auxiliary software models that creates essentially the phenomenon of life.
I don't think there's anything wrong with a simulated, and frankly, just more advanced Siri.
I already don't like Siri, she has too much attitude for me. Turn off my lights and don't give me lip.
Besides, if we did have true AGI, it would associate itself with particular tribes, and then people would be screaming for it to be reprogrammed to be neutral, or on the other team.
There are all sorts of different types of intelligence. Do you mean the intelligence of an adult human? An adolescent human? A pigeon?
If you mean, "I want to be able to talk to a computer, for it to have a personality, likes and dislikes, feelings, and learn over time." That's a pretty neat definition.
But then my question for you is, and here's where the magic starts to disappear and become a little depressing for some: why can't that be explicitly programmed today?
What is stopping a small handful of talented engineers from slapping together a set of technologies that fulfills this definition?
I suspect most of the work to do this would simply be in creating the systems explicitly.
An example would be, I want the agent to notice a user over some long period of time and trigger a refamiliarization and catch-up conversation.
But if you're asking, how do we throw the equivalent of digital neurons together in the hopes that something human-like will eventually come out: tough cookies, we don't even know every part of how actual humans today work.
And unfortunately, I suspect that if we did, the least amount of engineering labor to greatest amount of user experience pay-off would look the same: just explicitly create all of the things you want a supposed intelligent agent to do.
I suspect you could, for good measure, throw in some sort of process to provide an actual mechanism for "learning," but I doubt we will ever, ever see an abstract model for creating neurons and auxiliary software models that creates essentially the phenomenon of life.
I don't think there's anything wrong with a simulated, and frankly, just more advanced Siri.
I already don't like Siri, she has too much attitude for me. Turn off my lights and don't give me lip.
Besides, if we did have true AGI, it would associate itself with particular tribes, and then people would be screaming for it to be reprogrammed to be neutral, or on the other team.