Has anyone proposed something like an extended Turing Test? Meaning, like, it's one thing to have a conversation or two, or three, with a machine and been unable to tell it from human. Quite another to, say, work with it and make all sorts of requests over several weeks and not be able to tell.
Comments
Level 2: pass a week of regular interaction/coworking
Level 3: a month
Level 4: a year
Or something similar.
if they are a human, loop forever.
LLMs generate answers when prompted. Otherwise, they sit idle.
Consciousness requires autopoiesis, being-in-the-world, and agency. LLMs have none of these.
My monitor went to the farm with my dog.
Noc Noc, do you have dogs? And monitors?
Looks like I don't need your memory chip after all!
This was in 1993.
(Personally I’m not ultimately sure that consciousness is a well defined meaningful term to begin with, so any test for whether an AI has it tells us more about the test creators than the AI.)