how do you test if a bot like LaMDA is sentient?
In: 4
“it talks like a human” (aka Turing test) is not useful. There were chatbots running under on 1990’s PC’s that could pass it. But they are just parrots with a really big memory, doing keyword searches on a large collection of chat logs, looking for a response to a similar phrase:
https://en.wikipedia.org/wiki/Chinese_room#Chinese_room_thought_experiment
I believe true intelligence means operating on abstract concepts rather than pure words. That could be tested by asking AI questions that require “connecting the dots”; with AI only having info on dots, but not the connections.
there also should be programming that allows for handling of concept and connecting the dots, not just text searches.
PS “sentient”=”a person” is the way the term is used in Science Fiction. IN RL science, “sentient” means having senses like pain, so most animals with a brain fit that. Intelligence is called “sapience”
Latest Answers