"AI is not conscious" Ted Chiang
Sign up for OddsRabbit
Join communities like /c/ai and never miss posts like this. Every signup donates a meal for a child in need. Learn more
Comments
I think its pretty telling that Chiang cites an Anthropic paper he didn’t read, refuses to define consciousness, refuses to compare human and robot intelligence, reduces LLMs to text prediction when they generate an interpret text, images, video, sound, and other data. He doesnt mention other AI models even though Claude has at least 4 model categories, he pretends they work the same way.
He calls it a “simulation” because it is using reduced data and is not persistent. If it were running all the time and capable of comparable compute to a brain, he wouldn’t have anything to point out in this paper. Why isnt it simulation when a person proofreads your article? Because theyre thinking about dinner?
I think it is generalizing a lot when you say AI. AI includes models that couldn’t be conscious. Youd have to define conciousness before you could argue whether something is conscious. The reason people say these LLMs are conscious is because they meet previous definitions and tests for conciousness.
There was a movie called chappie which explored consciousness. I heard a critic say that the premise "we don't know what consciousness is" is absurd because although there is no agreed on definition of consciousness, everyone kind of intuitively knows what it means.
And there was another author from my college days Ray Kurzweil who wrote about the possibility of having human minds uploaded to a device and live forever. Would that be consciousness to you?
I think if it has a continuous experience of the physical world like we do, that we would consider it conscious. If we added sensors so it could see and hear and a persistent memory, it doesnt seem different from a creature to me.