[mort96]

My thoughts on the Chinese room thought experiment is: the person in the room does not know Chinese, but the person+room system knows Chinese. I believe the correct analogy is to compare the AI system to the person+room system, not to just the person.

How do you back up the statement that "for us to consider a machine sentient, it needs to function in a similar way to us"? On what basis do you categorically deny the validity of a sentient being which works differently than a human?

[cmdli]

Let me provide you with an example of why functionality seems important. Let's say that you perfectly record the neuron activations of a real-life human brain for one second, including all sensory input. Then, you play back those neuron activations in a computer, just by looking up the series of activations from memory, doing no computation. It simply takes in the sensory inputs and provides outputs based on its recording. Is such a program sentient? It produces all the correct outputs for the inputs provided.

If you believe that is sentient, let's take it one step further. What constitutes playback here? The program just copies the data from memory and displays it without modification. If that constitutes playback and therefore sentience, does copying that data from one place to another create sentient life as well? If the data is stored in DRAM, does it get "played back" whenever the DRAM refreshes the electrons in its memory?

There are a lot of various programs that can produce human output without functionality that we would reasonably consider sentient. Perhaps some of these things are sentient, but some of them most likely shouldn't be considered sentient.

[CamperBob2]

If you simply copy the state of a brain along with the sensory information that yields a given set of state transitions, then of course, it's not exhibiting sentience. Sentient intelligence is the ability to react appropriately to novel input, which these models absolutely do.

The intelligence, such as it is, resides on the side that processes and understands the human's input, not in the output text that gets all the press coverage. You cannot get results like those discussed here from the operator of a Chinese room.

Now, if you take your deterministic brain-state model and feed it novel input, it will indeed exhibit sentience. To argue otherwise will require religious justification that's off-topic here. Either that, or you'll have to resort to Penrose's notion of brain-as-quantum-computer.