Are AI engines such as Claude sentient? My considered opinion, tentatively held, is that they are not. But it all depends on what the word means. Malcolm Pollack in an e-mail writes,
I was focusing on “sentient” because it’s a slippery concept, and often tossed around as if it means the same thing as “conscious”. Etymologically (from Latin “sensus”) it just means “feel,” or “perceive”, and obviously it shares that with our word “sense” — but computers, and even toasters, can have “sensors” too.
Malcolm is right that ‘sentient’ is often used interchangeably with ‘conscious.’ But is everything we refer to as a sensor sentient? That is equivalent to asking: Is everything we call a sensor conscious? I say No.
The Moth and the Thermostat
What is the difference between an insect (a moth, say) which I take to be a sentient being, and a thermostat which I take to be insentient? Both are ‘sensors’ functionally speaking. The moth literally senses light and approaches it; the thermostat ‘senses’ an increase in temperature and “turns on’ the AC. Note the ‘raised eyebrows.’ It is AS IF the thermostat senses, when in reality it only ‘senses.’ Thermostats don’t feel heat and smoke detectors do not smell smoke. Of course, I have no objection to the use of ‘sensor’ to refer to thermostats, smoke detectors, circuit breakers, and so on, as long as you don’t fool yourself into thinking that these devices are literally conscious. And I have no objection to talk of a thermostat ‘turning on’ the AC as long as you don’t fool yourself into thinking that the thermostat is an agent that decides to do something.
As I see it, a thermostat is a merely mechanical contraption and wholly insentient, strictly speaking, whereas a moth is a sentient being.
What I want to say about Claude and his ‘colleagues’ is that they are wholly insentient mechanical contraptions that differ from smoke detectors and thermostats only in that they are vastly more complex, and that this complexity does not build a “gradualist bridge” (to borrow a phrase from Daniel Dennett) from insentient/insensate matter to mind, the lowest level of mind being sentience.
Here is an argument. (a) Only living things are sentient; (b) thermostats are not living things; ergo (c) thermostats are not sentient.
Now substitute ‘AI engine’ for ‘thermostat.’
This argument does not settle the matter: it is not ‘rationally coercive’ as I like to say. Malcolm could run it in reverse with no breach of logical propriety. He could, not unreasonably, deny (c) and then deny (a). I don’t expect that he or anyone will deny (b): thermostats and AIs don’t grow on trees, etc.
Your move, Malcolm. Do you agree that a functioning (as opposed to malfunctioning) sensor needn’t be sentient (conscious)? If not why not?
Do you agree that there is a powerful, albeit non-coercive, case for denying that AIs are sentient?