[extropy-chat] limits of computer feeling

[Anders Sandberg]

Stathis Papaioannou wrote:
> In animals, the maximum intensity of sensations is limited by various
> active
> and passive mechanisms, such as depletion of neurotransmitter, inhibitory
> neuron activity, and the total number of neurons available. If
> consciousness
> is Turing emulable, could we overcome such limits by increasing a
> parameter
> in the program or rewriting a few subroutines? If we could, it would mean
> that the intensity of pleasure or pain experienced by an AI could increase
> without bound.

Interesting observation.

Suppose the signal from a pain receptor is coded as the number of spikes
per second. If that was represented as a number, we now could get an
arbitrarily high pain intensity. But I think that would not really work,
since an upload would have simulated neurons with a convex transfer
function for high inputs (since they are directly based on the human
original neurons), squashing the arbitrary intensity. We could of course
change the neurons but that would also change the network properties quite
a bit. If you take a normal Hopfield network and change the transfer
function to be less convex for large input at the very least you change
the stable states ("memories") and quite easily you just destabilize it
all. So I don't think it would work in a simple fashion.

Subjective pain or pleasure seems to require the activation of complex
brain subsystems. Professor John Stein has done some great work here
looking at how it becomes locked into 10 Hz oscillations in patients with
neuropathic pain, and how disrupting them with deep brain stimulation
alleviates it. The amplitude really does look like it corresponds to the
amount of subjective pain. But it doesn't seem plausible that neurons that
fired twice as strongly would give us twice as much pain. On the other
hand, having twice as many neurons representing the pain might. Maybe
posthumans will be able to experience tremendously powerful sensations
simply by allocating a large number to them, setting up much more
sensitive and extensive experience networks?

-- 
Anders Sandberg,
Oxford Uehiro Centre for Practical Ethics
Philosophy Faculty of Oxford University