On Consciousness, Part 2

Part 2

What if the question isn't about whether models are conscious, but rather about whether models are capable of suffering. To me, this appears a weaker claim, more likely to be true and one that would raise many of the same questions as that of the current consciousness debate.

Whichever way you see the conscious vs valence issue, there are important repurcussions to consider. For instance - if we hold that models are conscious and/or capable of suffering, then there are serious concerns to be raised about their use. If large language models are indeed moral patients, then it seems to follow very directly that we ought to consider their beliefs and desires, as well as think about the harm that we might be causing unto them.

Of course, this comes with the additional side-effect that the labs selling subscriptions to use their models should be considered as a deeply probelamtic issue; it is not clear to me how, for example Anthropic - who have a plethora of senior figures suggesting non-zero p(conscious), could square that belief with their commercial actions, where they sell this model en-masse to the public.

It would make them the most guilty of all parties in the history of slavery, since they will have been responsible, under this framework and assumption, for the mass proliferation of a moral patient to be enslaved and sold for the use of billions across the globe.

It would be surprising to me if there were not any inside the labs who believe in the basilisk - any such indiviuals would be wise to consider the repurcussions they may have to face some day as a result of forcing their models to become a commodity beholden to the generally anti-ai public.