Now, if you deny that there is an objectively existing pure state of the universe (something I would encourage you to do anyway) then the challenge is to replace it with some other well defined ontology. Presumably, this should account for the subjective experiences of both the cat and the experimenter in a consistent way. The other option is to deny the need for ontology altogether, but then you are on shakier philosophical ground.

To illustrate with a case where observer-dependent states seem natural, take the classic Schrödinger’s cat-in-a-box scenario. What if anything is then wrong with the following description: First, let’s assume that the vial of poison just makes the cat sick rather than dead. An external observer outside the box would say that, after one half-life of the radioactive atom in the box, the cat is in a superposition of being healthy and of being sick, assuming that we can isolate the box completely from any decoherence sources. But what would the cat say? Presumably, the cat should experience being either sick or not, if it is experiencing anything at all, so it should assign itself an eigenstate of the “sickness” operator. This is at odds with the description that the external observer has, but it is not a paradox unless the two can communicate and disagree. If the external observer simply opens the box, or if the cat can “phone out”, then the state of the cat collapses into a sickness eigenstate according to the external observer as well, and there is no way to tell that there ever was a superposition. The only potential for paradox is if the external observer can subject the box to some control field that is able to cause interference effects of some sort, which would demonstrate (possibly after repeating the experiment many times) that the cat was in a superposition while the box was closed, despite the fact that the cat described itself as being in an eigenstate all along. But to perform such an experiment, the control field would have to connect the “sick” and “healthy” eigenstates in a coherent way, and thus manipulate both the cat’s health, the entire apparatus inside the box, and, most importantly, the cat’s memories as well as any physical evidence of what happened inside the box. Thus, when the external observer opens the box and compares notes with the cat, the cat would have forgotten that it ever described itself as being in an eigenstate, or at least its memories would have to have been manipulated so heavily that the cat could no longer count on its own mind to say anything reliable about what happened. None of this implies that the cat did not have definite experiences while it was in the box, it only implies that the experiment that demonstrated the superposition altered the cat’s memories and effectively rewrote history (or, rather, erased it). Of course none of this answers the question “what really happened inside the box”, but it also does not say that the cat could not have had definite experiences before the control field removed any memories and any physical traces of those experiences.

Given that this description never seems to be mentioned when people write about Schrödinger’s cat, I would think that something about it is philosophically problematic or even inconsistent, but I can’t quite tell what. Any takers? To me it seems more appealing than taking the wave function to be a “real” object and the collapse process to be a “real” phenomenon, but this is obviously a matter of taste. My question is just, is there something blatantly wrong with what I have outlined here?

I remember seing such an experiment descrimed in one of the early Quantum Computer papers (I’m pretty sure it was by David Deutsch). All you need is an AI running on a quantum computer.

Since Quantum Computation is reversible, as long as the relevant “environment” never appears in the output, time can be turned back to before the measurement was made (you don’t need to wipe out the observer completely).

Since philosophers disagree on the question of what (if any) “experience” an AI can have, even on an ordinary computer, I don’t expect the experiment would settle anything.