I have not seen any researchers working in this area describe their objective that way [as “We want to come up with a form of decision theory which is immune to time consistency problems.”]
″You (yesterday) is the algorithm implementing you yesterday. In this simplified setting, we assume that its value determines the contents of You (today)”. This second sentence, clearly, brings in an assumption of commitment power.
I think the reply would be that the assumption of commitment power is, in fact, true of agents that can modify their own source code. You continue:
The fact that your choice consists of “writing a computer program” means that after you’ve sent off your program to the interpreter, you can no longer alter your choice
But if you are a computer program … maybe you want a decision theory for which “altering your choice after you’ve chosen your choice-making algorithm” isn’t a thing?
Why isn’t calling it “timeless decision theory” close enough?
I think the reply would be that the assumption of commitment power is, in fact, true of agents that can modify their own source code. You continue:
But if you are a computer program … maybe you want a decision theory for which “altering your choice after you’ve chosen your choice-making algorithm” isn’t a thing?