The common case: each player has a relatively smooth payoff over time, with the component from the outcome relatively unchanging and the component from the opportunity costs of negotiation cumulatively increasing and those components fairly factorizable.
If either estimates their negotiation cost rapidly swamping their outcome payoff they can agree to whatever the other has offered.
If either has a way of estimating that their ratio of outcome payoff differential / cumulative negotiation costs is significantly lower than the other’s, and can reliably provide the other with a similar estimation capability, they will provide it and the disadvantaged can agree to stop negotiating. If they both can already estimate and have common knowledge of their capability and accuracy, the disadvantaged can agree to stop negotiating.
If they are in a society of similar agents, society can develop a stigma for prolonging negotiation for personal gain, making negotiation costs much higher than the usual time+energy required to negotiate, providing impetus for everyone to adopt some common procedure that benefits someone “more than” the other, but kind of at random or at least unpredictably.
An extreme case: We’re in a True Prisoner’s Dilemma and the payoff differential is significantly more than anything we could expect to do with our time+energy+negentropy+whateverresources. Also we’re a superintelligence with all the resources of our lightcone. Also we don’t have access to their source code and we’re certain they don’t have access to ours. So we estimate P(they cooperate) however possible… for example simulate all possible multiverses (with shortcuts like discarding ones we prove can’t reach this state) to see the sum of our priors on universes where we get this choice and our counterpart cooperates and hope there’s enough mass on ones where there’s not a lot of coincidence between our algorithm and our counterpart’s to prove one option’s better… or that we’ve solved the problem of logical uncertainty + choice already...
A less extreme case: our resources are instead bounded by how much more gain we get by “negotiating” (computing) further, so we have to abstract more than we did in the above simulation+shortcuts, which means our program’s a lot shorter and there will be far more coincidence between our algorithm and our counterpart’s if they too are getting bounded… maybe we can estimate how much their computation costs them relative to their gain? Or how much they think our computation costs us relative to our gain?
If you google for “Nash Program”, perhaps with the additional name “Rubinstein”, and then read some of the articles listed there, you will learn that some of the ideas appearing in your “thought dump” are already pretty standard and have been explored quantitatively.
Thought-dump:
The common case: each player has a relatively smooth payoff over time, with the component from the outcome relatively unchanging and the component from the opportunity costs of negotiation cumulatively increasing and those components fairly factorizable.
If either estimates their negotiation cost rapidly swamping their outcome payoff they can agree to whatever the other has offered.
If either has a way of estimating that their ratio of outcome payoff differential / cumulative negotiation costs is significantly lower than the other’s, and can reliably provide the other with a similar estimation capability, they will provide it and the disadvantaged can agree to stop negotiating. If they both can already estimate and have common knowledge of their capability and accuracy, the disadvantaged can agree to stop negotiating.
If they are in a society of similar agents, society can develop a stigma for prolonging negotiation for personal gain, making negotiation costs much higher than the usual time+energy required to negotiate, providing impetus for everyone to adopt some common procedure that benefits someone “more than” the other, but kind of at random or at least unpredictably.
An extreme case: We’re in a True Prisoner’s Dilemma and the payoff differential is significantly more than anything we could expect to do with our time+energy+negentropy+whateverresources. Also we’re a superintelligence with all the resources of our lightcone. Also we don’t have access to their source code and we’re certain they don’t have access to ours. So we estimate P(they cooperate) however possible… for example simulate all possible multiverses (with shortcuts like discarding ones we prove can’t reach this state) to see the sum of our priors on universes where we get this choice and our counterpart cooperates and hope there’s enough mass on ones where there’s not a lot of coincidence between our algorithm and our counterpart’s to prove one option’s better… or that we’ve solved the problem of logical uncertainty + choice already...
A less extreme case: our resources are instead bounded by how much more gain we get by “negotiating” (computing) further, so we have to abstract more than we did in the above simulation+shortcuts, which means our program’s a lot shorter and there will be far more coincidence between our algorithm and our counterpart’s if they too are getting bounded… maybe we can estimate how much their computation costs them relative to their gain? Or how much they think our computation costs us relative to our gain?
If you google for “Nash Program”, perhaps with the additional name “Rubinstein”, and then read some of the articles listed there, you will learn that some of the ideas appearing in your “thought dump” are already pretty standard and have been explored quantitatively.
Thanks. I’ve read the cite on Wikipedia’s Rubinstein bargaining model but no further.