The model of negotiation I think better than game theory Ronald Loui Ronald Loui Ronald Loui Published May 27, 2026 + Follow I'm going to start writing longer "facebook posts" as LinkedIn articles because it seems they are indexed well by search, even if they disappear quickly on the site itself. My main aim is to bypass publication but not have it all disappear as daily blogs when my server and domain disappear -- as is inevitable when I'm gone. The advent of AI chat to explain what isn't said reduces the burden of writing, which normally is for journals, which have always been bad gatekeepers in my life. I find writing for the next generation of scholars on ideas they would not otherwise hear about to be rewarding. So how could anything be better than Nash? I love JC in the BEAUTIFUL MIND movie more than most. I was reading Nash, and loving the work, back in 1983. Grew out of that mindset within the decade. Have started saying that the Academic freshness date for game theory is approaching. I actually still like Stackelberg and Shapley from college days, but a lot of micro is based on weird data structures for modeling people and preferences. Arrow was nice to me; feel bad about dissing those simple 1-variable functions. There were actually TWO solutions to the problem that Nash proposed: a two-person encounter defined by a row-column bimatrix of utilities (you choose row, i choose column, tempo undefined). People forget that one of Nash's ideas was based on simple arithmetic. The other, equilibrium, got famous. Because mathematicians found it useful. And its flaws are well known, but it keeps on going. Some people really like to think that way (static incentive to non-deviation from serendipitously discovered fixed point). I do not. Nash #1 was faintly about fairness and power (max product and hope for linear scaling!); Nash #2 was about self-enforcement when there is no contract and no legal system to enforce contracts. My early 90s work with Diana Moore (still unpublished) on busting out of the payoff matrix setup made good hay. The most famous multiagent people in UK picked up on the ideas of (1) formalizing cheap talk, (2) pushing the Pareto frontier by being creative (involving new dimensions), and (3) changing the payoff you get for a row and column by doing problem-solving search. Between Katia Sycara's PERSUADER work and George Ferguson's dissertation, with Roger Fisher's ideas of principled negotiation (loved Prof Fisher's Soc Sci 174), you can see the idea's ontogenesis. All are still good ideas. In fact the problem-solving aspect is why i now quip that "you need more out of the negotiation if you are low IQ, because you don't know how to do more with less." But around 1998, playing with software that fell out of Fernando Tohme's visit and Anne Jump's conversations, the probabilistic model with nonstandard utility also popped out. This was later shown at Sorbonne-Pantheon Paris (JURIX) and the national Spanish AI conference CAEPIA-Donostia during invited plenary talks. And written in mathematical form in the Festschrift for advisor Henry Kyburg. But the slides recently disappeared from repositories, so let's review the idea. It's so simple. As you make proposals and so does the other side, you start lowering your aspiration. Why? Because if you don't retract proposals, your concession level keeps changing. This is an improvement from the other side's point of view. But they make concessions too, so both aspirations are falling. At any time, one has a rough idea of which row and column are the more probable places of settlement, if there be an agreement. We learned this by repeatedly playing the Anne Jump game. Also, as time passes without progress, or even with minor progress, the probability of no agreement rises. This is an empirical claim, and certainly there are pathological subclasses of encounters where people look like they won't reach agreement, then suddenly break through. Talking probability here, not possibility. With no agreement, the BATNA (best alternative to negotiated agreement) or security position (walkaway/threatened row or column) determines a person's utility. With a probability pB at time t of breakdown utility, BU, and a probability pA(i,j) at time t over every admissible (non-dominated by concessions) remaining settlement possibility (i,j), one calculates an Eu(t), expected utility at time t. If any row-column proposal has utility higher than Eu(t), it is rational to extend that as an offer. So one can drive a process of rational concession just because that Eu(t) keeps falling. It's great to watch when you program a simulator. Dynamics, not statics. That's the pessimism part of the model. But it's too easy to manipulate another just by making no proposals, staying silent. So introduce a process utility ("nonstandard" because it values HOW you got there, not WHAT you get). The more time since the other made a nontrivial proposal, the greater the extra value of breaking down. If I don't like you, i'll take a 5 cent penalty just to screw with you. I actually learned that while watching our ultra-high-IQ questionable-EQ dean try to nudge, uh, manipulate people. He was from control theory math, which I'd studied too. So i knew where his blind spot was. Humans don't like to be controlled. At Donostia, should I say San Sebastian?, a colleague from UMass asked, why not add a positive utility when agreement is found in happy circumstances? Great idea. You negotiate with your loved one, I hope you attach positive process utility to the act of agreement! That's it. That's the whole model. The rest is simulation of "laissez-faire" paths, different functions for pessimism rate and punishment rate. And showing that different starting offers can lead to breakdown before agreement or vice versa. Worked on those simulations with Lance Cai who was on loan from Yale for the summer. A few others. You know who you are, thank you (if i could list all without omission, i would, but memory is imperfect). But is it a good model? Well, I always thought too many Near East minds like the wrong models of negotiation. I always worried about Tom Schelling's argument about mutual assured destruction. So perhaps goodness is relative here. I know that if you do not feel the satisfaction of breaking down on someone who is trying to manipulate you, this isn't for you. I will say I would be happy to have an automated negotiating agent programmed to behave this way for me. And the process behaviors are certainly more natural than what you see out of game theory. I think there is exactly one famous paper on process, trajectory, and dynamic utilities in game theory, Rubinstein. There should be more. Maybe there are more now, but Tohme and I looked back in the day. So is it better than Nash? I would ask if it not as bad as Nash.