The most useful sentence in Dr Daniel Bonevac’s Game Theory course for me is a short one:
A stable outcome and a good outcome are not the same thing.
Two people can each be doing the best they can, given what the other is doing, and still be stuck somewhere both of them hate.
Nobody is being stupid. Nobody is being cruel. Each is making the best available move, and the best available move for each produces a result neither wanted.
Game theorists call that a Nash Equilibrium, after the mathematician John Nash, and it is the idea I took out of the course.
I should say plainly that it is nearly all I took out of it.
I finished Dr Bonevac’s eight lectures on Peterson Academy without getting excited about the subject. Technical material interests me less than philosophy and human nature, and I would rather say so than pretend I was gripped.
I enjoyed the first lecture the most. It talked about Nash, which brought back memories of watching and later re-watching the film A Beautiful Mind.
He talks about other mathematicians, many of whom also had psychological problems. The Nash equilibrium, which is fundamental to game theory, opened my eyes to things I had never thought about before.
As the course progressed into more and more technical ground, I noticed that I was starting to lose interest. I picked up some interesting things from each of the eight lectures, but they were mostly related to human behaviour and making choices.
What the course covers
Dr Bonevac teaches game theory as the study of situations where your best move depends on someone else’s move, and theirs depends on yours.
Lecture 1 sets out the vocabulary: players, actions, consequences and preferences, and what it means for wants to be coherent.
Lectures 2 to 4 sort situations into three families.
In a coordination game you win by matching, as with driving on the same side of the road.
In a cooperation game everyone gains if they coordinate and each is tempted to defect.
In a competitive game one side’s gain is the other’s loss.
Lecture 5 covers mixed strategies, which means deliberate randomness.
Lecture 6 asks what changes when you meet the same person again and again.
Lecture 7 covers games played in turns.
Lecture 8 asks what happens when a whole group has to choose, and ends on a result by the economist Kenneth Arrow which says that a perfect voting system cannot exist.
The idea I kept
Take Bonevac’s caveman. He meets a stranger and can fight or seek peace.
He ranks the outcomes like this:
Best, he fights and the stranger does not
Next, both seek peace; then both fight
Worst, the stranger fights and he does not.
Work it through.
If the stranger seeks peace, fighting is better.
If the stranger fights, fighting is better.
So fighting wins whatever the stranger does, and the stranger reasons the same way.
Both fight, which is the third-best outcome for each, when the second-best, mutual peace, was available the whole time.
That is a Nash equilibrium: neither can do better by changing alone. It’s also a disaster, and it explains why telling the two of them to try harder doesn’t work. If one stops fighting while the other carries on, the peaceful one loses everything.
What the course establishes here is narrow.
In the prisoner’s dilemma each prisoner does better by testifying whatever the partner does, so both testify and both do worse than if they had stayed silent. I take that to mean that good intentions and clear heads don’t rescue a bad equilibrium.
The way out is to change what each side gains from defecting: a rule, a fine, a reputation that follows you. My bet is that many of the stuck situations people blame on character are this shape.
In my essay on The Elephant in the Brain I wrote about memorising for exams and forgetting it, from the signalling side of things. Game theory says something different about the same facts.
In my own life, going through the school system, the exams were designed to test the efficiency of the educational system, and the people who created it believed that structuring exams that way was the best and most reasonable approach.
I understood how the system works and was reasonably doing what I needed to do to get the best possible grades, which meant memorising everything word for word, passing the exam and moving on to the next one.
The outcome was bad for both parties.
The educational body was measuring the wrong thing and was rewarding me with good grades for doing something ineffective in the real world, without thinking about the long-term consequences of memorising rather than understanding and actually using the knowledge.
I believed I needed the school to give me good grades so I could use them as a CV in order to get a good job in the future, and the school system needed me to do what I was doing so they could measure how well the system works.
The second idea: the future changes the game
The other lecture that stayed with me is Lecture 6, on repeated games.
When playing once against a stranger, defecting is the dominant move.
Dr Bonevac says John Locke’s reply to Thomas Hobbes was that if you will keep meeting, defecting today buys a grudge that costs you tomorrow.
His example is Las Vegas. Tourists on the Strip are fleeced because nobody expects to see them ever again, while the same workers are warm and cooperative in their own neighbourhood, where they will deal with each other for years.
He calls the mechanism the footprint of the future: how large and how certain the expectation of dealing with each other again is.
In the tournaments he describes, a strategy called tit for tat, which starts by cooperating and then copies the other side’s last move, tends to win overall, although it never beats any single opponent. It wins by collecting many cooperative relationships while the exploiters win single exchanges and lose the long run.
What separates the strategies that do well, in his account, is a short list: never defect first, be legible, be willing to retaliate, be willing to forgive, and do not be envious of whoever is currently ahead.
My take is that niceness is, at least partly, a response to the length of the relationship. It explains why people change how they behave when a commitment loosens, often before anything else has changed.
Where it collides with what I already read
There is a problem with all of this, and Dr Bonevac half admits it.
Game theory starts from the assumption that a rational person has coherent, stable preferences.
In Lecture 1 he reports a researcher, Ariel Rubinstein, who found that only about fifteen percent of the people he surveyed showed no circular preferences at all.
In Lectures 4 and 7, people turn down offers that the model says they should accept.
Set that beside Jonathan Haidt’s The Righteous Mind, which argues that the rider invents reasons for choices the elephant has already made, and Daniel Kahneman’s Thinking, Fast and Slow, which maps how fast thinking produces answers before slow thinking is consulted. They do not sit comfortably with the premise of a payoff table.
(I wrote about Haidt here and about Kahneman here.)
One needs me to know what I want, the others say that much of the time I don’t.
I don’t think the books and the course can both be straightforwardly true of the same person. They can be read together if game theory is a model of the structure of a situation and not of the inside of a head.
I don’t think this makes game theory useless, but it changes what I think its payoff table describes.
I can use it to ask how incentives and constraints shape a situation without assuming that people consciously know what they want or can explain why they chose it.
The danger is treating the analyst’s payoff table as a map of the person’s mind, when it may be only a model that helps explain their behaviour. I’d trust the model more if it predicted choices across situations better than people’s own accounts of their motives did.
Where my enthusiasm stopped
It stopped at the grids. Payoff tables, the algebra of mixed strategies and the proof of Arrow’s theorem are well taught, and I could follow them, but none of it made me want to.
I don’t read that as a fault in the course. Dr Bonevac says himself in Lecture 3 that as a predictor the theory is often wrong and as a set of instructions it’s just as shaky, and that what it does well is spot the traps. It predicts better where cheating is hard to verify, such as arms control, and worse where it is easy, such as a friend turning up when they said they would.
Who should take the course
This is where I have to land somewhere.
Take Lectures 2, 3 and 6 if you want to understand why decent people end up in bad outcomes together, and why changing the payoffs works where appealing to goodwill does not. Those three are worth the time on their own.
Take all eight if you negotiate, set prices, write rules or chair meetings, because Lectures 4, 5, 7 and 8 are tools for those jobs.
If you want something to think with in ordinary life, expect two ideas and a lot of grids. That’s not a reason to skip the course. It’s a reason to know what you are buying.
I only finished the course less than a week ago, so I haven’t used any of it in real life yet, but it gave me enough to think about in regard to human nature.
For example, I could never understand why we humans still choose to go to war. Now, after finding out about the Nash equilibrium, this is much clearer to me and makes sense.
I don’t regret taking the course, as it gave me answers to questions I didn’t even know I had.
What I am less convinced by
That tit for tat “wins”.
I know it only through Bonevac’s account of tournaments and simulations, and the evidence is thinner than the sentence suggests, because those are worlds where everyone sees the last move and nobody lies about it.
Game theory models the situation and not the person. It’s how I’ve reconciled the course with Haidt and Kahneman, and I cannot yet say what evidence would show I was wrong.
Will game theory ever become something I think with?
Logic did. This has not, at least not as much.
What would settle it is whether I reach for it unprompted after the Plato course I start tomorrow, and I will say which way it went.
I sat through all eight lectures so you can decide whether to. Everything that survived is in a thirty-four page PDF: the fourteen ideas worth keeping, the five that matter if nothing else does, where the course collides with Haidt and Kahneman, a seven-day test, and 109 cards laid out to print and cut. Free, no catch.
The two ideas in this essay are two of the fourteen.
Sources & Disclaimer: Game Theory by Dr Daniel Bonevac, Peterson Academy; The Righteous Mind by Jonathan Haidt; Thinking, Fast and Slow by Daniel Kahneman. All linked inline above via affiliate links.
This post contains affiliate links. If you buy through them I may earn a small commission, at no cost to you.
The reading and the arguments are mine. I write here to find out whether I have understood what I read, because explaining something is the cheapest way to discover whether you actually understood it or only recognised it.
AI is the other half of that. I argue with it, it pushes back, and what you are reading is what survived. English is not my first language, so it also helps me produce the clean version.
Nothing goes out that I could not defend if you asked.
That distinction between stability and welfare feels crucial: an equilibrium tells us no one can improve things alone, not that the arrangement is worth preserving. Your schooling example makes the mechanism concrete—each side can meet its measure while the shared purpose gets lost. What kinds of rule changes do you think make room for a better outcome without relying on one side to sacrifice unilaterally?