Anyone who hasn't experienced the ecstasy of betrayal knows nothing about ecstasy at all.
~ Jean Genet
Trust seems to be a mystery only as long its outcomes hang in the balance. When trust is rewarded, the dilemma renews itself the next day. When trust is betrayed, a bond is broken, often never to be repaired. The give-and-take of trust and reciprocity always seems tenuous and preliminary; betrayal is more final. Why is that?
Laypeople and the psychologists who study them have long noticed that good reputations are difficult to earn but easy to lose, especially in the moral domain (Reeder & Brewer, 1979). Reciprocity after trust goes against self-interest and must be normatively regulated. All people learn and many internalize the injunctive norm that they are to give back after they have received. Trade, gift exchanges, and mutual pleasantries respect the norm of reciprocity.
When people trust, they bank on reciprocity, betting that its force will be stronger than the force of selfishness—with a sufficiently high probability. If reciprocity were assured, there would be no trust dilemma.
Suppose a trustor (the person who trusts) gives the trustee (the person who is being trusted) $10, knowing that this amount will turn into $30 in the trustee's hands, and expecting that the trustee will split the wealth 50/50 with a probability of .8. When the trustee walks away with the whole amount, the trustor, if she reasoned statistically, would simply conclude that events that happen with a probability of .2 (betrayal) happen from time to time, namely 1 time out of 5, on average. Future interactions would not be affected by this disappointing outcome. A trustor with a Bayesian bent of mind would revise the probability of reciprocity from p = .8 downward. The shift may be large enough to make future exchanges unprofitable.
I suspect that neither stable-p statistics nor Bayesian adjustable-p statistics are a good model for human reactions to betrayal. A single breach of trust is often enough to make a person withdraw from further interaction - particularly if alternative partners are available. A student of mine, Jaap Ruoff, suggested that people treat an act of betrayal as an existence proof. A single black rose refutes the theory that no rose is black. There is, in other words, a categorical difference between no betrayal and 1 betrayal, whereas the difference between 1 and 2 betrayals is only a matter of degree.
Jerker Denrell has shown that sampling tends to be truncated after negative events, which results in negatively biased impressions (Denrell & Le Mens, 2012). One of his examples is how we evaluate restaurants. If your first visit to Chez Gaulois is a disaster you will probably stay away and retain a negative memory. If, in contrast, your first visit to La Cucina is delightful, you will come back for more. What you miss is the regression effect. If Chez Gaulois was horrendous, it was probably so at least in part because of random day-to-day fluctuations in the quality of the fare. If La Cucina's food was out of this world on that day, it was probably so at least in part because of the same type of fluctuation. Few things are consistently terrific or terrible. As you go on to sample La Cucina's cooking, you will notice that on average your delight is not quite as intense as it was on the first day. Conversely, you will not notice that your horreur gaulois would have abated over time. You can't notice this because you aren't there to notice it. If you now average your impressions of the two restaurants, the result is more negative than it would be had you continued to sample both places.
The analogy with the game of trust is clear. If people use a one-betrayal stopping rule when choosing (and rejecting) partners for exchange, they will drift toward a cynical assessment of human trustworthiness. Ultimately, some will declare "Trust no one!", whereas no one will ever say "Trust everyone!"
Beyond the Popperian falsification of the hypothesis of perfect trustworthiness (Jaap's idea) and biased sampling (Jerker's idea), it ought not to be forgotten that betrayal hurts beyond the damage to one's material interests. Withdrawal after betrayal may rather be like one-trial fear conditioning. As in fear conditioning there is some resilience. Bitten once by a dog need not mean that all dogs are avoided forever. Yet, overgeneralization can occur; a person may lose the capacity to trust anyone after one too many betrayals or after a betrayal that was particularly painful.
Unlike fear conditioning, however, the emotional reaction after betrayal is partly social in the sense that it betrays (sorry) a sensitivity to reputation. Being a sucker is one thing; being a sucker in public is quite another. Even a person willing to give a traitor the benefit of the doubt has an incentive to abort contact if others are watching. A local college president once fired an otherwise respected coach because he, the president, did "not want to be finessed." Being able to terminate relations after one case of failed reciprocity communicates power and authority (until there is no one left to impress).
Power is asymmetrical. The powerful, whose trust is abused, can turn defeat into victory by using the termination of the relationship as a display of resolve. The less powerful have no such option, but then again, their humiliations are not as public. Betrayal itself has an element of power (see epigraph). To consider the possibility that betrayal triggers ecstasy is to contemplate the psyche's shadow. I am not sure what to make of X who betrays Y merely for the sake of a psychic rush. It seems sadistic. But there may be occasions when a public betrayal preserves power by instilling fear in those who are watching. It is not pleasant, but it has a certain Machiavellian logic.
Denrell, J., & Le Mens, G. (2012). Social judgment from adaptive samples. In J. I. Krueger (Ed.), Social judgment and decision making (pp. 151-169). New York, NY: Psychology Press.
Reeder, G. D., & Brewer, M. B. (1979). A schematic model of dispositional attribution in interpersonal perception. Psychological Review, 86, 61-79.