Proof from Morality (3)

Social Animals

One clue is that we see morality in other animals too.

Wild coyotes have play rules that ensure every pup is on equal footing. If one coyote can bite harder than another, it’ll hold back. If one is more dominant, it’ll send signals of submission to level or even reverse the social hierarchy. Pups that refuse to play by the rules are ostracised, to the point that they’re four times more likely to die.[115] The literature on elephants is full of heart-melting anecdotes, like the pack that deliberately slowed down to accommodate a disabled member, or the female elephant that swooped in to save another injured female from a male’s attack. Even rats can be remarkably chivalrous. A pair of them got stuck in a laboratory drain overnight. When they were found in the morning, only one was strong enough to drink or eat. This rat took a piece of food and placed it in front of the weaker one; as it nibbled on the gift, the strong one would tug it a short distance away, encouraging the weak one to crawl forward a little to resume its meal.

Can you guess where the stronger rat was tugging the food?

This does not rule out divine intervention, of course, it merely points out that any human-centred explanation of morality won’t do. Christianity, for instance, has traditionally claimed that only creatures of species Homo Sapiens Sapiens have a soul, and are thus blessed with intelligence, morality, or whatever noble trait the speaker wished. This black-and-white view fails in a world full of shades of gray.

While I can’t completely rule out the divine, I can provide a simpler explanation. In the introduction, I introduced Ockham’s Razor when deciding if oranges were gods. If I want to apply it here, I have to present a way for morality to develop that doesn’t need the help of a god. This gives us two theories for moral development and no evidence to prove either, the perfect situation to invoke Ockham and again make the god hypothesis useless.

The First Game

Let’s start with something abstract. Have you heard of the Prisoner’s Dilemma?[116]

Imagine you and a partner you’ve never worked with before team up to rob a bank. The police catch both of you, but not before you’ve stashed the loot in a safe place. They don’t have enough evidence for a conviction, so instead they split you up and press you to rat out your partner.

This puts you in a bind. Keeping quiet seems to be the smart move; so long as your partner stays quiet, both of you will be taking baths in your share of the money. But what if your partner rats you out as the ringleader? You’ll be making license plates while your partner can swim in both shares of the loot. It’d be better to rat them out instead and get your swimming trunks ready… unless they too rat you out. In that case, both of your swimming goggles will be sold at a police auction. You know your partner must be facing the same decision. What to do, what to do?

If you’re a logician, you always Rat out the other person. To understand why, I’ll summarize this game in a payoff table, with “points” instead of dollars:[117]

Your Partner’s Decision:

Stay Quiet

Rat You Out

You Earn:

They Earn:

You Earn:

They Earn:

Your Decision:

Stay Quiet

Victory Dances for All!

You’re screwed, big-time.





Rat Them Out


everybody loses





Tally up the “You Earn” portion of each column. Since the total value for “Rat Them Out” row is greater than the “Stay Quiet” row, you stand to gain the most points by Ratting. This will also protect you from the worst-case of zero points, while opening up the best-case of five. Your partner knows all this, too, and thus is likely going to Rat on you anyway. It makes perfect logical sense.

And yet when you present this scenario to non-logicians, they usually Stay Quiet. For some reason, they trusted their partner to go for the best-case scenario, even though there was no logical reason for it. Logicians were perplexed, and launched several studies trying to figure out why humans were so willing to be so forgiving.

The most infamous was started by Robert Axelrod. Instead of asking people to play this game, he created several computer program “players” and gave each a different strategy. These were treated like bacteria floating aimlessly in a dish; two programs were picked at random to compete, and the awarded points went to their type of strategy. As each strategy earned or lost points, individuals of that type were added or removed from the “dish”. This is an important change; in the scenario above, the Prisoner’s Dilemma was played for a single round, not several. In addition, individual players were also granted the ability to remember programs they interacted with before and every choice that individual picked before, but weren’t allowed to know what strategy they were interacting with.

Axelrod then did something interesting: he turned his study into a competition by asking his peers to create strategies for the digital players. There were no restrictions on program length, so long as the submissions followed the above rules. He received fourteen strategies in total, ranging from the self-explanatory “Always Rat” and “Always Stay Quiet,” to complicated versions that used advanced statistical techniques to predict the likelihood of the next outcome and act accordingly. He added a fifteenth strategy to the mix, “Pick Randomly,” and let the simulation run.

The results were a shock. The most successful strategy was the simplest non-trivial one! “Tit-For-Tat,” submitted by Anatol Rapoport, just repeated what its challenger did last round. If there was no last round, it chose to stay quiet. No other strategy, no matter how complex, could beat it.

Somewhat taken aback, Axelrod shared the results publicly and asked for more challengers. This time, he received 62 submissions. Rapoport re-submitted “Tit-For-Tat,” with no changes.

After 3 million rounds, “Tit-For-Tat” still beat all comers!

Axelrod and other researchers have re-run this simulation, with various tweaks. If a player’s choices are occasionally flipped from “Quiet” to “Rat,” or vice versa, “Tit-For-Tat” is beaten by “Generous Tit-For-Tat.” The only difference between the two is that “Generous” occasionally ignores history and plays a “Silent;” this prevents it from falling into a cycle of retribution over a flipped choice. If digital players can remember their own past choices too, an even simpler strategy called “Pavlov” can rule the roost. If the two players didn’t play the same strategy last time (say one picked “Quiet” while the other chose “Rat”), “Pavlov” plays “Rat,” and in all other cases goes with “Quiet.” The reasons why it beats “Tit-for-Tat” are complicated, but “Pavlov” seems to encourage an “ecology” of strategies that allow it to rise to the top.

It’s important to note that while these three simple strategies dominate, they rarely stand alone. The usual result of these contests is a balanced ecology, with one or two big players and a number of smaller ones surviving on the edges.

The “Tit-For-Tat” duo and “Pavlov” have only been bested once. On the 20th anniversary of his first contest, Axelrod staged another. This time the winners we some clever entries from Professor Nicholas Jennings and others at Southampton University. When these programs encountered a new opponent, they played back a pre-set series of moves. If their opponent responded with a certain pattern of choices, they began acting like “Always Rat” or “Always Silent” to them. If not, they gave that player the “Generous” or “Pavlov” treatment.

If you know what strategy your opponent will take, any game becomes trivial. If your opponent will always stay “Quiet,” the best strategy is “Always Rat,” which earns you the most points per round. If you know you’re playing against your own strategy, “Always Stay Quiet” is better since it has the highest point earnings collectively. If up against “Always Rat,” “Always Rat” will minimize the damage.

While the procedure hid the strategy of each program from all others, SU’s entries exploited a loophole. By watching the first sequence of moves, they could spot what program they were up against, as well as broadcast what strategy they were using. SU’s entries traded short-term failures for long-term gain, and this edge was enough to catapult the “Secret Handshake” strategies to the top.[118] In contrast, the “Simple Three” treat everyone equally. This is one key to their success: they make no assumptions about their opponent, and thus can’t have those assumptions turned against them. There’s a cost to be paid, however. By treating everyone equally, they cannot exploit any differences between strategies.

Had Axelrod done another round of competition, none of the “Secret Handshake”s would have cracked the top 10. Why? Their advantage was based on an assumption: every program with a fixed opening sequence is honest. Now that their initial sequence of moves is public, another competitor could write a program that impersonates these winners, only to change its behaviour after it detects one of them has taken the bait. To combat this the Handshakes would have to change between rounds, and the only legal way to pull that off in Axelrod’s contest is to resubmit new programs each time.

We’re ready to move back to the real world.

The Prisoner’s Dilemma originally featured real people. It makes a lot of sense to bring them back in and watch the results. Under the same rules and restrictions, it turns out, we imitate “Generous” and “Pavlov.” And yet few of us have heard of the Prisoner’s Dilemma, let alone studied it. This also explains why we tend to Stay Quiet during the one-round version; all three simple strategies play that on the first round.

Remember the infant experiment? The puppets were engaged in a game very similar to the Dilemma, except it was impossible for both to lose. The baby didn’t participate that time, but instead had a preview of how each player would behave. When it was entered into the game via the food bowls, it took advantage and behaved like “Tit-For-Tat” on the second round; Rat on the puppet that had Rat-ted last time, and Stay Quiet to the one that had Stayed Quiet.

Variations on the Prisoner’s Dilemma are extremely common in the real world. Should I ignore that bacterium floating towards me, and risk being attacked, or should I pre-emptively start a battle? Will I hunt in a group and share my food, or hunt alone and hoard it? I’ve spotted a distant predator; should I put myself at risk by chirping in alarm, or stay quiet and hide? To play or snub? Attack or flee? Submit or defy? Our choice in each case effects our very survival.

This is evolution’s turf. Since behaviour can be partially controlled by our genes, we have been evolving solutions to these dilemmas for billions of years. It’s remarkable how closely the artificial and natural versions match; both result in populations where the majority use “Generous” or “Pavlov,” but a minority get away with “Always Rat” and a host of other less-than-optimal strategies.

Hopefully this also sheds light on an oddity I glossed over earlier. The total points awarded for mutual Quiet are greater than Rat/Quiet because that’s a closer match to reality. I share half my genes with my parents, siblings, and children, so helping them indirectly helps my genes spread as well. I get less benefit by helping others in my species, but keeping the gene pool diverse is still a net plus.[119] Therefore, both players get a minor bonus when they co-operate. There can even be some cross-species benefit; dogs provide us companionship, hunting skills, and early warning to a mutual predator, and we’re happy to return the favour.

Both players also get a minor bonus when they harm one another. While I’m worse for the encounter, at least the other player isn’t any better off and is less likely to threaten me in future.

Game Theory and evolution can explain a lot of moral behaviour. Why are we kind to strangers? Because altruism can pay off later. Why do we rarely lie and steal? Because we may get punished for it, by the victim or a third party, and if too many people do it everyone suffers. Why do people do it anyway? Because we can get away with it in small doses. Why are we fairly consistent in our moral choices? Because the best strategies are simple enough to be bred into our very bones. No higher power is needed for this, so no higher power should be assumed.

A common critique against this approach is that it fails to provide an absolute morality. It’s true, this morality only applies to anything that evolves. But do we need a moral system which accommodates for the experiences of rocks or solar systems?

If this approach seems greedy or short-sighted, sit tight. I need to cover a gaping hole before I move on.

[115]  “The Ethical Dog,” Scientific American Mind, March 2010

[116] This is a classic problem in Game Theory, developed by Merrill Flood and Melvin Dresher back in the 1950’s.

[117]  Note that if both of you co-operate, the total reward is greater than the total reward when there’s one winner and one loser, and if both of you Rat you yourself do better than if you alone lost. As you’ll see later, this tends to be the most interesting award system, and thus most studied.

[118]  Can you guess where the worst performers of that round came from?

[119]  This is why homosexual behaviour has been found in every species we’ve studied. Why they are very unlikely to have children, non-breeders make excellent uncles and aunts and thus benefit even the greediest gene. [Future HJH: It’s ONE reason, not THE reason. Biology is complex, so it should be no surprise if multiple reasons exist, some of which are not due to adaptation.]