Meet Athena Aktipis--evolutionist, mother of two, and salsa dance instructor in her spare time. Perhaps it was the dancer in Athena that caused her to teach multilevel selection by having the students get up and move.

Each student is given a wooden stick with an A (for Altruistic) or S (for Selfish) written on one end. Information on the blackboard tells them that altruists give three fitness units to their partners at a cost of one fitness unit to themselves. Selfish individuals receive without giving. Then they are instructed to move around the room and find a partner at random without revealing their identities. At the count of three, they reveal who they are and write their gains and losses on a 3x5 card. After repeating the process of pair formation and social interaction several times, the students total their score and take their seats for a few minutes of instruction.

I have started to use Athena's method in my classes and it is wonderful to see how the students come alive when they are allowed to move. Before they were fighting to stay awake in their chairs, but now a party atmosphere develops as they find their partners. Groans and cheers erupt when they present their sticks to each other. Altruists experience the joy of receiving, in addition to giving, when they are paired with another altruist, resulting in a net gain of +2, but they experience the agony of betrayal when paired with a selfish individual, receiving a loss of -1. Selfish individuals score big when paired with an altruist, for a gain of +3, but it's a drag to encounter another S, resulting in a 0.

Athena's method enables the students to understand the concepts of within- and between- group selection more vividly than from a dry lecture. Altruism is locally disadvantageous, which I have called "the original problem" throughout the T&R series. The students experience this for themselves when S trounces A in every mixed pair. But altruism is successful at a larger scale, as the students also experience when AA pairs produce a net gain of 4, compared to only 2 for AS pairs and nothing for SS pairs. What evolves in the total population depends upon the net effect of the opposing selection pressures. With random pair formation, within-group selection trumps between-group selection. The average S accumulates more points than the average A and wins the Darwinian contest. A message to the wise: beware of your partner in anonymous social interactions.

Now the exercise is repeated with a single twist. Partners are allowed to stay together by mutual consent. Only students on the rebound from unhappy unions must seek new partners and they can't break up happy couples. After several rounds of play, most of the altruists are smiling and most of the selfish individuals are looking pretty grim. AA pairs elect to stay together while AS and SS pairs break apart. After several rounds, most of the A's have found each other, forcing the S's to interact by default. Within-group selection still favors selfishness and between-group selection still favors altruism. Only the balance between levels of selection has changed in the second example, by concentrating individuals into AA and SS pairs and gradually eliminating the AS pairs. In her scientific research, Athena creates theoretical models similar to this version of the exercise, which she calls "walk-away" models.

In a third version of the exercise, each student is given both an A and S stick to present as they see fit and several interactions take place within each pairing. Now the mind games begin, as some students try their luck with the selfish option, typically resulting in retaliation from their partner during the next round of play. No matter how complicated the strategy of employing A and S, however, the basic logic of multilevel selection applies. Selfishness is the unbeatable strategy within each pair; altruists never beat their own partner; they can only lose or draw. It is necessary to increase the scale of comparison to find fitness differences weighing in favor of altruism.

I have created a fourth version of the exercise to illustrate the principle of kin selection, which begins by forming the students into groups of six. Two members of each group are designated as Mom and Dad, who are about to have four children. Mom and Dad are given two sticks each, which might be 2 A's, one A and one S, or two S's. These are their two chromosomes, since it is necessary to include sexual reproduction to illustrate kin selection. The A allele is assumed to be dominant, so AA and AS genotypes behave altruistically and SS genotypes behave selfishly.

Now Mom and Dad proceed to have four kids by randomly showing one of their sticks to each of their four children. It's not as much fun as having sex, but it will have to do! In this fashion, each kid obtains their genotype from their parents, just like Mendel's peas. For example, if Mom is SS and Dad is AS, then all four kids get an S from Mom, while half (on average) get an A from Dad and the other half get a S from Dad. When the kids start interacting with each other, half will behave altruistically (AS) and half will be selfish (SS). The composition of each group of siblings will depend upon the genotypes of their parents in similar fashion.

What is the result of this exercise? When the siblings interact with each other, the basic logic of multilevel selection remains unchanged. Mixed groups are still formed (as in the above example) and selfishness still wins within each mixed group. The new twist that is introduced with kin selection is in the way that the groups are formed. Sibling groups are formed through the funnel of their parents' genes, which increases the likelihood of altruist-altruist and selfish-selfish pairings and decreases the likelihood of altruist-selfish pairings, even though they still occur. This way of forming groups has the same effect as the second version of the exercise, when individuals could select their own partners. Perhaps this explains why we are so nice to both family and friends. In both cases, altruism evolves because between-group selection trumps within-group selection.

Athena's teaching method gives students a visceral feel for the generality of multilevel selection. Social interactions almost always take place within groups of individuals that are small compared to the total population. The minimum group size is 2, but everything that my students learned in pairs applies to larger groups. No matter how the groups are formed--at random, on the basis of experience, or through the funnel of genetic relatedness--no matter how flexible the choice of behaviors--altruism is locally disadvantageous and requires higher-level selection to evolve. It doesn't matter whether you call it group selection, kin selection, reciprocity, game theory, selfish gene theory, or anything else. All evolutionary theories of social behavior include the original problem and solve the problem only by identifying factors that enable between-group selection to overcome within-group selection. As Ed Wilson and I concluded at the end of our review article titled "Rethinking the Theoretical Foundation of Sociobiology", "Selfishness beats altruism within groups. Altruistic groups beat selfish groups. Everything else is commentary."

William D. Hamilton, the legendary founder of inclusive fitness theory (dubbed kin selection by John Maynard Smith), required several years and lots of math to reach the same conclusion.