Games, culture, and the Turing test (Part II)
March 17, 2013 11 Comments
To most evolutionary game theorists, Henrich is know not for how culture shapes brain development but how behavior in games and concepts of fairness vary across cultures. Henrich et al. (2001) studied the behavior of people from 15 small-scale societies in the prototypical test of fairness: the ultimatum game. They showed a great variability in how fairness is conceived and what operationalist results the conceptions produce across the societies they studied.
In general, the ‘universals’ that researchers learnt from studying western university students were not very universal. The groups studied fell into four categories:
- Three foraging societies,
- Six practicing slash-and-burn horticulture,
- Four nomadic herding groups, and
- Three small-scale farming societies.
These add up to sixteen, since the Sangu of Tanzania were split into farmers and herders. In fact, in the full analysis presented in table 1, the authors consider a total of 18 groups; splitting the Hadza of Tanzania into big and small camp, and the villagers of Zimbabwe into unsettled and resettled. Henrich et al. (2001) conclude that neither homoeconomicus nor the western university student (WEIRD; see Henrich, Heine, & Norenzaya (2010) for a definition and discussion) models accurately describe any of these groups. I am not sure why I should trust this result given a complete lack of statistical analysis, small sample size, and what seems like arithmetic mistakes in the table (for instance the resettled villagers rejected 12 out of 86 offers, but the authors list the rate as 7%). However, even without a detailed statistical analysis it is clear that there is a large variance across societies, and at least some of the societies don’t match economically rational behavior or the behavior of WEIRD participants.
The ultimatum game is an interaction between two participants, one is randomly assigned to be Alice and the other is Bob. Alice is given a couple of days wage in money (either the local currency or other common units of exchange like tobacco) and can decide what proportion of it to offer to Bob. She can choose to offer as little or as much as she wants. Bob is then told what proportion Alice offered and can decide to accept or reject. If Bob accepts then the game ends and each party receives their fraction of the goods. If Bob declines then both Alice and Bob receive nothing and the game terminates. The interaction is completely anonymous and happens only once to avoid effects of reputation or direct reciprocity. In this setting, homoeconomicus would give the lowest possible offer if playing as Alice and accept any non-zero offer as Bob (any money is better than no money).
The groups that most closely match the economists’ model are the Machiguenga of Peru, Quichua of Ecuador, and small camp Hadza. They provide the lowest average offers of 26%-27%. They reject offers 5%, 15%, and 28% of the time, respectively. Only the Tsimane of Bolivia (70 interactions), Achuar of Ecuador (16 interactions), and Ache of Paraguay (51 interactions) have zero offer rejection rates. However, members of all three societies offer a sizeable initial offer, averaging 37%, 42%, and 51%, respectively. A particularly surprising group is the Lamelara of Indonesia that offered on average 58% of their goods, and still rejected 3 out of 8 offers (they also generated 4 out of 20 experimenter generated low offers, since no low offers were given by group members). This behavior is drastically different from rational, and not very close to WEIRD participants that tend to offer around 50% and reject offers below 20% about 40% to 60% of the time. If we are to narrow our lens of human behavior to that of weird participants or economic theorizing than it is easy for us to miss the big picture of the drastic variability of behavior across human cultures.
What does this mean for the economic Turing test? We cannot assume that the judge is able to decide how distinguish man from machine without also mistaking people of different cultures for machines. Without very careful selection of games, a judge can only distinguish members of its own culture from members of others. Thus, it is not a test of rationality but of conformation to social norms. I expect this flaw to extend to the traditional Turing test as well. Even if we eliminate the obvious cultural barrier of language by introducing a universal translator, I suspect that there will still be cultural norms that might force the judge to classify members of other cultures as machines. The operationalization of the Turing test has to be carefully studied with how it interacts with different cultures. More importantly, we need to question if a universal definition of intelligence is possible, or if it is inherently dependent on the culture that defines it.
What does this mean for evolutionary game theory? As an evolutionary game theorist, I often take an engineering perspective: pick a departure from objective rationality observed by the psychologists and design a simple model that reproduces this effect. The dependence of game behavior on culture means that I need to introduce a “culture knob” (either as a free or structural parameter) that can be used to tune my model to capture the variance in behavior observed across cultures. This also means that modelers must remain agnostic to the method of inheritance to allow for both genetic and cultural transmission (see Lansing & Cox (2011) for further considerations on how to use EGT when studying culture). Any conclusions or arguments for biological plausibility made from simulations must be examined carefully and compared to existing cross-cultural data. For example, it doesn’t make sense to conclude that fairness is a biologically evolved universal (Nowak, Page, & Sigmund, 2000) if we see such great variance in the concepts of fairness across different cultures of genetically similar humans.
Dunbar, R.I.M. (1998) The social brain hypothesis. Evolutionary Anthropology 6(5): 179-190. [pdf]
Dunbar, R.I.M., & Shultz, S. (2007) Evolution in the Social Brain. Science 317. [pdf]
Henrich, J., Boyd, R., Bowles, S., Camerer, C., Fehr, E., Gintis, H., & McElreath, R. (2001). In Search of Homo Economicus: Behavioral Experiments in 15 Small-Scale Societies American Economic Review, 91 (2), 73-78 DOI: 10.1257/aer.91.2.73
Henrich, J., Heine, S. J., & Norenzayan, A. (2010). The weirdest people in the world. Behavioral and Brain Sciences, 33(2-3), 61-83.
Lansing, J. S., & Cox, M. P. (2011). The Domain of the Replicators. Current Anthropology, 52(1), 105-125.
Nowak, M. A., Page, K. M., & Sigmund, K. (2000). Fairness versus reason in the ultimatum game. Science, 289(5485), 1773-1775.