Results and discussion
1. Cultural drive in an undivided population with ‘Machiavellian’ culture. First, we explored the dynamics of the simulated population under the following simple conditions: all population is a single group (no between-group competition), only TrE and Useless memes are allowed, social learning is limited only by MC, brain volume = 20 + MC (which means that parents have to spend two additional resources for each additional unit of the child’s memory capacity), initial MC = 0, MC gene is the only gene that can mutate and therefore evolve, LE=1 and does not evolve (perfect congenital copying ability), TE = 0 (see Suplementary Table 1 for further details).
Under such parameters, population stabilizes at 450-500 individuals, a new meme is invented approximately each 5-6 years. Population is quite viable from the start and quickly reaches the bearing capacity of the environment. There is no need for change, but the changes are forced by within-group competition and ‘cultural drive’. As soon as any individual invents and remembers an efficient TrE meme, she is at advantage. One needs MC to learn the meme, and so selection for MC starts. The process is autocatalytic: the higher is the average MC, the higher is the probability of preservation and spread of newly invented memes. The larger is the number of useful memes in the meme pool, the more beneficial it is to have large MC. As group members become more sophisticated tricksters, it becomes harder for individuals with low MC (and small brain) to survive and reproduce in such a group.
The runaway brain-culture coevolution forced by cultural drive is reminiscent of Baron Munchausen pulling himself out of a swamp by his hair: there are no external incentives and no reason to change, but the growing brain and developing culture continue to push each other further and further.
The initial stages (first 1500 years) of brain-culture coevolution under these parameters are shown in fig. 2. At first, there is a ‘dormant phase’: no memes are preserved because individuals do not have enough MC. Initially, MC=0, but MC gene mutates, and mutant individuals with MC>0 appear. As long as there are no useful memes in the meme pool, MC is a slightly deleterious trait (more resources are required to produce offspring). Thus the average MC tends to mutation-selection equilibrium which in this case corresponds to MCavg≈0.3. Average size of invented memes is 1, but variance is high. So sooner or later a small enough meme is invented by an individual brainy enough to remember it. Efficient TrE meme can promote selection for MC and thus initiate the ‘cultural drive’.
Useless memes cannot initiate cultural drive. Moreover, if the first meme to spread happens to be a small but efficicent Useless meme, then large MC may become considerably more deleterious, and mutation-selection balance may shift to a lower equilibrium value of MC. This can prolong the ‘dormant phase’. As the cultural drive proceeds, Useless memes infiltrate the meme pool (in the example shown in fig.2, three Useless memes have spread, marked by arrows). However, under the current set of parameters parasitic memes cannot stop cultural drive. Littering of memory by Useless memes results in continued brain/memory expansion so that it can accommodate TrE memes despite being littered.
Further events (up to the year 70000) are shown in Fig. 3.
Cultural richness (total number of unique memes in the meme pool) follows an S-shaped curve. Average number of memes known by an individual decelerates somewhat faster because this parameter is limited by lifespan and learning speed. Average individual tends to know only about 4% of the total knowledge of the population.
Average brain volume grows driven by more and more efficient Machiavellian culture, but then it stops growing and even declines. The limit to growth is explained by the fact that at some point the benefit from the ability to learn more TrE memes stops to exceed the cost of further brain expansion. The cost grows linearly with brain volume (the larger the brain, the more resources are spent by parents for each child). The benefit, however, increases with deceleration, because it takes time to fill a large memory. As culture and brain expand, the ‘juvenile’ period of incomplete use of memory becomes longer. Individuals start to fully exploit their large memory at progressively later age, and the benefit from further brain/memory expansion decreases. At some point the benefit stops to exceed the cost, and the brain stops growing.
The reasons for further decline in brain volume are more subtle. They stem from the gradual decrease in the average size of memes stored in memory (fig. 3), a process observed under virtually any parameters, unless we make the sizes of all invented memes equal. This process , which we call ‘meme simplification’, is driven by meme selection: smaller memes spread faster irrespective of their phenotypic effects, because there are always individuals in the population whose free MC is insufficient to learn a large meme, but is still large enough to learn a small meme. Meme simplification was observed by Gavrilets and Vose in their model (Gavrilets, Vose, 2006) and is clearly reproduced in our study.
For the first ~15000 years the average size of the memes stored in individuals’ memories was increasing (fig. 3). This is because meme’s size is positively correlated with its efficiency, and average MC was growing rapidly, thus making it possible to remember larger (and, on average, more useful) TrE memes. Later, however, selection for smaller memes led to meme simplification. Accumulation of simple (small) memes in culture makes large MC less beneficial, because learning takes time, and at some point individual lifespans become insufficient to fill large memory with progressively smaller memes. The efficiency of memory usage decreases, and the brain starts to shrink.
The integral efficiency of Machiavellian culture (average phenotypic TrE of individuals, fig. 3) continues to grow despite the decrease in MC, because memes not only become smaller: they also become relatively more efficicent (efficiency-to-size ratio increases) due to the interplay of meme and individual selection.
Interestingly, the proportion of memory occupied by Useless memes decreases only slightly with time, and the proportion occupied by TrE memes does not increase (fig. 4, bottom right diagram), although TrE memes are highly beneficial and Useless memes are slightly deleterious. This reflects the fact that individual selection has only a very limited ability to remove maladaptive memes from the culture (Enquist, Ghirlanda, 2007) or, more generally, to control the content of the meme pool. This is because memes, unlike genes, spread horizontally. If a maladaptive meme does not kill its host immediately, which Uselless memes do not, the host has time to disseminate the meme. Conversely, between-group competition and group selection can purge the culture very efficiently (see below).
The evolution of the simulated population can be divided into three stages (observed under other sets of parameters as well).
1. Start of the runaway brain-culture coevolution: ‘rough’ culture stimulates rapid brain expansion. At first, the brain starts to expand to accommodate the first small but efficient memes. As the brain grows, the average size of the memes also grows, while their efficiency-to-size ratio decreases (‘extensive’ development of culture). The culture formed during this stage is ‘rough’: it consists of large memes with low efficiency-to-size ratio. It is this culture, however, that stimulates the fastest expansion of the brain.
2. The potential for ‘extensive’ development is exhausted. Meme selection results in gradual meme simplification. As memes become smaller, brain expansion slows down and stops; this results in further selection for smaller memes. The integral efficiency of Machiavellian culture continues to increase, as seen from the increase in average phenotypic TrE of individuals. The efficiency-to-size ratio of the memes starts to increase again.
3. ‘Intensive’ cultural development. Further meme simplification promotes the decrease in brain volume. ‘Childhood’ (the period needed to fill one’s memory with memes) becomes longer. The integral efficiency of culture is still growing; the culture becomes more ‘sophisticated’.
Perhaps it is not too far-fetched to note that the three stages are somewhat reminiscent of the Early, Middle and Late Paleolithic, although there are prominent differences as well. Most importantly, in our model the cultural development is generally decelerating (rather than accelerating as in the real history of Homo ), and the relative durations of the stages are reversed. We believe that the descrepancies stem from the fact that the simulated culture is not cumulative: memes cannot be modified, improved or used as the basis for further development. Modeling cumulative culture is a task beyond the scope of the current study.
2. Cultural drive in a population consisting of competing groups; ‘Machiavellian’ culture. Next, we asked how between-group competition (BGC) influences the brain-culture coevolution. We used the same set of parameters with one exception: we varied the value of G (maximum group size), which was initially set to 800 so that the population always consisted of a single group. We tried two other values: 40 and 15, which correspond to approximately 20-25 and 55-65 groups in the population, respectively. This results in moderate to strong BGC and group selection.
The results are shown in fig. 4. BGC alters the pattern of brain-culture coevolution. The outcome of BGC generally depends on cultural differences between groups (cultural group selection) and on brain volume (large brain hinders reproduction and group propagation). In this case, culture per se is useless for competition with other groups, because only TrE and Useless memes are allowed (TrE memes only influence the distribution of resources within the group and have no significant impact on group competitive ability and propagation). Still TrE memes initiate cultural drive. In a way, Machiavellian culture harms the group by enforcing brain expansion which, in turn, results in reduced fecundity. Those groups that can somehow curb cultural drive will be at advantage. As seen from fig. 4, what really happens is that BGC promotes stronger selection for smaller memes and effective purging of culture from Useless memes. Both processes help the brain to remain small despite the cultural drive.
BGC results in lower cultural richness (fig. 4), mostly because of cultural drift: groups compete, often die out (with their culture), or split and propagate. This results in lower overall meme diversity. However, average individual knows approximately the same number of memes as in the previous case (without BGC). Overall, the culture is less diverse and more stereotypic.
BGC also results in lower MC and brain volume. When there is no BGC (G =800), only individual selection is working against brain expansion. When BGC is present, individual selection is aided by group selection. Consequently, the limitations on brain expansion become more stringent. This, in turn, results in stronger selection for smaller memes, more radical ‘meme simplification’ and higher efficiency-to-size ratio for TrE memes. Smaller memes, once again, reduce selection for larger brains. We call this feedback mechanism ‘the vicious circle of meme simplification’ (fig. 5).
The integral efficiency of Machiavellian culture is, somewhat unexpectedly, even higher than without BGC. This is because TrE memes which accumulate in the meme pool are extremely small and efficient, and Useless memes are effectively removed. BGC cannot stop ‘selfish’ brain-culture coevolution driven by the TrE memes, but it is fully capable of removing Useless memes from the culture. Groups with additional costly MC filled by maladaptive memes loose the competition. Expectedly, such efficient cultural group selection is possible only if between-group migration rate is comparatively low (in the simulations discussed so far, individuals move to another group with probability 0.001 per year; this means that only 2-3% of individuals migrate in their lifetime). The higher the migration rate, the more similar is the pattern of brain-culture coevolution to that observed in the absence of BGC (see below).
3. ‘Machiavellian’ culture vs. ‘cooperative’ culture. Next, we simulated brain-culture coevolution in a population with ‘cooperative’ (rather than ‘Machiavellian’) culture. The same parameters were used as before, with the only exception that ‘Hunting efficiency’ (HE) memes were allowed instead of TrE memes. The parameters for HE memes (mean efficiency, standard deviation of efficiency, C, R [see Supplementary Table 1 for explanations]) were 4, 6, 0.25, 2. This means that an average HE meme increases the phenotypic value of HE by 4 units, has size 1, there is a weak positive correlation between the efficiency and size, but the variance of both values is high. The situation is thus completely symmetrical to the previous one, except that TrE memes are beneficial for the individual but useless for the group, while HE memes are beneficial for the group rather than for the individual. The cost of having larger brains is paid by the individual, while the benefit from remembering more HE memes is shared between all group members.
The results are shown in fig. 6 (middle row of diagrams). For comparison, simulation results for Machiavellian culture (discussed above) are shown in the top row of diagrams.
When brain expansion is driven by cooperative culture, the intensity of BGC is the major factor promoting cultural drive. Contrary to Machiavellian culture that provides maximum brain expansion under minimal BGC, cooperative culture results in larger brains under strong BGC. If BGC is absent (G =800, all population is a single group) or low (G =300, population consists of 3-5 groups, data not shown) cooperative culture fails to initiate cultural drive, and the brain remains small. Thus Machiavellian culture appears to be able to initiate cultural drive under a wider range of BGC levels than cooperative culture.
If BGC is strong (G =15), cooperative cultural drive is very efficient, and brain volume reaches even higher values than in the case of Machiavellian culture and no BGC. Moreover, under strong BGC and cooperative culture, both meme simplification and decrease in brain size are less pronounced and start later. This is because every HE meme, regardless of its size, is very important for group survival when competition with other groups is high. Cultural group selection strongly favours groups with the most efficient cooperative (hunting) culture. The resulting culture is ‘rough’ and uniform on the population scale, but very efficient.
BGC appears to be a powerful mechanism to prevent the spread of the Useless memes. This is true for cooperative culture as well as for Machiavellian culture. Strong BGC generally results in a uniform, efficient culture with a minimum amount of unnecessary elements.
The comparison of the top and the middle rows of diagrams in fig. 6 shows that the average meme size covaries with the average brain volume (MC). This is apparently because smaller brains (or stronger constraints on brain expansion) promote stronger selection for smaller memes, and saturation of culture with smaller memes results in weaker selection for larger brains (fig. 5).
4. Complex culture vs. specialized culture. Next, we simulated brain-culture coevolution in a population with a complex culture, that is, with all three types of memes allowed (TrE, HE, Useless). All parameters were as in previous simulations, except that creativity was set to 0.0004, in order to retain the fixed rate of invention of memes of each type (0.000133 per individual per year, see Supplementary Table 1).
The results are shown in fig. 6 (bottom row of diagrams). The complex culture appears to be a more powerful driver of the runaway brain-culture coevolution than both types of specialized culture discussed above. The differences are as follows:
a) The complex culture promotes strong cultural development and brain expansion at all levels of BGC (contrary to Machiavellian culture which works best at low levels of BGC, and to cooperative culture which initiates cultural drive only at sufficiently high levels of BGC);
b) The complex culture results in a slightly richer meme pool and slightly lower levels of memory clogging by the Useless memes. Under strong BGC, ‘cooperative’ (group-beneficial) HE memes outcompete ‘selfish’ (individually beneficial) TrE memes (fig. 7, bottom row of diagrams).
c) The complex culture makes it possible for some hunting skills to develop even when HE memes are not supported by group selection. E.g., the average phenotypic value of HE under complex culture and G =800 is much higher than under specialized cooperative (hunting) culture and G =800. When G =800, there is no BGC and the HE memes are virtually useless, because the initial (congenital) phenotypic value of HE is high enough for the individuals to survive and reproduce. The population quickly grows up to the bearing capacity of the environment, after which all available resources are effectively extracted and used. Thus in the absence of BGC the HE memes provide no benefits to either individuals or the group. This is why cooperative culture is unable to initiate runaway brain-culture coevolution in the absence of BGC. However, when G =800 and the culture is complex, cultural drive is initiated by the TrE memes, the brain and MC expand, and the HE memes begin to spread as ‘parasitic’ memes despite their uselessness under the circumstances.
d) The complex culture prevents the extreme meme simplification.
The dynamics of memory loading (fig. 7) shows that cultural group selection (which is efficient when BGC is high) is capable of shaping the content of the meme pool, while individual selection is not. When G =800, the proportions of the meme types stored in memory are similar to the proportions of meme types invented, whereas under G =15 cooperative HE memes prevail in the meme pool.
The advantages of the complex culture discussed above imply that the chances for a large-scale runaway brain-culture coevolution may improve when the population finds itself in a situation where the individuals have the opportunity to invent many different, cognitively demanding and highly beneficial memes of different types, e.g. cooperative andMachiavellian. In this case, cultural drive is possible irrespective of the intensity of BGC.
This situation is conceivable for early Homo. Chances to invent valuable and cognitively demanding ‘cooperative’ memes could have increased due to changes in foraging behaviour and new feeding strategies, e.g., cooperative scavenging or hunting for large prey in savannah habitats (Rogers et al., 1994; Braun et al., 2010; Patterson et al., 2019). Early hominin scavangers and hunters are thought to have relied heavily on within-group cooperation in order to effectively compete with large carnivores and other hominins (Flinn et al., 2005; Moll, Tomasello, 2007; Bickerton, Szathmáry, 2011; Gavrilets, 2015). Stone tool production and use is an example of behaviour which probably was highly beneficial for the group because fast and effective butchering of large carcasses, along with other types of cooperative behaviour, could have been essential for avoiding direct aggressive encounters with stronger competitors and predators (Rose, Marshall, 1996; Plummer, 2004). At the same time, even the production of seemingly simple and primitive Oldowan tools appears to be cognitively demanding and requiring high-fidelity social learning (Morgan et al., 2015). We suggest that socially transmitted skills needed for stone tool production and use by early Homo may be comparable with the ‘Hunting efficiency’ memes in our simulation. Confrontational scavenging, hunting and butchering were most probably collective endeavors whose success benefited the group rather than an individual hunter or butcherer.
Chances to invent highly beneficial and cognitively demanding ‘Machiavellian’ memes could have increased due to the proposed evolutionary shift towards lower within-group aggression, social monogamy, increased social conformity, parental care and paternal investment in offspring (Lovejoy, 2009; Stanyon, Bigoni, 2014; Raghanti et al., 2018). In a society where direct physical aggression against group memebers is not encouraged, ‘Machiavellian’ skills and tricks may become the main way to achieve higher status and reproductive success (Humphrey, 1976; Byrne, Whiten, 1988).
5. High-fidelity, costly social learning is a powerful driver of brain expansion. Selection for high-fidelity social learning is the cornerstone of the ‘cultural drive’ hypothesis (Lewis, Laland, 2012; Laland, 2017). In the simulation experiments described above, social learning was limited by memory capacity which was costly in terms of brain expansion (each unit of MC required one additional unit of brain volume). This cost is nearly optimal for achieving the largest brain size: under the current parameters, it is not possible to achieve much larger brain volumes by making it higher or lower. If the cost is lower, the extent of cultural development ultimately will be similar (limited in the long term primarily by learning speed and lifespan), but the brain will be smaller. With higher cost, the ‘vicious circle of meme simplification’ will be more intense, the average meme smaller, and selection for larger brains weaker. The resulting culture will again be similar, and the brain volume smaller.
In an attempt to find additional factors promoting brain expansion, we varied the parameter LE (‘learning efficiency’) which determines the probability of success when an individual with sufficient MC attempts to learn a meme. In the experiments discussed above, LE was set to maximum (LE=1), did not evolve, and did not impose any costs. In the next experiment, we set the initial value of LE to 0, made it very costly (30 units of brain volume per unit of LE), and allowed it to evolve genetically (LE gene mutation rate 0.04, mean mutation effect 0, standard deviation 0.1). All other parameters were left unchanged. This means that brain-culture coevolution was now limited not only by the costly memory capacity, but also by the very costly learning efficiency. In other words, much higher price must be paid now for the same level of cultural development.
This is different from just making MC more costly, because there is one major difference between MC and LE as the limiting factors of brain-culture coevolution in our model. Costly MC limits the spread of large (complex) memes to a much larger extent than the spread of smaller memes, thus providing selective advantage to simple memes and facilitating ‘the vicious circle of meme simplification’. Conversely, costly LE limits the spread of all memes irrespective of their size (complexity). Thus costly LE is not expected to promote meme simplification. Of course, LE can also be programmed to select for smaller memes (as was done, e.g., by Gavrilets and Vose [Gavrilets, Vose, 2006]), but we aimed to explore a complexity-insensitive limiting factor (LE) along with a comlexity-sensitive one (MC).
Rather unexpectedly, we found that very costly LE does not prevent the runaway brain-culture coevolution. Under some combinations of parameters ([complex culture, G =15] and [cooperative culture, G =15]), average LE evolves up to 0.89 – 0.91 within 70000 years despite the fact that the cost of such evolution is 26.7 – 27.3 additional units of brain volume. The results are summarized in fig. 8 (compare with fig. 6 to see the effects of evolvable, costly LE; note different scales on the brain volume diagrams).
Costly, evolvable LE (fig. 8) results in cultural development similar to that observed under free, fixed LE (fig. 6). Runaway brain-culture coevolution starts under all combinations of parameters except [cooperative culture, G =800] and [Machiavellian culture, G =15]; under [Machiavellian culture, G =40] it starts in approximately half of model runs. Total cultural richness is somewhat lower than in the case of free, fixed LE, but the integral efficiency of the resulting culture (phenotypic values of TrE and HE) are comparable. Meme simplification is slightly less pronounced (average meme size is larger). Parasitic memes are more efficiently removed from the meme pool. The most striking differences are in the brain volume: when LE is costly and evolvable, brain volume reaches much higher values, especially if the culture is cooperative or complex and between-group competition is high.
Why does the costly learning efficiency result in evolution of larger brains than the costly memory capacity? The apparent reason is that LE is insensitive to meme complexity, and so the constraints on LE do not result in the ‘vicious circle of meme simplification”, whereas the constraints on MC do.
The results imply that the runaway brain-culture coevolution can be strongly facilitated by the development of mechanisms of social learning that are (1) costly in terms of brain expansion or structural optimization, and (2) tolerant to the complexity of knowledge. In other words, these mechanisms must be neurologically demanding (e.g., rely on sophisticated neuronal circuits), but they must make it possible to transfer complex information relatively easily. Human language seems to fit this definition accurately.
TribeSim program allows to simulate genetic and cultural evolution of LE and TE (teaching efficiency). Evolution of high-fidelity teaching is thought by some scholars to be essential for human evolution (Morgan et al., 2015; Laland, 2017). We found that LE can evolve both genetically and culturally (by propagation of the LE memes) with similar success. Interestingly, TE ‘prefers’ to evolve culturally rather than genetically. This is because TE memes have viral properties: when an individual learns a TE meme, she becomes a more efficient machine for meme dissemination. In other words, TE memes are the only memes that help themselves to propagate. When TE is allowed to evolve culturally, TE memes tend to occupy disproportionally large portions of memory (fig. 9) . We tentatively suggest that this may have some relation to the fact that human ability to acquire language appears to be mostly congenital (analogous to the LE gene in our model), while the language itself is learned. By learning words and grammar, people acquire the ability to transfer their knowledge to others, that is, to teach; some components of language thus can be regarded as analogs of the TE memes in our simulation.
6. Additional factors that affect the dynamics of brain-culture coevolution. Finally, we tested the effects of four additional factors: population size, meme invention rate, life span, and between-group migration (fig. 10).
Larger population size enhances brain-culture coevolution: the values of brain volume, learning efficiency, hunting efficiency, cultural richness were higher after 50000 – 70000 years of evolution (fig. 10).
The effect of higher creativity (higher rate of meme invention) is much smaller. This means that the positive effect of larger population is not explained by higher number of potential inventors, but rather by higher number of brains that can store and disseminate knowledge.
Increased life span resulted in a much more powerful runaway brain-culture coevolution, apparently because longer-lived individuals are more efficient ‘machines for meme accumulation and dissemination’. Doubling of lifespan resulted in only 20% increase of population size, because the latter is limited primarily by resources (R ). Lower death rate automatically results in lower birth rate, because it becomes more difficult for the individuals to accumulate sufficient resources for reproduction.
Absence of between-group migration (isolated groups) generally hinders brain-culture coevolution, especially the development of Machiavellian culture is compromised (average phenotypic TrE remains low). Conversely, higher migration rate is good for Machiavellian culture, but does not support the development of efficient cooperative culture (HE remains low). This emphasizes the advantage of complex culture, which can ensure powerful brain-culture coevolution at different levels of between-group migration as well as at varying levels of between-group competition (see above).