Claire Watson, University of Kyoto:
Cultural Complexity in Japanese macaques: Literature Survey of Potential Behavioural Traditions
A huge amount of evidence has now accumulated for traditions in nonhuman animals, particularly nonhuman primates. Behaviours seen in free-ranging Japanese macaques (Macaca fuscata) were among the first potential traditions reported for any nonhuman animal, most notably ‘sweet potato washing’ and ‘wheat placer mining’ on Koshima Island, Japan. Numerous possible cultural variants have since been reported for this species, sometimes in Japanese only. Multiple long-term research sites have been subject to decades of detailed observation including: Arashiyama, Takasakiyama, Koshima, Kinkazan and Yakushima Island. We reviewed the existing literature for behaviours reported as potential traditions.  We aim to present a comprehensive synthesis of potential cultural variants in this species. We categorised potential traditions into food (foraging/processing behaviour or food choice) and non-food related (social/communicative, comfort/hygiene or other) behaviours. We further categorised each as either directly related, indirectly related or unrelated to human influence, including provisioning. Prevalence was categorised, at the time of report, as customary, habitual or unknown within each site. We could not determine the pattern of variation across all sites for each behaviour because cross-site comparisons were mostly limited to a few populations (with exceptions: stone handling, stone throwing, non-conceptive sexual behaviour and embracing), resulting in missing ‘data-points’. This species shows a high number of potential traditions but further research is needed to reveal occurrence of multiple potential traditions, and sub-variants, across the studied populations and whether this represents complex cultural variation. Delineating traditions in wild and free-ranging monkeys will likely help us to better understand the phylogenetic origins of human culture.

Cara Evans, Max Planck Institute for the Science of Human History:
When does it pay to copy the detail? High-fidelity learning mechanisms, learning outcomes and the implications for complex culture
High-fidelity social transmission is commonly associated with enhanced mechanisms of social learning (e.g., imitation), and is considered to play a key role in the extraordinary cultural achievements of our species. Comparative studies of social learning have sought to identify which species are capable of using high-fidelity learning mechanisms, but have typically neglected consideration of whether these enhanced learning mechanisms are necessary for attaining successful learning outcomes. Thus, although it is commonly assumed that learning mechanisms associated with increased fidelity to a demonstrator’s actions or object manipulations are required for the successful transmission of instrumental culture, there is currently little evidence to support this
claim, and some evidence to the contrary. A literature search was performed for all published studies of primate social learning that were conducted in controlled laboratory settings and which presented subjects with instrumental, two-action learning goals. Data were extracted on learning mechanisms, tool-use behaviour, task complexity, and learning outcomes (i.e., success/failure in attaining the instrumental goal) for each of the N=125 experimental test groups that met the inclusion criteria (~1000 individual test subjects spanning 10 different primate species), along with information about the experimental protocol, and the test group’s demographic composition. A phylogenetic, multilevel analysis of the resulting meta-dataset suggests that high-fidelity learning mechanisms are important in achieving successful learning outcomes, but only at higher levels of learning complexity. These results support the notion that high-fidelity social transmission is necessary for the emergence and evolution of increasing cultural complexity.

Marina Bazhydai, Lancaster University:
Information transmission in two-year-old children: preference for less complex over pedagogically demonstrated actions
Social transmission of information is achieved through observation, imitation, and explicit teaching. Human capacity to engage learners in pedagogical contexts is at the core of transmission of complex human culture (Burdett, Dean, & Ronfard, 2017). Infants exhibit early receptive sensitivity to pedagogical cues (Butler & Markman, 2016; Csibra & Gergely, 2009), but less is known about developmental origins of children-directed teaching (Strauss & Ziv, 2012, Kline, 2015).
It has been argued that both pedagogical communication (direct eye contact and child-directed speech; Csibra & Gergely, 2009, 2011) and intentional but non-pedagogical communication result in learning in children (e.g., Gopnik & Schulz, 2004). In a study of child-driven transmission of information learnt in these two ways, Vredenburgh, Kushnir and Casasola (2015) showed that 2-year-olds are more likely to demonstrate an action to an adult after learning it in a pedagogical than in a non-pedagogical context. However, in addition to pedagogical cues, other factors may affect transmission, such as complexity of learnt information. In two laboratory experiments, we investigated whether manipulating action complexity as well as pedagogical cues when demonstrating an action selectively affects the likelihood of the action being shown to an ignorant adult.
In Experiment 1, 24-month-old children (N = 31) interacted with two unfamiliar adults who demonstrated two actions. One demonstrator showed a simpler action in an intentional, but nonpedagogical manner, while the other showed another, more complex action in a pedagogical manner by using explicit verbal cues (“This is how you do it!”), direct eye contact, and childdirected speech. Following demonstrations, children were equally likely to imitate both actions, but achieved the action outcome significantly more often with the simpler action. At transmission phase of the experiment, children were encouraged to teach an ignorant familiar adult who was not present during demonstrations how to play with the toy. While 79% of children showed both actions, they were significantly more likely to teach the simpler action first, even though it was presented without explicit pedagogical cues (Figure 1, t(30) = -2.68, p < 0.01), suggesting preferential transmission due to the ease of action execution rather than pedagogical demonstration. Additionally, and unexpectedly, 45% of children utilized and combined the demonstrated actions in a novel way, essentially exhibiting rudimentary tool innovation (Carr, Kendal, & Flynn, 2016), previously shown to be difficult to achieve (Cutting, Apperly, Chappell, & Beck, 2014).
In Experiment 2 (N = 31), we attempted to replicate the results by Vredenburgh et al (2015) to test whether pedagogical communication modulates information transmission in absence of other salient factors. We used the same procedure as in Exp. 1, except that action complexity was matched for both pedagogically and non-pedagogically demonstrated actions in each trial. At transmission, children were equally likely to perform either the pedagogically or nonpedagogically demonstrated action first, both in simple and complex trials, thus producing a null result (Figure 2, t(30) = .00, p = 1.0), failing to replicate the original study.
While receptivity to information presented using both intentional and pedagogical cues has been studied extensively, children’s active transmission of information following these cues is understudied. Here we investigated whether and how pedagogical communication and action complexity interact to affect information transmission.
We find no evidence for preferential transmission of pedagogically communicated actions. However, we do find evidence for preferential transmission of less complex actions. Pedagogical cues and ease of execution both enhance the saliency of an action, increasing the likelihood of children transmitting this action first. However, when pitted against each other, ease of execution wins. These results suggest that action complexity may undermine the role of explicit pedagogical cues in information transmission mechanisms in early childhood.

Bill Thompson, University of California:
A Bayesian Model of Cumulative Culture
Unlike other species, human populations naturally pool computation over generations. Our system for achieving this – cumulative culture – caches human innovations to construct a technology-rich environment that is increasingly adaptive, and increasingly peculiar: we are surrounded by technologies that are exceptionally complex and highly efficient, that almost nobody understands, and which no individual has designed (Boyd et al., 2013). How do human technologies become so effective? Why do some technologies appear to grow ever more complex without limit, while others converge on relatively simple designs? One way to explore questions like these is to develop formal models of the processes that underpin human culture (Boyd & Richerson, 1985). A large literature on the dynamics of cultural evolution is now emerging, but the principles of *cumulative* complexity in culture are only beginning to be understood (Kempe & Mesoudi, 2014). Recent work has suggested that a key priority is to establish connections between models of cultural processes and existing theories of human cognition (Heyes, 2018). We present a simple mathematical model of cumulative culture as transmission of a technology among populations of innovative Bayesian learners. By deriving a model of cumulative culture from a computational model of probabilistic inference — Bayesian learning – we extend existing Bayesian analyses of social learning (Perrault et al, 2012) and cultural transmission (Griffiths & Kalish, 2007) to the cumulative case, and provide a formal connection between properties of individual cognition, technological fitness landscapes, and the accumulation of complexity. More generally, our analysis leads to an equivalence between Bayesian cumulative culture and a widely studied optimization algorithm from artificial intelligence. This demonstrates that there is a natural computational formulation of cumulative culture as an optimization algorithm executing in human populations, with well understood connections to evolutionary processes and a large existing literature characterising it’s behaviour under varied conditions. We lay out the equivalence using the simplest possible model of cumulative culture, but show that more realistic models correspond to more sophisticated (and more efficient) versions of the algorithm. These analyses lead to a general prediction that we are in the process of testing experimentally: limitless accumulation of complexity, in the face of cognitive constraints, requires a supra-linear utility landscape for new technologies, over and above high fidelity transmission and flexible cognition.

Christian Rutz, University of St Andrews:
Cumulative technological evolution in New Caledonian crows: from simple sticks to highly-efficient, crafted hooked tools
Human technological advancement relies on the stepwise accumulation of innovations. Over time, small modifications to once-basic designs lead to increasing tool efficiency. Similar cumulative refinement may also occur in the New Caledonian crow – a species that employs a diversity of tool types for extractive foraging, including elaborately-crafted hooked tools. An important requirement for any cumulative process is the differential performance of variants, an assumption my team recently tested in a series of controlled experiments with wild-caught, temporarily-captive crows. To assess the efficiency of different tool types, we measured how quickly our subjects managed to extract prey from standardised, naturalistic foraging tasks. We found: (1) that crows acquired food several times faster when using hooked rather than non-hooked tools, regardless of tool material, prey type and extraction context (St Clair et al. 2018, Nature Ecol. Evol. 2, 441–444); and (2) that the efficiency of hooked tools increased significantly with hook depth (Sugasawa et al. 2017, Curr. Biol. 27, 3885–3890). While the mechanism of tool-type transmission (genetic or ‘cultural’) remains to be determined, these results demonstrate that tool shape strongly affects energy-intake rates in this species, highlighting a powerful driver for cumulative technological change. Based on the results of these tool-efficiency assays, and our earlier work on tool morphology (St Clair and Rutz 2013, Phil. Trans. R. Soc. B 368, 20120415) and tool-manufacture techniques (Klump et al. 2015, BMC Biology 13, 97), we can tentatively reconstruct a scenario for the cumulative evolution of crows’ stick tools: unmodified non-hooked tools (sticks sourced from leaf litter) → manufactured non-hooked tools (distal sections of branches snapped off from live vegetation) → ‘basic’ hooked tools (branches pulled off close to nodal joints) → hooked tools with well-defined, deep hooks (branches removed through controlled cuts above and below nodal joints) → complex hooked tools with crafted hooks and two additional design features (bark removed near the working end; tool shaft bent until markedly curved). Taken together, these studies establish the New Caledonian crow as an invaluable non-primate model system for investigating the conditions and processes that allow rudimentary technologies to evolve, becoming increasingly complex, efficient and diverse.


Maxime Derex, University of Exeter:
Cultural evolution without insight? An experimental investigation
The tools essential for life in even the simplest foraging societies are very complicated artefacts with multiple interacting parts made of many different materials. A central claim of cultural evolution theory is that these complex artefacts can arise though the retention of adaptive changes even though individuals do not understand why these changes are beneficial. This prediction, however, has never been properly tested and it is not clear whether cultural evolution can operate in absence of insight. Here we provide such a test by asking chains of 5 participants to improve the configuration of a wheel going down on rails. The wheel was composed of 4 spokes and 4 weights that could be moved along these spokes. Two treatments were compared. In the first one, participants had 5 trials to improve their wheel before passing their last two configurations to the next participant in the chain. The second treatment was similar except that participants could also transmit their theory about what makes the wheel going fast to the next participant. At the end of the experiment, participants’ understanding about the wheel was evaluated in isolation. In both treatments, we found that later generations produce wheels that are much faster than earlier ones. In contrast, participants’ understanding about the wheel barely changes across time. This experiment suggests that cultural evolution results in the emergence of complex solutions even in absence of insight.

Carsten Bergenholz, Aarhus University:
Cooling down atoms: How humans engage in collective, adaptive search when solving a novel and complex problem
How humans solve complex problems is a question at the core of multiple disciplines. Researchers within cultural evolution, cognitive science, psychology and management have set up many different kinds of experimental tasks to shed light on this issue. One major challenge is to create a complex task that not only consists of selecting between relatively few given options, while still allowing a mathematically tractable comparison of the quality of the different solutions provided, as well as constituting a realistic problem. We set up a task where citizen scientists had to solve a truly complex problem; how to cool down atoms. We introduce a novel remote interface that allowed 600 citizen scientists to engage in a real-time optimization of the production of Bose-Einstein condensates in a physical lab. Participants could manipulate 3 curves sampled at 90 different points equaling to a search in an interdependent search space of 270 parameters. Thus, the dimensionality of the search space was finite, albeit exceptionally large. The quality of the proposed solution was then determined via a real-time test in the lab. Quantitative studies of search behavior demonstrates that citizen scientists collectively engage in a combination of local and global search. This form of collective, adaptive search prevents premature convergence since low-performing participants explore while high-performing participants locally refine their solutions. In comparison with a computational algorithm controlled by experts, these amateur citizen scientists identified just as high-scoring solutions. Furthermore, their search was particularly useful since they explored more of the search space than the computational search. Overall, we contribute to an understanding of how humans search truly complex search spaces, and how human search propensities compare with state-of-the-art optimization algorithms. Finally, we contribute to physics and provide knowledge about how to produce Bose-Einstein condensates.

Thomas Morgan, Arizona State University:
The empirical study of cognitive and demographic complexity in cultural evolution
Cultural evolution is both a producer and product of complexity. For instance, cognitive processes, such as imitation, teaching, language and innovation, have been invoked to explain the unique complexity of human culture. Simultaneously, demographic factors, such as population size and connectedness, have also been shown to foster or limit cultural evolution.
Despite the importance of cognitive and demographic complexity, it is challenging to combine the two in single studies. The former requires the inclusion of human participants, but this typically limits experiments to small, simple networks such as transmission chains. Conversely, the latter requires mathematical models and agent-based simulations as they can be scaled up with little cost, but these rely on simulated cognition in lieu of actual human brains. To bridge this gap, I have developed an experimental automation framework, “Dallinger”, that recruits thousands of human participants over the internet and then arranges them into whatever demographic structure is requested. Participants can thus take part in experimental simulations of cultural evolution that include both the cognitive complexity of real human participants as well as the demographic complexity of realistic population sizes and structures.
To demonstrate this approach, I conducted a series of experiments investigating the gene-culture coevolution of social learning in a changing environment. Across experiments, constraining participants’ behavior to match theoretical assumptions replicates classic results from the cultural evolution literature, but when these constraints are removed the evolutionary dynamics produced do not match theoretical expectations. This discrepancy highlights the utility of inserting human brains into large-scale evolutionary simulations and I conclude that the function of human social learning is not to adapt to an unstable environment, but instead to reliably transmit cultural traits.