The emergence of collective knowledge and cumulative culture in animals, humans and machines

In recent years, the phenomenon of cumulative cultural evolution (CCE) has become the focus of major research interest in biology, psychology and anthropology. Some researchers argue that CCE is unique to humans and underlies our extraordinary evolutionary success as a species. Others claim to have found CCE in non-human species. Yet others remain sceptical that CCE is even important for explaining human behavioural diversity and complexity. These debates are hampered by multiple and often ambiguous definitions of CCE. Here, Professor Mesoudi reviews how researchers define, use and test CCE. He identifies a core set of criteria for CCE which are both necessary and sufficient, and may be found in non-human species. He also identifies a set of extended criteria that are observed in human CCE but not, to date, in other species. Different socio-cognitive mechanisms may underlie these different criteria. He reinterprets previous theoretical models and observational and experimental studies of both human and non-human species in light of these more fine-grained criteria. Finally, Professor Mesoudi discusses key issues surrounding information, fitness and cognition. He recommends that researchers are more explicit about what components of CCE they are testing and claiming to demonstrate.

In this talk, Professor Bredeche will present how social learning can be implemented in a swarm of robots when sensing, action, and communication are performed on a local basis. From a roboticist point of view, social learning can be seen as a method to perform online distributed reinforcement learning and can be extremely useful when the environment in which the robots are deployed is not known in advance. In particular, he will show how social learning helps to diffuse individual innovations throughout the swarm, whether for exploiting unexpected environmental cues or complex collision-based physical interactions between robots. Experiments with real robot swarms will be shown to illustrate how tasks such as foraging, phototaxis, and self-aggregation can be learned, as well as how the morphology of the robots can be exploited to perform some sort of morphological computation. Beyond robotics engineering, Ihe will also show how social learning in a robot swarm can lead to the emergence of collective behaviours even in the absence of any user-defined task, through the emergence of behavioural patterns that ensure survival and social strategies. Finally, Professor Bredeche will discuss how the tools from swarm robotics can be used to simulate how mutual cooperation and cooperation with partner choice can be learned.

In this paper, Professor Tomasello and Professor O'Madagain argue that the distinctive features of human culture derive from humans' unique skills and motivations for coordinating with one another around different types of action and information. As different levels of these skills of ‘shared intentionality’ emerged over the last several hundred thousand years, human culture became characterised first by such things as collaborative activities and pedagogy based on cooperative communication, and then by such things as collaborative innovations and normatively structured pedagogy. As a kind of capstone of this trajectory, humans began to coordinate not just on joint actions and shared beliefs, but on the reasons for what we believe or how we act. Coordinating on reasons powered the kinds of extremely rapid innovation and stable cumulative cultural evolution especially characteristic of the human species in the last several tens of thousands of years.

In this talk Professor Winfield reviews a series of experiments in collective social robotics, spanning more than 10 years, with the long-term aim of building embodied models of (aspects) of cultural evolution. These experiments address the question 'how do we have culture?', by modelling the low-level processes and mechanisms of cultural evolution, with robots.

The initial experiments demonstrated the emergence of behavioural traditions in a group of social robots programmed to imitate each other’s behaviours (which we call copybots). These experiments show that the noisy (ie less than perfect fidelity) imitation that comes for free with real physical robots gives rise naturally to variation in social learning; they also show how the robots’ morphology and sensorium ‘normalises’ these behaviours. 

Professor Winfield then outlines more recent experimental work which extends the robot’s cognitive capabilities with simulation-based internal models, equipping them with a simple artificial theory of mind. With this extended capability it is proposed to explore social learning not via imitation but robot-robot storytelling (storybots), in an effort to model this very human mode of cultural transmission.

Professor Winfield concludes the talk by showing how Dennett’s Tower of Generate and Test provides a unifying framework for this work. It is hoped that this work stimulates not only discussion but suggestions for hypotheses to test with the storybots.

Human societies are collective brains. People within every society have cultural brains – brains that have evolved to selectively seek out adaptive knowledge and socially transmit solutions. Innovations emerge at a population level through the transmission of serendipitous mistakes, incremental improvements and novel recombinations. The rate of innovation through these mechanisms is a function of (1) a society’s size and interconnectedness (sociality), which affects the number of models available for learning; (2) fidelity of information transmission, which affects how much information is lost during social learning; and (3) cultural trait diversity, which affects the range of possible solutions available for recombination. In general, and perhaps surprisingly, all three levers can increase and harm innovation by creating challenges around coordination, conformity and communication. Here, Robin Schimmelpfennig and Dr Muthukrishna will focus on the ‘paradox of diversity’– that cultural trait diversity offers the largest potential for empowering innovation, but also poses difficult challenges at both an organizational and societal level. They will introduce ‘cultural evolvability’ as a framework for tackling these challenges, with implications for entrepreneurship, polarization and a nuanced understanding of the effects of diversity. This framework can guide researchers and practitioners in how to reap the benefits of diversity by reducing costs.

Human cognition is not solitary, it is shaped by collective learning and memory. Unlike swarms or herds, human social networks have diverse topologies, serving diverse modes of collective cognition and behaviour. Here, Dr Momennejad reviews research that combines network structure with psychological and neural experiments and modelling to understand how the topology of social networks shapes collective cognition. First, she reviews graph-theoretical approaches to behavioural experiments on collective memory, belief propagation and problem solving. These results show that different topologies of communication networks synchronize or integrate knowledge differently, serving diverse collective goals. Second, she discusses neuroimaging studies showing that human brains encode the topology of one's larger social network and show similar neural patterns to neural patterns of our friends and community ties (eg when watching movies). Third, she discusses cognitive similarities between learning social and non-social topologies, eg in spatial and associative learning, as well as common brain regions involved in processing social and non-social topologies. Finally, Dr Momennejad discusses recent machine learning approaches to collective communication and cooperation in multi-agent artificial networks. Combining network science with cognitive, neural and computational approaches empowers investigating how social structures shape collective cognition, which can in turn help design goal-directed social network topologies.

Various studies have investigated cognitive mechanisms underlying culture in humans and other great apes. However, the adaptive reasons for the evolution of uniquely sophisticated cumulative culture in our species remain unclear. Professor Migliano proposes that the cultural capabilities of humans are the evolutionary result of a stepwise transition from the ape-like lifestyle of earlier hominins to the foraging niche still observed in extant hunter–gatherers. Recent ethnographic, archaeological and genetic studies have provided compelling evidence that the components of the foraging niche (social egalitarianism, sexual and social division of labour, extensive co-residence and cooperation with unrelated individuals, multilocality, fluid sociality and high between-camp mobility) engendered a unique multilevel social structure where the cognitive mechanisms underlying cultural evolution (high-fidelity transmission, innovation, teaching, recombination, ratcheting) evolved as adaptations. Therefore, multilevel sociality underlies a ‘social ratchet’ or irreversible task specialization splitting the burden of cultural knowledge across individuals, which may explain why human collective intelligence is uniquely able to produce sophisticated cumulative culture. The foraging niche perspective may explain why a complex gene-culture dual inheritance system evolved uniquely in humans and interprets the cultural, morphological and genetic origins of Homo sapiens as a process of recombination of innovations appearing in differentiated but interconnected populations.

How do cultural learning mechanisms interact with artificial evolution in a population of evolving robots? This talk will provide an overview of how learning (in the form of individual learning and/or cultural learning) can augment evolutionary approaches to the joint optimisation of the body and control of a robot. The overview is grounded in a general framework for evolution which permits the interaction of artificial evolution acting on a  population with individual and cultural learning mechanisms. Professor Hart will discuss examples of variations of the general scheme of  'evolution plus learning' from a broad range of robotic systems, and reflect on how the interaction of the two paradigms influences diversity, performance, and rate of improvement. Specifically, Professor Hart will describe a novel form of cultural learning in which an individual’s ability to learn over its lifetime is influenced by drawing on a structured knowledge-store that captures historical knowledge from past generations, enabling learning to be bootstrapped. The talk will also touch on how such knowledge should be represented in order to be most easily exploited.

Patterns of social interactions at regional and local scales are increasingly being framed as underpinnings of human cultural evolution. From this perspective, the socio-spatial structure of prehistoric hunter–gatherers is key for understanding the selective pressures underlying cumulative culture. However, very little is known about the interplay between ancestral patterns of social connectivity and culture. In order to address this gap, Dr Romano focuses on the relationship between long-term changes in socio-spatial connectivity and cultural transmission during the Late Glacial and Holocene periods in Europe. In this talk, she will first introduce a network-based framework that considers multi-scalar interactions over time and space. She will then discuss different examples of cultural phenomena in which the analysis of social networks could unravel the role of social connectivity on cultural transmission. In doing so, she will discuss the challenges of reconstructing archaeological networks and the perspectives in the field. Dr Romano will finally argue that social connectivity should not be considered as an independent factor driving cumulative culture, but one that influences cultural transmission in prehistory. She hope this talk highlights the potential contribution of archaeology to the multidisciplinary debate around cumulative culture.