Exploring the science of complexity series (part 19): Concept 10 – Co-evolution

This article is part 19 of a series of articles featuring the ODI Working Paper Exploring the science of complexity: Ideas and implications for development and humanitarian efforts.

Complexity and agency – Concepts 8, 9, and 10

Certain kinds of systems are made up of individual adaptive agents acting for their own purposes, and with their own view of the situation. Such agents can be powerful in shaping the system. A special class of complex systems is made up of adaptive agents (Concept 8), which react to the system and to each other, and which may make decisions and develop strategies to influence other agents or the overall system. The ways in which these actors interact can give rise to self-organised phenomena (Concept 9). And as agents operate in a system, changes in the system and changes in the other actors can feed back, leading to co-evolution of agents and the system (Concept 10).

Concept 10 – Co-evolution

Outline of the concept

When adaptable autonomous agents or organisms interact intimately in an environment, such as in predator-prey and parasite-host relationships, they influence each other’s evolution. This effect is called co-evolution, and it is the key to understanding how all large-scale complex adaptive systems behave over the long term. Each adaptive agent in a complex system has other agents of the same and different kinds as part of its environment. As the agent adapts to its surroundings, various elements of its surroundings are adapting to it and each other. One important result of the interconnectedness of adaptive bodies is the concept of co-evolution. This means that ‘the evolution of one domain or entity is partially dependent on the evolution of other related domains or entities’ ¹.

Detailed explanation

Co-evolution brings a focus on the ‘evolution of interactions’, whereby the characteristics or tendencies of an agent may be powerfully shaped by its interactions with other agents or the wider system. Agents continuously influence and are influenced by their environment in a reciprocal fashion that changes the interacting environment and the agents themselves. To take a commonly cited example, elephants thrive on acacia trees, but the latter can only develop in the absence of the former. After a while, the elephants destroy the trees, drastically changing the wildlife that the area can sustain and even affecting the physical shape of the land. In the process, they render the area uncongenial to themselves, and they either die or move on. The land is adapting to the elephants just as they are to it². As the Maasai proverb has it: ‘cows grow trees, elephants grow grasslands’.

Kauffman³ argues that there are two distinct levels of co-evolution: interspecies and system wide. Interspecies co-evolution is widely recognised and there are numerous ecological examples. Numerous examples exist of avian and insect species with co-evolved feeding and breeding strategies that depend on the parasitic or predatory practices of other species. For example, Angraecoid orchids and African moths co-evolve because the moths are dependent on the flowers for nectar and the flowers are dependent on the moths to spread their pollen so they can reproduce. The evolutionary process has led to deep flowers and moths with long probosci. Mutualism can also be used to explain co-evolution between predator and prey species. For example, in the case of the rough-skinned newt and the common garter snake, the newts produce a potent nerve toxin that concentrates in their skin, to which garter snakes have evolved a resistance, enabling them to prey upon the newts. The co-evolving relationship between these animals has resulted in an ‘evolutionary arms race’ that has driven toxin levels in the newt to extreme levels.

System co-evolution is a large-scale process through which the interaction of one or more co-evolved species with the system results in changes so fundamental that all species in the system must adapt and the system itself change in significant ways. Human interaction with the global environment and the increasing numbers of extinctions is a commonly cited example of this phenomenon. Given the above, adaptive agents can be said to have ‘porous boundaries’, in that they are significantly shaped by their interrelations with others, with no hard division between the doer and those that are ‘done-to’ ⁴.

Work in biology on ‘fitness landscapes’ is an interesting illustration of competitive co-evolution⁵ A fitness landscape is based on the idea that the fitness of an organism is not dependent only on its intrinsic characteristics, but also on its interaction with its environment. The term ‘landscape’ comes from visualising a geographical landscape of fitness ‘peaks’, where each peak represents an adaptive solution to a problem of optimising certain kinds of benefits to the species. The ‘fitness landscape’ is most appropriately used where there is a clear single measure of the ‘fitness’ of an entity, so may not always be useful in social sciences.

However, if fitness is defined broadly as ‘optimal trade-offs’, then there is greater applicability. For example, in the Balinese water temples cited below, ‘fitness’ is viewed as the deliberate efforts of farmers to cooperate in setting water irrigation management patterns so as to maximise both pest control and harvests. Adaptive agents continuously move around a changing fitness landscape, continually changing their and others’ environments and continually adapting to the changes in others. This has been taken by some as a reflection of how the success of any organisation’s strategy crucially depends on the strategies of others. Fitness landscapes can be a powerful metaphor to guide thinking about co-evolution, and to highlight various features of complex adaptive systems.

Co-evolution has implications for many aspects of human systems, but perhaps the most important relates to setting of targets – once a measure becomes a target, it ceases to become a measure. The implication is that measurement is distorted because of the existence of a target: the creation of a target is in fact an intervention in itself that has consequences and incentives, such that the system and the agents evolve together to maintain the status quo.

This is highlighted with respect to the concept of homeostasis. Originating in biology and used heavily in cybernetics, homeostasis is a concept that illustrates how adaptive agents seek to maintain certain factors within a desired range, often exerting energy to maintain stable levels (this is a negative feedback process).

Perhaps the clearest example is how the human body works to regulate its temperature. In the social realm, there is evidence to believe that a partially effective vaccine for AIDS or malaria may increase the rate of infection as people respond by cutting back on other precautions⁶, achieving a certain desired trade-off between risk and behavioural convenience. One of the most famous examples of homeostasis is that articulated by James Lovelock in his Gaia theory, which states that the entire mass of living matter on Earth functions as a vast homeostatic super-organism that actively modifies its planetary environment to produce the environmental conditions necessary for its own survival. In this view, the entire planet maintains homeostasis.

Other kinds of homeostasis include risk homeostasis, where, for example, people who have anti-lock brakes have no better safety record than those without anti-lock brakes, because they unconsciously compensate for the safer vehicle with less safe driving habits. It has also been applied to scientific thinking by postmodernists, who argue that there are societal power centres governed by a principle of homeostasis. An example is the scientific hierarchy, which will sometimes ignore a radical new discovery for years because it destabilises previously accepted norms⁷.

There has been increasing attention in the social and political realms to understand individual behaviour and its relationship to social institutions by application of the concept of co-evolution. Work by the Santa Fe Institute^{8 9}, focusing on small-scale societies, has identified a mutual interaction between social behaviour and social interaction – over time, behaviour and institutions evolve in an interrelated way. In this evolution, the relative fitness of behavioural strategies and of specific social institutions determines their success over time.

This has been demonstrated with regard to institutions such as through resource sharing, alongside which supporting social group behaviours have co-evolved, despite the costs this poses at the individual level. This is explained by reference to the contribution such institutions make to the evolutionary success of the group as a whole.

An implication of this work is that altruistic behaviours and warfare as a group practice may have co-evolved, the frequency of warfare contributing to the evolutionary success of altruism and the presence of a significant fraction of altruists in a group contributing to a group’s war-making capacity¹⁰.

This example highlights how co-evolution emphasises not only how a particular agent adapts to and is shaped by the agents in a system, but also how agents adapt to the overall properties or institutions that result from individual interactions. In this way, the various agents of a complex adaptive system co-evolve with each other, as do the macro properties and institutions.

Example: Balinese water temples

In Bali, rice is grown in paddy fields fed by irrigation systems, which are dependent on rainfall. Rainfall varies by season and elevation and, in combination with groundwater inflow, determines river flow. Traditional irrigation begins with a weir in a river, which shunts all or part of the flow into a tunnel that emerges downstream, lower ground, routed through a system of canals and aqueducts to the summit of a terraced hillside. By controlling flow of river into terraced fields, farmers are able to create pulses in important processes, e.g. the cycle of wet and dry phases alters soil pH, with numerous effects on the quality of the soil. Flooding and draining of terraces has effects on pest populations. Cooperation between farmers with adjacent terraces can create a fallow area over a sufficiently large area to deprive rice pest of their populations, but if too many farmers follow an identical process, there will be peak irrigation water demand at the same time, and there may not be enough water for all, especially as there are only weirs every few kilometres. Water sharing and pest control are opposing constraints – optimal level of coordination depends on local conditions.

Simulation models are uniquely appropriate for addressing issues of adaptation and determinism in the development of complex social systems. Extending the use of simulations in biology, which might focus on the co-evolution of algae and Antarctic sea ice, this analysis moved from natural ecosystems that evolved through a process of ‘blind’ natural selection. Systems of interest to anthropologists are by definition shaped by conscious human intentions. Life in the sea ice of Antarctica is thought to have evolved opportunistically through the random effects of natural selection. On the other hand, centuries-old Balinese rice terraces would cease to function if the Balinese farmers stopped managing them. The introduction of human agency into natural ecologies blunts the tools designed for the study of such ‘blind’ co-evolutionary processes. As the authors of the study¹¹ suggest:

Historical evolution of irrigation systems, rice terraces and water temples [is an example of] engineering of the landscape as generations of Balinese farmers cleared forests, dug irrigation canals, and terraced hillsides to enable themselves and their descendents to grow irrigated rice … [There would be] false starts, abandoned irrigation works, conflicts … [but] records would show the traces of conscious design … the realisation of each generations plans changed the world for their descendents…

The study uses ecological simulation modelling to illuminate the role of human agency in reshaping the ecosystem and the emergence of cooperative behaviour among Balinese farmer. The ‘fitness value’ or payoff of different farming strategies changes as a result of the complex interactions between irrigation networks and the domesticated ecology of the rice terraces. A spontaneous process of self-organisation occurred when temples were allowed to react to changing environmental conditions over time in a simulation model. Artificial cooperative networks emerged that bore very close resemblance to actual temple networks. Tellingly, as these networks formed, average harvest yields rose to a new plateau. Subsequently, the irrigation systems that were organised into networks were able to withstand ecological shocks such as pest outbreaks or drought much better than those that lacked networks. The networks have a definite structure, which leads to higher sustained productivity than would be the case if they were randomly ordered. Structures emerge without conscious planning through process of coevolution. Water temple networks may represent a type of social organisation: a self-organising managerial system shaped by a process of agents co-evolving to a changing environment.

Implication: Look for and work with the effects of co-evolution

An understanding of tendencies of adaptive agents to co-evolve in response to opportunities and constraints put on them highlights the drawbacks to certain kinds of development and humanitarian interventions. This means it is important to pay attention to the multi-faceted nature of many problems in development, and the frequency of unintended and indirect effects. In this environment, ‘policies are often found to overlap or be in conflict and the policy system is unduly complicated producing inefficient or even ineffective solutions and generating new problems’ ¹².

For example, there is evidence of a mutual construction of events by bilateral donors and relief agencies in a way that suits both agents’ ends. This is a clear example of a potentially damaging kind of co-evolution. As the authors of a HPG study suggest:

The [decision making] process involves finding a common “narrative” about the situation in question that fits the priorities of agency and donor alike, and allows the two to be reconciled. While this narrative may indeed be based on sound analysis, and may lead to appropriate responses, there are structural reasons why it may not do so, given the potential organisational interests of both parties in the acceptance of one narrative over another. … [There is] evidence of mutual “construction” of crisis by agencies and donors in a way that suits both their ends. Given the tendency of contract-based relationships to be evaluated against contracted input and output rather than actual outcomes, there is a danger of circularity – problems are “constructed” and “solved” in ways that may bear little relation to actual needs ¹³

Given the above situation, the natural inclination of agencies may be to not ‘rock the boat’ – the perception of such interdependence may give rise to highly risk-averse behaviours. More worryingly, such interdependence can be used to further the interests of specific agents who control resources. Examples abound, such as the debate on conditions attached to aid, and the conflation of military and humanitarian goals in the war on terror. At a more day-to-day level, there are legitimate concerns that co-evolution shapes the way in which particular actors behave and perceive local contexts. When the idea of the log frame is successfully passed on – usually ‘downwards’ from a donor to an international NGO, or from an international to a local NGO – and adopted, it can create a false belief that everyone is reading the resultant logical frameworks in the same way. For example, an expatriate development worker might feel s/he understands the work of a local NGO if the staff can follow the LFA and produce a clear logical framework. Unfortunately, the most successful local NGOs, those which repeatedly get foreign funding, are those which have learned to play the game and can present their thinking in a logical framework in order to get funding. This creates a distortion of the relationship between so-called ‘partners’; the local NGO has to adapt to the alien way of thinking, whereas the foreign partner, whether NGO or donor, does not need to adapt to the local context, and relates to the nature of adaptive agents highlighted in Concept 8. Some have argued, on the above basis, that the log frame represents an ideology rather than being an objective technical management tool. This highlights the fact that co-evolution in social systems is intimately tied to issues of power and control.

It is crucial to see oneself and one’s own organisation and actions as part of a wider system, and pay attention to the way that various actors may adapt and react to various constraints put upon them and opportunities made available, incorporating into the design an understanding of how they may react to a proposed policy and a willingness to engage in an iterative interaction if the reaction is unanticipated (Khan, personal communication, 2007). It also means taking into account the possible effects of homeostasis, which can often negate any hoped-for gains made by policies, in much the same way ‘risk homeostasis’ leads to drivers who react to improved safety measures by driving faster and hence more dangerously. Where there are clear incentives to co-evolve because not co-evolving would lead to a sub-optimal position, we can expect a degree of homeostasis.

In the context of international aid agencies, there may be a degree of what can be called power homeostasis. The aid system tends to locate power, control and legitimacy in the hands of richer countries and their representatives. This may mean that there are some groups of actors not able to self-organise or to co-evolve with other actors.

One somewhat authoritarian approach is to seek to constrain other actors in order to minimise the number of interconnected factors and reduce unintended consequences. For example, many rock music festivals do not publicise the availability of free medical care, in order to prevent a ‘moral hazard problem’. However, an approach that requires withholding information about one’s actions and purposes may be inappropriate, impractical or impossible in some circumstances. Alternatively, one could plan specifically based upon the likely ‘unintended’ consequences of an action. One such method is known as the Lijphart effect, where ‘the belief that undesired results are likely if decision-makers do not take unusual steps may lead them to take such steps and prevent the “natural” outcome from occurring’ ¹⁴. Again, this may not always be a relevant option.

This issue brings into question the nature and effectiveness of clear and explicit indicators of progress which are conceptualised and defined by external parties. As Jervis¹⁵ argues, the interactions among perceptions, behaviour and measurement mean that the meaning of a particular yardstick or indicator will be altered by its being used to measure progress or performance. Regardless of whether adaptive agents are co-evolving in conscious ways to manipulate results, a consistent reading of the indicator will be undercut by processes set in motion by the behaviour of various actors.

This is similar to the warning against the assumption of ‘all other things being equal’. If all things were equal, it might be preferable to financially bolster medical aid programmes with the highest survival rate of beneficiaries. The multiple direct and indirect effects of using such a yardstick, however, as the system reacts to an actor’s behaviour, cause undesired effects. For example, this ignores the fact that better programmes may focus on beneficiaries who are in greatest need. However, using survival rates would give those managing aid programmes incentives to avoid more difficult cases and leave those in the greatest need for others to deal with. This is not to say that no indicators can be useful, but it does suggest that they must be carefully formed. It is likely that they will have to be rooted in the context in
which they are used.

Next part (part 20): Conclusions – summary.

Article source: Ramalingam, B., Jones, H., Reba, T., & Young, J. (2008). Exploring the science of complexity: Ideas and implications for development and humanitarian efforts (Vol. 285). London: ODI. (https://www.odi.org/publications/583-exploring-science-complexity-ideas-and-implications-development-and-humanitarian-efforts). Republished under CC BY-NC-ND 4.0 in accordance with the Terms and conditions of the ODI website.

Header image source: qimono on Pixabay, Public Domain.

References: