New England Complex Systems Institute

APRIL 23 | 7 to 9 PM

Transcending the Group Selection Controversy in Evolution

with Implications for the Human Condition


The event can be live-streamed for free on this page and on our YouTube channel.

Note: The live event has been shifted to virtual due to logistical issues.


Your generous contributions will support future programs and events.

Kin or Group Selection?


For over a century, a debate has raged in the heart of evolutionary science. Is natural selection limited to selfish action by genes, or can altruism within groups of genes or organisms improve fitness? The distinction can seem slight, but it raises important questions.

Does altruism exist?

The debate exists at two levels: ideas about how genes contribute to the success of organisms, and the mathematics describing that process. The ideas are embodied in scenarios that we can think about; the mathematics is used to prove that we understand something completely general. However, the scenarios have assumptions that may be put into the math, and the proofs depend on those assumptions. We have shown that underlying the debate is an over reliance on statistics that misses relevant variables—variables that need to be included to understand what is happening. Even when equations are solved correctly, if the right variables are not included, the conclusions are not necessarily correct. Translating this back into the ideas is a key step that can help us understand what can happen with altruism. In these pages we will explain the ideas and why they are not included in the way evolution is taught in both popular books and textbooks.

Yaneer Bar-Yam, Brief discussion of the mathematics of kin and group selection, New England Complex Systems Institute (January 22, 2019).


Statistics and Evolution

In the 1920s, statistics was developed as a powerful new approach to understanding genetic inheritance, describing how parental genomes combine to become those of offspring. Each offspring is treated as one instance of all possible combinations of parents. These methods continue to play a central role in the analysis of heredity and population biology.

However, statistical methods rely on approximations, whose significance are still not widely understood.

Y. Bar-Yam, From big data to important information, Complexity doi: 10.1002/cplx.21785 (April 25, 2016).

Mendelian inheritance is an example of a statistical approach.


Gene-Centered View

The gene-center view of evolution, popularized in 1976 by Richard Dawkins in The Selfish Gene, is a statistical approach. He argued that natural selection exerts its force on single genes. As far as the individual genes are concerned, the rest of the genome, organism, and species are merely vehicles for its own reproduction.

Dawkins uses what he calls the “rowers’ analogy.” By following this example, we can see the assumptions of the statistical approach—and how they break down.

The analogy involves a group of rowers running races. The rowers represent the gene pool, and the boats represent organisms. Pairs of rowers run heats, and the winners return to the rower pool (duplicating themselves to keep the population size the same).

To represent alleles (different version of the same gene) each rower either speaks English or German. Pairs of same-language rowers are better able to coordinate and win more raises. If the initial pool happens to start with more speakers of one language (say English), then that language will proliferate, eventually wiping out the other language group entirely.

According to this analogy, the only thing that matters is the statistical distribution of the rowers (genes) in the pool (population).

This assumption is an example of the mean field approximation.

Yaneer Bar-Yam, Non-technical explanation of the breakdown of Neo-Darwinian — Gene Centered view, New England Complex Systems Institute (February 29, 2016).


 Breakdown of the Mean Field

Dawkins’ analogy seems reasonable, but a hidden assumption has surprisingly far reaching consequences.

The mean field approximation places winning rowers back into the pool and pairs them up again at random. This is like assuming random mating and well mixed populations in real organisms. But what if, instead of returning to and from the pool randomly, winners went to the back of a line, with new pairs being selected from the front.

The outcome is remarkably different.

Same-language pairs still have an advantage, but instead of the more populous group wiping out the other, they both forms clumps of their own type. Within its own patches, the minority group is not affected by the larger group. The boundaries between these patches will move and change over time, which is an important behavior not captured by the mean field approximation. Even if one group eventually wins out, it will take much longer.

This result reflects the real world, where English and German speakers both still exist within their own countries.

H. Sayama, Y. Bar-Yam, The gene centered view of evolution and symmetry breaking and pattern formation in spatially distributed evolutionary processes, in Nonlinear Dynamics in the Life and Social Sciences, W. Sulis and I. Trofimova, Eds. (NATO Science Series A/320, IOS Press, 2001) 360-368.


 Altruism and Selfishness

What do both views have to say about the evolution of altruism? Altruism can be defined as sacrificing your own reproductive success to benefit the reproduction of another organism.

Kin Selection

According to the gene-center view, genes are only looking out for copies of themselves. A gene might allow an organism to behave altruistically if it would benefit the reproductive success of an immediate relative. After all, an organism’s nieces and nephews have a good chance of carrying copies of the selfish gene.

In this view, natural selection can only act above the scale of single genes if copies of the same genes are found in a closely related group of organisms, but not on a higher scale. Hence the name Kin Selection.

Altruism directed toward individuals without an immediate blood relation is then considered unsustainable. If an altruistic allele and a selfish allele are both found in one population, the selfish allele will eventually win out. The selfish gene will get the benefit of the altruistic individuals’ reproductive sacrifices, while offering nothing in return.

Group Selection

Group selection is the idea that natural selection can act on scales larger than a gene, i.e. at the organism or social group level. Let’s look at a case of a selfish allele and an altruistic allele in a population that breaks the assumptions of the mean field approximation.

If altruistic organisms form isolated communities, their self-sacrificing tendencies can benefit the members of their community without having to worry about free-loading selfish organisms. They can work together to increase each other’s reproductive success. The key to how this applies in the world is to understand the dynamics of the boundaries between the groups. These boundary dynamics were not treated in kin selection models. For a spatial model, where mating is local, the neighboring association of altruistic individuals is enough for altruistic behavior to evolve naturally.

The formation of groups for selection only requires that individuals have a preference for mating with organisms in the same geographical location where they were born. As with German and English speakers, this is evident from real world examples.


In Kin Selection, altruistic organisms help genetically related individuals to reproduce. In Group Selection, altruistic organisms help members of a group they associate with. It turns out that Kin Selection requires related organisms to associate with each other; that’s what it means to help genetically related individuals. At the same time, Group Selection also requires those who associate with each other to be genetically related. Shared genetics and association mean that kin and group selection are flip sides of the same coin. When statistical averages are used, they can be proven to be equivalent mathematically.

Both sides are identifying the relevant variables differently. How can we identify what is really important?

M.J. Wade, D.S. Wilson, C. Goodnight, D. Taylor, Y. Bar-Yam, M.A.M. de Aguiar, B. Stacey, J. Werfel, G.A. Hoelzer, E.D. Brodie III, P. Fields, F. Breden, T.A. Linksvayer, J.A. Fletcher, P.J. Richerson, J.D. Bever, J.D. Van Dyken, P. Zee, Multilevel and kin selection in a connected world, Nature 463: E8-E9 (2010).


Evolution and Complexity Science 


One of the hardest things to explain is why complex systems (like evolutionary populations) are actually different from simple systems. The problems is an interlocking set of assumptions: Every system has a set of properties that can be understood with experiments and modeling. All the collected data and the finished model will tell you everything you need to know.

The flaw in these assumptions is that we may be starting with the wrong set of properties. Key properties might be missing or their importance might change over time. Why don’t we just keep studying properties until the system is understood? The amount of data will be overwhelming and the process will never end. The key is to identify which properties are important, which itself is a dynamic property of the system.

Complexity science allows us to understand complex, dynamic processes like evolution by identifying the right variables. A key component of complexity is scale. As we saw above, focusing at the smallest scale of individual genes fails to capture important features of population biology. These multiscale considerations are why complexity science supports the theory of Group Selection. Kin Selection is not sufficient.

A complex systems understanding can provide insight into other aspects of evolutionary science, besides altruism, which are also missing from the mean field approximation.

Yaneer Bar-Yam, Why complexity is different, New England Complex Systems Institute (March 16, 2017).


 More Results Beyond the Mean Field


Populations can be much more diverse than conventional population biology predicts. Prior theories can not describe the high level of biodiversity found in nature. In a well mixed population, the diversity disappears exponentially fast. It persists much longer when it is not well mixed. Interestingly, many of the experiments that test population biology are done in a laboratory where populations are mixed. So the assumptions of the theory and the experimental conditions match. Laboratory populations are known to be very homogeneous in genotype and natural (“wildtype”) populations are much more diverse, consistent with the expectation that they are not well mixed.

E. M. Rauch, Y. Bar-Yam, Theory predicts uneven distribution of genetic diversity within species, Nature 431: 449-452 (Sept. 23, 2004).


The separation of one species into two or more species over time has been a subject of much controversy. If we start by considering the species to be mixed at every generation of mating, then how does it stop being mixed? If on the other hand we include a non-mean field description of a species, then speciation results from progressive separation of types that are more likely to reproduce with those of the same type. The spatial population version of this idea has been shown to describe speciation very well.

M.A.M. de Aguiar, M. Baranger, E.M. Baptestini, L. Kaufman, Y. Bar-Yam, Global patterns of speciation and diversity, Nature 460: 384-387 (2009).

A. B. Martins, M. A. M. de Aguiar, and Yaneer Bar-Yam, Evolution and Stability of Ring Species, PNAS 201217034 (March 11, 2013).


In the real world, increasing rates of travel by airplanes and other means is leading to a lot of mixing. We saw that mixing leads to mean field type of behavior that is very different from the case of spatial models. Increasing transportation brings us closer to the well mixed case. Biodiversity should decrease. We see this effect in invasive species eliminating local variations. Another effect is a change in the pathogens responsible for infectious diseases. When a pathogen can only spread locally, the most aggressive strains go extinct as they kill off the local hosts in their local patch. As transportation increases, patches become irrelevant and the successful strains are those that are more aggressive and deadly. The practical implications are present in the increasing prevalence of deadly epidemics, like the recent Ebola outbreaks.

E.M. Rauch and Y Bar-Yam, Long-Range Interaction and Evolutionary Stability in a Predator-Prey SystemPhysical Review E 73: 020903 (2006).


The mean field approximation suggests that evolution cannot select for specific lifespans. Living longer means more chances to reproduce, so selfish organisms should live as long as they can. However, local reproduction links altruistic organisms to their descendants through the local availability of resources across generations. By dying “early,” they can leave more resources for their descendants. If ecological conditions can select for limited lifespans, this has important implications for longevity research. If our bodies have an evolutionary mechanism to begin the aging process, research into interventions could greatly extend our lifespans.

J. Werfel, D.E. Ingber, Y. Bar-Yam, Theory and associated phenomenology for intrinsic mortality arising from natural selection, PLOS One (March 29, 2017).