The Taxonomy of Selection
This post consists of some notes that looking at the analogy of natural & artificial selection to design and its consequences. A worthwhile paper on a related but different topic is Christina Cogdell’s Products or Bodies? Streamline Design and Eugenics as Applied Biology (2003) Design Issues, 19(1), 36-53. doi:10.1162/074793603762667683
Types of Selection
The purpose of this page is to describe how natural selection can be used as a framing tool for recognizing how artifacts, services, and interventions can affect individuals and natural populations of humans and other species. The point is not to draw a direct analogy, but to try to link the effects of the things we make to the behaviors, growth, and flourishing of living things. These are not so much set rues as they are a set of guides that can help us reconsider the expected effects of changing our environment in order to evaluate the risk and alternative future possibilities involved in the production of technology from the most precursory to the most complex.
I was intrigued after a reading group discussion we had about anthropometrics. Wikipedia defines anthropometrics as the measurement of human to gather statistical data about the distribution of body dimensions in the population are used to optimize products. I would alter this definition slightly to say design products rather than optimize. Humans change and so do products.
We were a little unsettled by the focus only on human needs and the intent that anthropometry be entirely in support of comfort and ease of use. Taking a more critical approach, we started to brainstorm all of the different ways that design structures human and non-human behavior. We started to keep an eye open for ways that design and evolution can begin to interact. We hit some dead ends so I reached out.
I asked a group of colleagues if they knew of any comprehensive taxonomy of selection, and here is what one of them (Joel) contributed:
There are so many different ways to split selection up that it can be mind-boggling. To make it worse, those who study molecular evolution use different terms (positive, balancing) than those of us who study phenotypic selection. I don’t think there’s a way to taxonomize the terms satisfactorily, at least in a tree. It would probably look more like a convoluted Venn diagram.
That said, Joel laid out four areas that can be used to focus our attention. I’ve modified them from his interpretation, but they are basically agents, episodes, modes, and scales.
Here is how he originally wrote about it in his response to me:
In the phenotypic selection realm, I tend to split selection up in four different ways, based on agents, levels, fitness components, and mode.
The agent of selection, that is, the factor that causes fitness differences to arise, can be either ecological (phyisical or heterospecific) or conspecific. I would call the former ecological selection and the latter social selection (sensu West-Eberhard). The latter would tend to subsume sexual selection, which tends to be caused by male-male or male-female interactions. Includes frequency-dependent and density-dependent or other x-dependent.
The level of selection describes the units that exhibit fitness differences (which, annoyingly, Gould call “agents” of selection). This can be individual selection, and at higher levels, family selection, group selection, kin selection, social selection (sensu Wolf, Moore, and Brodie), etc. Hard and soft selection can fall under this category as well–Wade and Goodnight have good papers discussing this.
Third, selection can be split into different “episodes” by splitting total fitness into multiple components. This is usually done because it is empirically convenient, or to examine evolutionary trade-offs. This gives rise to terms like survival selection, fecundity selection, and sexual selection.
Finally, you can describe selection based on the shape of the fitness surface, i.e. the “mode.” This includes directional (linear), stabilizing, disruptive, and correlational (all three quadratic). Of course, the shape of the fitness surface is often complex, and you can have elements of all of these going on at once when you’re considering multiple traits.
Reframing Selection
We might think about what Joel said differently and transform it as the grammatical structure of a sentence. Where:
AGENT = SUBJECT
SELECTION =VERB
EPISODE = DIRECT OBJECT
MODE might be akin to diagramming the entire sentence. while SCALE is more like the context that the sentence takes place within (e.g. the paragraph or passage).
Agents
Agent refer to the most causal explanation for the response to selection. Agents provide the mechanistic explanation and frequently are the antagonists to the entity/entities experiencing the effects of selection.
From a designer’s perspective, these agents should be the artifacts or services we create either with the intent to exert some selective force or ameliorate it.
We can understand these as ecological agents that affect anything from the climate of our surroundings, our food supply, the structure or our living and working spaces, interactions with outer species (as in pets, disease, or domesticated laborers), and even perhaps to our conventional definitions of time that enable further articulations of the environment.
Similarly social agents work along the lines of our own perception, learned, and innate behaviors to enumerate male-male, male-female, family, and cooperative interactions. Sensation and display are extremely important because they distinguish among individuals to allow decisions about how to interact. Social agents range from clothing, jewelery and other status symbols to weapons, traditions, and business plans as agents of cooperation or competition.
There is a nice hybrid space too where ecological and social meet in the production of artifacts favoring or disfavoring reproduction–in vitro fertilization on one hand…and condoms on the other, for example.
Often, agent-based selection is described as selecting for trait ‘x’ and can even be more complicated when traits x, y, and z covary as a result of this selection. As a consequence we find that selection can be multi-facited and not reducible to a single interaction. Hence, we need to reconsider the cumulative effects of each agent’s contribution.
Episodic
Moving along the causal chain (if we can indeed identify it), we would then want to understand the factors or physical attributes that are on the receiving end of the agents’ work. From an empirical perspective, this is often where conflict begins and fluctuates in an ever-present set of trade-offs. We can split the effects into many different components looking at reproduction, lifespan, health, outlook, social status, niche, range, communicativeness, and, perhaps most importantly, agency (as the ability of an individual to act as its own agent).
This is the main point of interest in design–i.e. what, where, when do the effects of the design work manifest in nature?
Mode and Variance
In order to understand what patterns are present, evolutionary biologists look at variation and the response of a particular trait or episode to selection from agents. Here attention is focused on the values of the entire population in contrast to just the trait itself. We can certainly use these visualizations and modes to describe the distributions of episodic traits, but here there is explicit quantitative emphasis on the response to selection over one or more generations.
We can think about it in different ways: populational and interactional or hard and soft.
Populational
Populational patterns include the ideal types of directional (linear), stabilizing, disruptive, and correlational (all three quadratic), and null (no selection). The shape of the fitness surface is often complex, and you can have elements of all of these going on at once when you’re considering multiple traits.
The graph below is composed of four of these types where the axes show the proportion of the population as a function of trait values through time. The shaded areas represent the part of those populations that is being selected for. The color simply differentiates between types. Correlational selection is not shown because it consists of the interaction of multiple traits in response to selection, and we would need a 3-dimensional graph to show just two of those traits changing.
From the graph you can see the response of a population to null selection. Because mutation-based variation is not selected out of the population, the shape of the distribution randomly changes and will not fit a “standard” distribution.
The blue, directionally-selected distributions move (you guessed it) directionally because of pressure against one end of the distribution.
Likewise, stabilizing selection in green is like directional, but instead of one end the pressures are exerted to stabilize the mean.
For diversifying (aka disruptive) selection in red, the mean is selected against–leaving greater proportions near the previous tails of the distribution.
Interactional
The second way is what I call interactional, meaning it depends on the interactions among agents, often in space. Here, ecological agents and social agents exert their effects. The goal of this description is to capture the meaning behind the mechanism (thus, interaction) rather than the change over time. When coupled with population visualization techniques, one begins to get a dynamic picture of evolutionary change.
We may be able to consider correlational selection as a special case of interactional (and not populational) because the internal constraints within a population’s gene pool and genomic regulation are effectively a suite of internal genetic interactors at a different scale.
Normally however, we can think of interactional patterns in terms of frequency-dependence, density-dependence, or some other x-dependent factor related to ecological or social agents. So frequency-dependence, for example, is just a way of describing the total effect of agents…or of saying that the trait in question responds in a way that is frequency-dependent.
The main difference is that interactional describes the mechanism of selection itself (within a generation), while populational describes the response to selection (change between generations).
Putting the two together means we could graph dynamic change.
Hard
Both hard selection and soft selection are relative to the population as a whole. Hard selection is like a bat chasing an insect. The insect has some maximum speed that it can flee and the bat has some speed that it can chase. Assuming it is only the bat and the insect, then there is hard selection for the speed at which the insect can flee.
Soft
Of course insects are probably not alone since they tend to aggregate in large populations. Soft selection takes this into account and considers the effect of more than one insect fleeing. Here the insect needs not be faster than the bat, just faster than the other insects that the bat is following!
The major difference is one of absolute value or of percentage. Hard selection works on the absolute value of a trait while soft selection works on a percentage of the distribution of trait values.
Scale
The scope of impact of a particular service of artifact is also important, especially when we ask the question, “for whom?” Is it working on a emergent trait or even creating one? Examples might include political systems or policies that increase or decrease emmigration, the locating of a hazard that increases mutation rates, or one child policy.
Whereas before we were only considering a single ideal population, what happens when we include multiple populations? Does the work of the designer or design team affect traits that span across individuals and include qualities that can only be formed from collective-action?
Some levels of scale might include: individual, family, kin, group, social, community, or ecological.
