Inspired by Sarah Perry's great post "Why Cultural Evolution is True (And What It Is)." She does a better job with the substantive work of explaining cultural evolution- this is more of a meta-point about how we should treat models and metaphors.

It is unfashionable to use the concept of a "meme," in large part because it has been co-opted by The Internet [1]. This is a shame since it has the potential to be so useful as a base model from which to explain culture. Many people levy the accusation that the idea of a meme is just a metaphor, but I find that to be slightly misguided. Rewording George Box, all scientific concepts are metaphors, some are more useful than others [2]. It is better to explicitly analyze the ways in which a metaphor holds or does not hold in a specific situation.

In talking about cultural evolution, it's important to remember the distinction between evolution and natural selection, since the two ideas are so frequently conflated. (So much so that I feel like I may be cheating by trying to separate them.) Evolution is simply change over time. Natural selection, on the other hand, is a specific mechanistic model with four important components. In his essay "The False Allure of Group Selection," Stephen Pinker defines natural selection as having four key components:
  • The units of selection must be discrete
  • The units of selection must influence reproductive fitness
  • Mutation must be random relative to the fitness environment
  • Success must be defined by the number of copies present in the environment
But we can treat each of these components as a knob or a switch that we can twiddle to see what happens. Sexual selection can be viewed as a way of changing the "random with respect to fitness environment" requirement. Organisms don't choose mates at random- they choose mates that signal increased reproductive fitness. This can be viewed as a sort of agent-guided evolution. (There's actually a hypothesis that sex stuck around because it allowed us to out-evolve parasites). But I would hardly consider the existence of sex a point against using the model of natural selection in biology. Similarly, the fact that we can make predictions about, say, what songs will become popular doesn't automatically disqualify songs from being treated in a selective framework.

Some models of memetic transfer can explain cultural phenomena without even appealing to differences in fitness levels! In one paper, researchers showed that the rate at which popular songs shuffle on and off "Top N" lists (that is, lists of most popular songs, baby names, dog breeds, etc.) fits well with a netural random-copying model. In this model, people randomly either copy some other person's meme or generate a new meme. A similar concept in biology is that of genetic drift, in which genes spread across populations completely at random. Genetic drift is explicitly not the same as natural selection- it specifically violates "the correlation between the trait and reproductive success must be nonzero"- but it's an indispensable tool in explaining aspects of biological evolution.

We can in this situation treat the memetic view as a cultural null hypothesis. The null hypothesis answers the question: what do we expect to happen if nothing special is going on? In this case, it's: what do we expect to happen to societal trends in the absence of any specific trend or desire?

Another reason that people find the memetic view annoying or worthless is that many of its proponents (at least among laypeople) jump from "something analogous to natural selection is at play in cultural evolution" to "the only relevant factor in cultural evolution is a simply definable form of reproductive fitness." They then conclude that the only reason people hold ideas is because those ideas happened to be particularly good at replicating. Holding this position is like learning about biological evolution and concluding that the only organisms that should exist are viruses that reproduce as quickly as possible. In culture, just as in biology, there are a huge number of constraining forces that determine things like fitness or otherwise make change non-random.

So yes, we should be careful about taking the meme/gene metaphor too literally. But on the other hand, the fact that ideas and culture are not exactly like biology should not prevent us from assembling a toolkit of biologically-inspired models. We shouldn't come to a full stop when we run into differences, but rather ask how we can develop new, ideally mathematical, frameworks that account for these differences.

Why do I say mathematical? Mathematical models allow us to more explicitly lay out the metaphors that we're using for prediction. One might say that math is just the practice of constructing precise, publicly accessible metaphors. Mathematical models, even (in fact, especially) stupidly oversimplified ones, are useful in that they can identify the minimal set of assumptions needed in order to explain some phenomenon [3]. For instance, it is reasonable to conclude that communities become segregated by race when people are racist, and that intentionally constructed oppressive organizations are what keep communities segregated. But this great simulation by Vi Hart and Nicky Case shows that we don't need to postulate either explicit racism or structural oppression in order for segregation to exist. A slight preference for clustering with people similar to you can explain the global behavior. (Obvious disclaimer: I am not trying to make the argument that forces of oppression don't exist in society. They do.) Even if there are other much more complicated forces at play (and there are), the model tells us unambiguously that just telling people to not be racist will not desegregate communities. It can't necessarily tell us what will work, but it can say some things about what won't.

There are many ways in which evolution can be guided by both agentic and non-agentic forces. We don't need to draw a line in the sand between cultural change driven by purposeful individual action and that caused by uncontrolled collective behavior. [4]

[1] It also doesn't help that the term originated in a Richard Dawkins book. Any self-respecting scientifically-minded countersignaller wants to distance themselves as much as possible from Richard Dawkins, who is associated mostly with asshole atheists.
[2] See Lakoff and Johnson's Philosophy in the Flesh, or Luke Muelhauser's summary.
[3] See this great article in The Atlantic, which discusses the segregation model as well as a number of other interesting simulations.
[4] For many more thoughts on purposeful vs. non-purposeful forces in society, see Scott Alexander's magnum post-us Meditations on Moloch. (Warning: discussion of strong AI.)