The Elephant in the Brain
I.
If you’ve been subscribed to this blog for a while, you might’ve heard me mention signalling. When I’ve written about it in the past (see here), I’ve often been talking about signalling in the context of education policy. The claim there, made by Bryan Caplan among others, is that a big part of the reason that people who go to university earn more money than those who don’t is that a university degree signals that they have qualities that employers are likely to value.
Someone who went to Oxford is more likely to be bright, hard-working and conformist than someone who never went to university. So employers might dismiss a CV from someone who never went to university, even if they don’t know anything else about the candidate, because of the signal that the lack of a degree provides.
But signalling isn’t only relevant to education policy. Kevin Simler and Robin Hanson’s book The Elephant in the Brain makes the case that signalling is an important reason why we do many of the things that we do. And the book has had an enormous influence in the sort of online circles I inhabit, so I thought it might be worth writing a sort of review/summary type thing, to give a quick primer on what exactly people mean when they talk about signalling (or refer to ‘the elephant in the brain’).
The main argument of The Elephant in the Brain is that a lot of what you do is not for the reasons that you might think. If we do something, we like to think (or least claim) that our motivations are pure or virtuous. In fact, the elephant in the room brain is that we often have selfish or otherwise unattractive motivations. We can start by using an example from the animal kingdom of animals doing things that, on first glance, seem irrational.
Baboons spend a lot of time grooming each other. This makes sense - it’s difficult for a baboon to groom its own back, so George the Baboon might ask his friend Greg the Baboon to do it instead. But the sheer amount of time that baboons spend grooming each other is, on the surface, weirdly high. They spend about 17% of daylight hours doing it, in comparison to 0.1% of daylight hours in other primate species. Birds spend about 0.01% of daylight hours carrying out similar cleaning behaviours. So, why do they spend so much time grooming each other, when they could be eating, having sex, and whatever else it is that baboons do?
The answer is that the main purpose of grooming is not hygiene. Grooming another baboon comes with a number of social benefits. Grooming partners share food and support each other when fights break out. In larger social groups that have more complex structures, more time is spent grooming than in smaller groups - the larger groups are more tricky to navigate, and these relationships are more important, so baboons spend more time grooming each other.
Baboon social grooming involves a lot of signalling. When George grooms Greg, he’s signalling that he’s willing to spend his spare time doing what he can to help Greg. And by turning his back to George to let him carry out the grooming, Greg signals that he trusts George.
II.
Human behaviour also involves a lot of signalling and other hidden motivations. If I give money to charity, part of the reason I’m doing it is probably that it signals to others that I’m a good person, or that I’m rich enough to afford to give some of my money away. How do we know that charity is at least partially signalling (or other hidden motivations) rather than only a desire to do good? There are a few studies that can help us here.
Landry et al. (2005) run a door-to-door fund-raising experiment, and find that a one standard deviation increase in the attractiveness of the (female) fundraiser increases the average donation by between 35 and 72 percent.
There seems to be fairly strong evidence for the ‘watching eye effect’, which is where people are more likely to behave altruistically when they believe they’re being watched. There’s even some evidence that an image of watching eyes can make people behave more altruistically, even if there’s no indication that they’re actually being watched. That being said, I’m unsure if this effect has fallen to the replication crisis. The evidence from meta-analyses seems to indicate that it might, kinda, sorta be true, but there have also been a load of failed replication attempts. The jury’s out on this one.
Simler and Hanson also write that the fact that anonymous donations are extremely rare means that donating to charity probably isn’t all about helping other people, and notes that ‘only 1% of donations to public charities are anonymous’, citing Glazer and Konrad (1996).
Is giving money to charity the only time when humans do things for motivations that are, *ahem*, different to those that they advertise? No. In the Simler and Hanson model of human behaviour, signalling is everywhere. Did you invite a date to an art gallery? You’re probably signalling to others that you’re a person of culture.
Or maybe you’re even more dedicated than the gallery-inviter, and are actually an artist yourself. Humans have been making art and music for millennia - and this is true universally, rather than being confined to a few weird cultures. Evolution selects for traits that increase your chance of survival and should eradicate doing things that are both pointless and costly. Why, then, would anyone spend time painting instead of fishing?
Unsurprisingly, Simler and Hanson turn again to signalling. Making art is another example of a fitness display - it signals to potential partners that you’re worth shacking up with. In their model, art is boasting. I have the spare time to sit around and put painstaking effort into making something that is pretty much useless, what does that tell you about me?
This might help us to understand why people are often more approving of art that seems like it took a lot of effort to create. Simler and Hanson point to one study, Smith and Newman (2014), that apparently shows that people think more of art created by a single artist than multiple artists. This might also give us a clue as to why some people dislike modern art: ‘My five year old could have done that!’. And the response from modern art defenders is generally not to ask whether that even matters, but instead to claim that modern art is actually hard, as Susie Hodge does in her book Why Your Five Year Old Could Not Have Done That.
Or how about the recent complaints about art made by Artificial Intelligence (such as the depiction of the elephant at the top of this post)? Maybe those can also be partially explained by the ‘art as boasting’ model.
What else is partially signalling? I’m not going to summarise the entire book, but a brief look at the titles of the chapters in the second part of the book is instructive: Body Language, Laughter, Conversation, Consumption, Art, Charity, Education, Medicine, Religion, Politics.
III.
What makes for a good signal? The most useful signals are honest - they signal something that is actually true about the person sending the signal. And the best way to make it clear that a signal is honest is to make the signal costly.
So, if I want to let people know that I have impressive financial resources, there are a few ways to go about it that actually work. As already discussed, donating lots of money to charity is one way (and as a bonus, it also signals that I’m altruistic). Wearing very expensive designer clothes is another. Buying a sports car is another. Posting photos on Instagram showing luxurious holidays is another. And on it goes. A known teller of tall tales claiming to have millions of pounds in a secret bank-account will likely be less persuasive.
Throughout the book, there are plenty examples of costly signals that are worth quickly touching on:
As discussed at the beginning of the piece, getting into and graduating from a top university is a good signal that you might make for a good employee. And a big part of the signal here is that you actually got in. If you wanted to, it wouldn’t be particularly hard to sneak into Stanford and listen to the lectures for free. In fact, one of the authors actually did sneak into lectures at Stanford without much trouble. But only tiny numbers of people do this, because it doesn’t signal that you’re smart enough to get into Stanford1.
Sacrifices can show that a person really is dedicated to some cause that they claim to be dedicated to. Why are some members of the Clergy celibate? Why do Effective Altruists (and members of many other groups) give 10% of their income to charity? Why do Muslims undertake Hajj to Mecca? Hanson and Simler take the view that sacrifices work well as costly signals that someone really is as dedicated to their religion/social movement/political affiliation as they claim to be.
Among animals, the go-to example is often a Peacock’s extravagant plumage, which signals genetic fitness. I believe that this was one of the first examples of a costly signal, given by the biologist Amotz Zahavi. This is an example of the ‘handicap principle’, the claim that animals give themselves a handicap that could not be afforded by other animals with less of some desirable trait.
The Elephant in the Brain is really good, and you should go and read it! People talk about signalling constantly, and this book is largely why. In fact, the book, as well as the idea of signalling, have become something of a meme. So much so that Jeremy Driver, in his explanation of why some political activists seem not even to attempt to convince other people, writes ‘Some losers would probably call this “signalling” or something’. And yes, I would call it tribal signalling! And now, hopefully, you know why.
As with so many bold ideas, I worry that it leaves the impression that one causal factor (signaling, in this case) has overwhelming explanatory power. One of the few things we can say for sure about social sciences is that the effect sizes are almost always small, whether of immutable traits or interventions (<<0.5, probably <0.2). From my armchair, I suspect it's because people are so different, particularly when considering their contexts when making a specific decision.
To get specific, maybe the educated urban 20something makes art because of signaling. But maybe Frida Kalho really had something that she needed to say. Maybe a toddler just likes orange. Maybe that same 20something transitions into doing it for the sake of quiet mastery-driven self-esteem after getting some random praise early on from their now-wife. So the effect (like many others) can be real, but it's a lot more useful to specify the people+contexts where we'd expect it to be most predictive.
+1 to Matthew's comment. I'd add: many of the counter-arguments that signalling advocates use against perspectives like Matthew's end up inflating the concept to the point of near-meaninglessness.
The most fundamental phenotypical fact about Homo sapiens is that we live in cultures, meaning (a) we act collectively but (b) the 'script' for our collective action is not hardcoded into our psychologies; it's flexible and emergent from our interactions with others. For example, every human culture has complex collective courting/mating rituals, but they each have *different* rituals. Evolution didn't give us a psychological module for using Tinder; it gave us a generalised motivation to seek a mate, but left a lot of things blank (who is an appropriate mate? how do I meet them? how do we coordinate our actions?) to be filled in through cultural learning. This cultural learning is learning *from* and *about* other people, looking at their actions and gleaning some credible information about their characters and motivations. In other words, almost all of our behaviour must *necessarily* have a signalling aspect, because humans must transmit information to each other about their behavioural dispositions in order to live under culture.
But when interpreted like this, the claim 'a lot of human action is signalling' is nearly tautological, essentially equivalent to 'human beings live under culture'. The claim that art is signalling, for example, comes down to the claim that artists create art in large part to transmit private information about themselves to other people - which is just obvious! The claim that love and dating is just signalling is even more blatantly trivial if poked at a little, since we don't have a cognitive module for Tinder. The claim that morality is just signalling is at least as old as Adam Smith, if not Aristotle: human beings would never learn to make moral judgments if they were alone outside of society, so the phenomenon of moral judgment is emergent from learning information about other humans. If 'signalling' is interpreted broadly, as meaning just 'conveying information about oneself through one's behaviour', then none of this is particularly interesting or insightful.
Now, I think this inflated concept of 'signalling' still has a lot of use in e.g. mathematical models of human behaviour: an account of signalling can help us incorporate culture into our rational choice models, which is incredibly helpful. But economists and rational choice theorists tend to get this backwards: rather than 'signalling' being a concept used to revise our rational-choice models so they match up more closely with common-sense accounts of human behaviour, they see it as a concept from rational-choice models that we should use to revise our common sense!
Now, as we've already seen, in fact there's nothing contrary to common sense in the claim that human behaviour is largely signalling. So in order to vindicate a revisionist account, people like Hanson and Simler must be using the word 'signalling' in a more narrow way, limiting their focus to instances where our behaviour conveys *certain types* of information. Typically, people like this have a conception that we might call signalling-as-bragging: the information that our behaviour conveys is roughly 'Oh, look at me, aren't I great?'. To take a specific example, rather than the artist signalling information about their emotions or their moral convictions or etc., they are signalling 'I am talented and have good taste'.
The claim that most of human behaviour is braggadocious *is* contrary to common-sense; but it's also not obviously true. Interpreted in this narrow way, the hypothesis that 'much human behaviour is signalling' is subject to the kind of arguments Matthew makes: it's probably underpowered, and while bragging might be a big part of human psychology there just isn't enough evidence to make a stronger claim than that. In response to these claims, the advocates of 'signalling hypotheses' often then retreat to the motte, interpreting signalling broadly.
One can see this move in the example of baboons. Sam introduces this example as a case from the animal kingdom that might illustrate, via analogy, how 'we often have selfish or otherwise unattractive motivations'. But, while baboon grooming is signalling behaviour, it's not unattractive signalling: the baboons are signalling their willingness to aid each other! 'I am doing this to show how much I am willing to do for you, in the hope that the feeling is reciprocated' is not an unattractive motivation, never mind a selfish one, for animals (including humans) to have - it's the kind of motivation that we *hope* our romantic partners have, for example, in remembering anniversaries or purchasing flowers.
If we approach human behaviour with a lens that is overwhelmingly dominated by considerations of economic rationality, then signalling might initially look like a small and unimportant quirk; it could then be a surprise to learn how prevalent it is. But, if we are interpreting 'signalling' broadly and using any other lens to analyse human behaviour, its prevalence isn't shocking at all - it's basic and even obvious. Another way to put this is that signalling is not, in fact, just one causal factor (as Matthew put it); it's a whole range of causal factors, corresponding to essentially every culturally-influenced part of our psychology. The issue is that in economic contexts, signalling is often introduced with examples of bragging or showing off (conspicuous consumption, for example, or metaphorical peacock tails), and so the prevalence of signalling-interpreted-broadly is conflated with the prevalence of signalling-as-bragging, a single causal factor that is attributed huge significance. These two different ways of interpreting a 'signalling hypothesis' are not often disambiguated, creating slippage between a true-but-boring claim and an interesting-but-false claim.