There is a narrative hidden underneath the emergent social networking phenomenon.  It’s a story about our fundamental psychological reality and our absolute potential as a species.  Thousands of pundits are trying to tell it – but none of us really knows how to craft a narrative so epic in scope.  With the release of Google Plus, the story tellers are all out in force once again but largely its an underwhelming cacophony.  The one pure note sounded has been ex-googler Paul Adam’s slideshow – The Real Social Network.  And yet even this effort, I will argue, fails to cast any real light on what is driving the incredible social change that we’re seeing all around us.  We suffer from a lack of vision and an all too fervent pessimism about the possibilities opening out to us.  We need to remove those aspects of our theoretical foundations that are blocking us from uncovering the hidden narrative of social networking.  It’s in this spirit that I present today’s review of the theoretical underpinnings of Google’s new online social network.

There’s a lot to admire about Paul Adams’ research.  A lot of his conclusions about how people interact with one another are undeniable.  Most people do have lots of varied groups with which they interact, for instance – and he was spot on when he claimed that existing online social networks weren’t meeting this need.  His underlying philosophy that we should be building technology that matches the psychological reality of human sociality is also spot on.  An engineering culture which ignores this dictum might as well build hammers for one fingered aliens – we need hammers for human hands!

It’s unfortunate that we only have access to his slide show and that Google blocked publication of his book – because I feel some of the criticisms that follow may we have been considered in detail – and certainly couldn’t be considered in the context of a slideshow.  But in anycase – we can only work with what we have.

My biggest concern is that the presentation seems to conflate the current structural features of the way people interact with one another with the deeper, psychological reality that our evolutionary heritage has bequeathed (It’s hard to say if this mistake is being made from the slides presented – but it does seem to be implicit in the discussion).  You can’t just assume that the empirical snapshot given to us by Adams actually models this psychological reality.  Obviously, as good naturalists we all agree that this empirical snapshot is determined by that reality.  But that doesn’t mean that the current snapshot is a good representation of that reality.  What we need to see are the ways in which social networks are effected by changes in the surrounding context – to examine which features are preserved and which aren’t.  Because it’s clear that the structure of human social networks have changed significantly in various ways since the dawn of man.  We have to get clear on what facets have changed and what have stayed the same.  A particular empirical snapshot taken in one point in time doesn’t give us this.

I only see one data point in Adam’s presentation which seems to invoke research on what features of human social networks have persisted through time and that’s his reference to Dunbar’s Number – the maximum number of weak social connections that we are able to maintain before group social hegemony begins to significantly break down (around 150 connections).   I’ll explain Dunbar’s Number in a little more detail in a moment.  But the basic idea is that this limit has been observed across a wide cross section of cultures and in different periods of history – and so has a wide degree of confirmation.  Such persistence can be thought to be evidence of a component of our fundamental psychological limitations and reality.

Before I get deeper into Dunbar’s number, I want to point out first that the persistence of this data point throughout history doesn’t confer any persistence on the data yielded from the other particular empirical studies conducted on human social networks in the present by Adams and others.  We have to do the work to somehow find the data on this presumed persistence by consulting our palaeontologists, ethnographers, historians… etc.  I don’t see that in Adam’s presentation.  Yes, the research does seem to have been conducted in multiple countries – across multiple cultures – with near universal results.  But without knowing in detail the socio-economic realities in which the subjects are situated it’s impossible to rule out that perhaps  there just isn’t the kind of contextual variation in circumstances of the subjects that would cause variation in networking structures.

So maybe we’ll see more information in Adam’s book.  Until then – we need to be really careful about interpreting Adam’s results.  (Or until we can find the time to hit the literature or conduct our own studies).

 

Dunbar’s Number and its Interpretation

But even when you have a data point like Dunbar’s number which does seem to persist through time – you still have to be very careful about how you interpret it.  You have to be even more careful about how you use it to inform your choices when building social networking products.  To just see this as a blanket limit that can’t be circumvented, I’ll argue, is a big mistake.  Yet these are the sorts of conclusions drawn by people researching tools for developing online social networks.  This research paper on recommendation systems in social networks, for example, uses Dunbar’s Number as the rationale for stopping your recommendation engine from recommending contacts to people who are maxed out already at 150 connections and over.  To see why this sort of conclusion is wrong, we’ll have to dig a little deeper into Dunbar’s Number to understand just what it means.

Dunbar came up with this number by observing a correlation between the size of the neo-cortex and the maximum size of groups in various kinds of primates.  By applying this correlation to the size of the human neo-cortex – this yielded 150.  Dunbar then began to look for at examples of human organisation throughout history and he found many examples that supported the thesis that human groups begin to lose hegemony at around the number of 150.  This result, combined with the correlation between neo-cortex size and primate group size, led Dunbar to conclude that there was a causal relationship between neo-cortex size of humans and the size limit of effective social engagement in human social groups.

But why did we evolve this limit?  Dunbar’s answer to this question is fascinating. As primates hanging out in small groups we developed certain social procedures like grooming which conferred obvious benefits like controlling parasites and improving health.  Now it turns out that we weren’t very nice to each other back then (there’s one data point that remains the same through time) – with many sub-groups harassing and making life unpleasant for various individuals.  Given that you didn’t want your grooming buddies to get disrupted and thereby jeopardising your own comfort and health, it was natural to step in to their defence when they got harassed.  Over time, as our brains got big enough, grooming came to be seen as a signal that you wouldn’t harass and that you would help defend in the event of harassment.  So primates came to groom one another just to let others know that everything between them was tight.

The most important part of this story comes next.  Given that grooming behaviour takes a lot of time and energy – sending out the necessary social signals to the other primates in your group comes with a hefty economic cost.  For a species that has to spend a large amount of its time gathering resources, it means that you can’t groom as many primates as you would like.  And so it comes to be that you have to choose who you’re going to groom and who you’re going to ignore and thereby signal that you’re just not that into them.  It’s for this reason that there is an upper bound on the number of members in a group before social cohesion begins to break down.  Large groups have an economic overhead associated with appropriate signalling procedures that can’t be borne without the appropriate technology.

So then – it’s not that primate brains evolved in a certain way to only handle a certain number of connections and that THIS causes the group size limit.  The causal direction goes the other way.  Economic limitations involved in signalling procedures causes signalling failure – and this signalling failure constrains possible group size.  Primate brains adapted to this natural economic limit.  After all – why keep track of all those people that are just going to get the annoyed when you don’t signal to them correctly?  It’s not efficient.  Hence we get the correlation between brain size and group size.  The brain got as large as it needed to deal with the largest group sizes it would have encountered.

The natural limit for humans is much larger than primates.  Dunbar’s explanation for this is that our survival depended on the development of larger cohesive groupings, but in order to achieve this we needed a way to reduce the costs of the various signalling procedures used to maintain cohesion.  To this end, he claims, we developed language.  Language allows humans to signal on the cheap.  Rather than having to actually groom your fellow primate, you could just tell him that you would be there for him in a fight.  So the economic cost of signalling decreased dramatically.  Also it allowed us to cheaply gain information about others in the group through gossip as opposed to having to spend time observing an individual.

One  problem for Dunbar is in accounting for the obvious fact that much larger societal structures developed.  How did we manage to do this if our brain craps out at around 150 peeps?   It turns out that language came with some ancillary benefits.   Our representational systems allow us to categorise people into types easily – so we can quickly decide how best to behave in front of a particular individual (think uniforms on police officers… etc).  Secondly, they allow us to teach others how to behave in different types of circumstances – with none of this implying an increase on the upper bound of individuals we can personally keep track of.

So that’s Dunbar’s story as to how we circumvented the limit.  And it’s a good answer.  Unfortunately, however, it undermines any suggestion (implied by Dunbar and Adams as well) that the causal relation between brain size and network limits goes back the other way – from brain to group size.  Because as we’ve seen – the real causal determinant here is economic – not physiological.  The brain adapted to a particular economic reality.  The question is – how would the brain respond to a sudden drop in the signalling costs involved in group cohesion and maintenance?  And have we ever seen such a drop since language developed?

In my view – the answer to the latter question has been no, up until very recently.  Certain peripheral costs started dropping in a dramatic way around the beginning of the industrial revolution.  For instance – Dunbar did a study on groups involved in the sending of Christmas Cards.  He found that the Dunbar Number limit applied to these groups just as in many other cases.  The cost of sending Christmas Cards dropped dramatically upon the development of rail and road infrastructure – so this enabled one to signal to people that were physically a great distance away from you.  But the usual costs of writing the cards, employing the appropriate language – writing the symbols, keeping a list of who to buy cards for (either mentally or by means of a written ledger),  etc… that all remained constant.  Still much cheaper than spending an afternoon over a Christmas Ham, but relatively as expensive as it has ever been to signal to one another by means of language.

Enter online social networks.  We can now understand just what’s so revolutionary about them – they reduce the signalling costs involved in social cohesion.  The drop in costs are dramatic.  I don’t have to keep a mental list, or even a spreadsheet of my contacts.  I come across a person that I find interesting on Google+ and I just drag them to a circle.  I don’t have to laboriously track birthdays.  Facebook tells me when someone I know was born, and I can signal that I care about that person by writing on their wall – as everyone seems to do.  We know for a fact that signalling costs dropped dramatically since tens of millions of 30/40 something, estranged high-school friends suddenly started wishing one another happy birthday.  If I’m right, then there hasn’t been an equivalent drop in signalling costs since the birth of language.

There are still plenty of costs.  I spent pretty much an entire day trawling through Google+ profiles looking for people to add and only got up to around 60 people.   That’s way more people than I’ve ever been able to interact with on any level in such a short space of time – but it already feels like an enormous cost relative to the incredible pace at which those costs are dropping.

Over time these problems will get solved and signalling costs will continue to decline rapidly.  Robert Scoble, for instance, recently alerted his followers to a technology that will allow you to automatically group your contacts – so you don’t have to bear the cost of manually creating a taxonomy for them.  People seem to give Scoble lip for jumping on every new tech bandwagon – but I think he’s totally correct in this case.  We’ve only just begun to realise the potential benefits of this technology.

The last time something like this happened – around the dawn of language – humans managed to create institutional structures that changed the face of this planet.  This happened without any change in Dunbar’s Number.  So might a similarly massive drop in signalling costs allow us to create new institutions that progress us forward just as much?

Besides this – somewhat hyperbolic – supposition – we don’t actually know how the brain is going to respond to a massive drop in signalling costs.  Could it be that the 150 connection limit has been around for so long precisely because there hasn’t been an equivalent drop in signalling costs since?  After all – some of the costs are directly related to the effort the brain has to expend in order to track the people it wants to track.  It seems reasonable to suppose that,  given the large degree of automation of signalling related tasks the brain would otherwise would have to perform manually, the reduction of such costs would allow a larger amount of tracking.

Think about some of the concrete reductions in costs that social networks have achieved in terms of allowing you to quickly access information about your friends.  This New York Times article  by Clive Thompson describes how the invention of the Facebook Wall meant that you no longer had to surf from page to page to get information.  The ease with which we manage to garner information about our friends leads to what Thompson describes as an ambient awareness of a much larger group of people than that suggested by the Dunbar limit.

 

 

We Need a New Yardstick of a Real Relationship

Ah – but you reply, but these aren’t real relationships.  In fact, the practice of friending people on online social networks has been the object of some fantastic satire.  Cue this classic Daily Show Clip with Demetri Martin:

 

What is the source of suspicion that many people have with this kind of behaviour?  The intuition seems to be that the relationships formed aren’t ‘real’ in some sense.  And Dunbar’s Number is commonly used to defend this idea.  Dunbar and Adams claim that the same limits of human attention are to be seen in human interactions on online social networks.

From what I’ve been able to find – the data actually seems to be mixed.  I think we’re going to have to wait for more research to be conducted.  (Although I’ll be the first to admit that I’m not an expect in this literature so feel free to correct me – my larger point doesn’t depend on this.)  One study I read has observed an increase in the number from 150 to the 200-300 range in online social networks.   Results like this would support the view that the reduction in signalling costs will allow the brain to handle a larger number of connections.

A different study by Goncalves, Perra and Vespignani (GPV study for short here on out) measured social interactions using the Twitter Firehose API and came up with a result which supports the existence of the standard Dunbar Number as a constraint on human relationships.  I also believe that Dunbar himself has come out with similar research – although I haven’t been able to find it.

But even if it proves to be the case that the GPV study gets it right and the first study proves to be aberrant, this STILL doesn’t make the case that there is a biological limit for social relationships hard wired into us.  To see why – we’ll need to look at the GPV study in a little detail.

What the study did was analyse a bunch of tweets in a cross section of the whole twitter network.  They filtered out large uni-directional exchanges – like those between celebrities and their followers – and included only straightforward bi-directional exchanges between individuals.  Conversational nodes – i.e. tweets that begin a series of replies between participants – are identified, and the distance from the node of each tweet is measured.  The total distance of a conversational tree is called the ‘weight’ of that conversation and is taken to be a measure of the strength of the connection between participants.  What they saw was that the number of messages sent by participants to each contact maxed out at around 150 contacts.  After that point the number of tweets sent per contact declined.  As a result – the weight of each conversation suffered a corresponding decline, indicating a decreasing strength in the connection.  And so – because of this result the GPV study authors conclude that the Twitter data “offers support to Dunbar’s hypothesis of a biological limit to the number of relationships”.

So what’s wrong with this?  After all – it seems to support our intuitions that those folks that add thousand of people to their online social network are somehow behaving in a superficial way.  Well – one problem is that the authors are ignoring the fact that the amount of signalling is not necessarily a good measure of the quality of a relationship.

Human signalling in its very nature is arbitrary.  Our great innovation as a species was to employ signs that have no causal relationship to what they signify.  Thus the sentence: “I will protect you if other primates harass you” can signal your willingness to help another primate – even though you may not be in a grooming relationship.  But there is nothing forcing me to follow through on the promise.  It’s in this way that signalling behaviour tries to get value out of social relationships on the cheap.  And assuming that everyone is signalling honestly – it allows us to spread risk in a much more diverse way.  We can promise to help more people than we ever could possibly follow through upon – but that’s okay, because not everyone needs our attention all the time.  It’s like a bank that promises to pay anyone who decides to withdraw all their deposits.  That promise is only good on condition that not everyone decides to withdraw their capital all at once since the bank only keeps a fraction of those deposits on hand.  Assuming that disaster doesn’t happen – the bank is free to make use of the total capital sum deposited.  Similarly, we enjoy the advantage of having a wide range of people to potentially call upon should we need to through our signalling behaviours.  But we  know that we couldn’t possibly meet all obligations if everyone demanded our help all at once.

The problem is that due to the arbitrary and cheap nature of signalling, people signal stuff all the time that they never follow through upon.  So the value of a particular signalling procedure to the recipient comes from the sincere intentions of the signaller.  And the value of an act of signification to the signaller comes from the success in eliciting the desired increase in attachment from the receiver without a large degree of investment on the part of the signaller.  But the mere fact that a transmission of a signal between two parties has occurred does not establish whether or not either of these two conditions have obtained.

So let’s apply these insights to GPV study.  The study supposes that connections beyond the 150 limit suffer a loss in quality because of a reduction in the length of those exchanges.  But we don’t actually know much about the quality or not because we know nothing about the intentions or beliefs of the people involved.   It’s also going to be really difficult to gather this information.   Given that signalling procedures tend to be largely subconscious (as many like Adams argue),  how do we know for instance what a given engagement actually signals to the receiver?  How do we tell if they believed it?  How do we measure the sincerity of the signaller?

What’s more – there’s plenty of evidence to suggest that we engage less in signalling type behaviours with people with whom we have a greater degree of intimacy.  (I go further and define intimate relationships as those which  minimise arbitrary signalling behaviour – see my review of The Social Network).   Such a phenomenon is often interpreted as taking people for granted  (not always correctly in my opinion).

In many cases there are lots of factors determining whether or not a particular strategy of signification is likely to yield good results.  For instance, the single biggest predictor of being cheated on as a man among personality traits is agreeableness.  That might give you reason to think that a long twitter conversation between an agreeable man and his mate is in many cases actually weakening the relationship not strengthening it.

Similarly, many men will tell you that they received an increase in positive signalling behaviour from women when they reduce the  number of positive signals offered.  Successful dating profiles for men include pictures of men that aren’t smiling and are looking away from the camera.  There’s reason to think that people will signal a large amount to contacts that they value, when they doubt the level of the connection.  The flurry of signalling can be an attempt to remove this worry and strengthen bonds.   A high level of signalling, therefore, can actually imply a lower degree of connection between individuals in some cases.

So given cases such as these – it’s hard to maintain that Dunbar’s Number does anything like provide us with an upper bound on the number of strong relationships that a person can maintain.  Nevertheless, assuming the GPV study is replicated, it does seem to be telling us something.  But what exactly?  Well – just read it off the study, it’s all very clear.  It’s telling us the number of contacts with whom we can maintain a high degree of bi-directional signalling activity.  And given the restriction of the study to traditional, bi-directional exchanges, they are simultaneously selecting the kinds of exchanges that come the with same sort of economic signalling costs that have been with us since the dawn of language.

As such, the study can’t give us information about the limit on connections given a large drop in signalling costs.  Consider some of the ways in which these costs have dropped.  My favourite act is the act of following someone.  It’s a very low cost act – much lower than a short conversation, or even a single tweet.  Yet, it still is a signalling act.  What it means probably varies from person to person – but it’s not uncommon to think that to follow someone is to signal that you are interested in that person to some degree – that you will read at least some of their posts – that you might even share their content with your other contacts.  And while these might be intended as further acts of signalling that do take an investment of time – in many cases they won’t be.  If one values the content of the one you’re following – then reading it and sharing it has its own reward in value besides any signal that you’re a more devoted fan.  Now a mere click of that follow button probably doesn’t signal the sort of connection that would have to serve as godfather to their child – or even that you’d be willing to simply meet them for coffee – but it’s a connection of a sort and it has real value to the people that you follow.

Consider also signalling acts that are not generally bi-directional like when celebrities tweet to their followers.  Such acts play a powerful role in allowing personalities to signal their appreciation to their fans in a way that bi-directional relationships can’t possibly allow.  Celebrities can also establish a kind of faux intimacy by selecting a lucky fan with whom to engage directly in a bi-directional conversation – something that was difficult to do using television.

If you want an example – consider what is perhaps the most famous tweet of all time, Conan O’brien’s tweet about Sarah Killen.  This tweet made Sarah Killen an internet celebrity in an instant and resulted in an enormous amount of goodwill toward the Conan ‘brand’ for a gesture that cost him next to nothing.  Thus social networking has allowed for influential personalities to signal to their fans in ways that just weren’t possible before.  Now – again – these signalling acts don’t constitute relationships that can be characterised by traditional bi-directional signalling.  But they are real – and they have real value for the celebrities if not the fans as well.

As I mentioned, the GPV study excluded these sorts of uni-directional relationships from its analysis.  Thus it has its bias toward what it thinks are real connections built into the structure of the study.  Therefore at most it’s only telling us about a particular biological limit for a certain specific kind of signalling based relationship.  That’s a whole lot less interesting than the claim that there is a biological limit to relationships in general.

The discussion so far brings us to another important question.  Why do we tend to latch on to data like Dunbar’s Number as evidence of  value (or lack of it) in a relationship?   Because the economic cost of a signal is itself a signal of intentions and we’re going through a transition period while we are getting used to lower costs.  Dunbar’s Number measures limits on bi-directional communication given traditional signalling costs.  So a failure to match those costs by violating that limit signals to us that the intent is less than sincere.

Think of it this way:  when humans started using language to signal to one another that they would be there to defend against harassment – consider how shallow that might have appeared to those that used more expensive signalling procedures like grooming to signal their allegiance.  It would have appeared cheap!  And certainly at first one would be more suspicious of a person who only used language to signal loyalty and didn’t engage in grooming as well.  But obviously language did take off as a signalling medium because it’s the main one currently in use today.  And I think this is the same sort of thing that is going on now with the dramatic drop in signalling costs made possible by social networking.

 

The Hidden Narrative of Social Networking

 

Whether or not you understand the value of it, many people are making use of these new signalling opportunities.  What we need to understand is why.  At the same time we have to balance this against some of the other data that Adams has provided us.  Simple applications of data such as Dunbar’s number gets in the way of this understanding, rather than enhancing it.  It’s tempting to take particular data points and see in them aspects of our fundamental psychological reality.  But in actual fact, a true understanding of what is going on is still a long way away.

I get the impression that Adams is acutely aware of this since he points out that we don’t know yet whether online social networks should simply replicate existing offline social structures or seek to structure them differently.  This question is directly related to the issue of understanding the psychological reality which governs our lives.  Because to model online networks on the structure we currently observe in the offline variety is to assume (at least implicity) – that these observations represent the fundamental reality.

I agree with Adams that we don’t ultimately know yet what the answer should be – but obviously I lean toward the view that this is wrong.  I don’t doubt that our fundamental psychological nature is evolutionarily determined – but it’s a fallacy to infer from this that the surface phenomena we observe in human interaction are somehow static and insensitive to dramatic changes in context.  As I’ve argued above, the invention of social networking is itself a dramatic change in that context because of the reduction in signalling costs that it enables.

When provided with all these new tools tools of expression and social management we see an explosion of many different kinds of behaviours that aren’t common in the offline world.  It’s almost as if there was some latent human potential that we scarcely even dreamed was just sitting beneath the surface and waiting to be released.  Why chain it with our theories, when we can release it with our imagination?

It’s time to liberate ourselves from the view that our new found scientific perspective of our own nature dramatically limits the narratives we can craft for ourselves.  The potential of the human animal is vast and as yet unimagined.  There’s no need to try to cage it yet.