Humans live in groups. So do many other animals, fungi and bacteria. Nature is awash with collectives, in fact, in which individuals communicate and signal to each other. Over the last hundred years, science has attested to the enormous benefits of pooling information across individuals – after all, how else would you choose a new phone, if it weren’t for all those Amazon reviews? It therefore seems pretty clear that there are considerable advantages to paying close attention to the behaviour of others when making your own decisions… Right?
Perhaps not as much as we think, according to a new theoretical study by scientists working at Princeton and Exeter Universities. Interested by the common occurrence of surprisingly preventable disasters, such as the 2008 financial collapse and the Challenger shuttle accident, the team wanted to find out whether there was perhaps a relatively simple explanation for why individuals sometimes seem to forgo their own best judgement when they are working with other people. Their research now shows, they say, that not only do individuals in groups often behave in a way that negatively affects them, but they might just have evolved that way.
At the root of the problem is the fact that humans and other animals often seem to behave very strangely, particularly when they are in groups. That is to say that it’s difficult to come up with biological explanations for why humans do certain things, if their aim is to stay alive (which we generally believe it is, for the purposes of scientific enquiry). Scientists have studied the phenomenon of apparently “irrational” behaviour in a number of social situations, but one of the most familiar comes from the everyday occurrence of getting from one side of the street to another without being hit by a bus.
In late 2009, a team led by behavioural scientist Jens Krause spent what must have been a rather long few days watching people cross a main road in Leeds. They noticed that pedestrians frequently stepped out into the road when the cross signal was red, and observed a few near-misses between vehicles and people in the process. But looking at the data more closely, they found that the probability of someone stepping into the road was much greater if another pedestrian had already done so in the previous few seconds. People were often more strongly influenced, it seemed, by the neighbours’ actions than by their own good sense.
This result might sound familiar – many of us have stood at the side of the pavement at the red signal, only to step out into the road without so much as a glance at the oncoming traffic, confident that if other people are crossing the road, “it must be fine”. But why are we so swayed by our neighbours’ behaviour, if we can judge the safety of the situation for ourselves?
To answer this question, the Princeton researchers built a theoretical model of a group of “deliberately abstract” individuals which each have to make multiple choices between two possibilities, A and B. Importantly, at each point in time, one of these options is “correct” (you can think of it as getting a reward, or not getting hit by a bus), and the other is “incorrect” (entailing no reward or a collision). The right decision is not constant, but changes between A and B at a rate unknown to the group.
In deciding what to do, each individual could make its own estimate of the right choice, but could also look at the choices made by its neighbours to inform its decision. The researchers introduced a specific term in their model, called a “weight”, which controlled how much importance each individual gave to the choices of its neighbours. In road signal terms, a low weight means that you cross the road when you have judged it safe to cross the road, while a high weight means you’re easily convinced to cross the road if many other people seem to be doing so. Importantly, the “weighting” means that there’s a trade-off going on here: your opinion is discounted against your neighbours, so you can’t make good independent decisions if you’re using all your attention to copy others.
Instead of simply assigning fixed parameters to each individual in the group, the researchers designed the model in such a way that the group itself could ‘evolve’. The model allowed individuals that were better at making the right choice to be better represented in the next generation of the group as compared to individuals that often chose poorly.
Over time, the composition of the group changed from a collection of independent individuals that relied on their own evaluations to a more and more socially dependent group. But intriguingly, the authors found that performance was not correlated with sociality in the way one might have expected.
To begin with, the dependence of individuals in the group on their neighbours gradually increased, and so did group performance. Pooling information from neighbours, it seemed, initially conferred quite dramatic increases in the number of individuals choosing the right option. However, at a certain point, group performance began to decrease – but the sociality of individuals continued to increase. This continued increase meant that individuals in the group were doing worse and worse, but continued to evolve into more and more social versions of themselves. When the group finally settled at a stable value of the weighting parameter (i.e. it ceased to evolve one way or another), the average level of the weight ended up being much higher than it was when the group was attaining its highest accuracy in predicting options A or B.
The explanation, according to the study, has to do with the fact that at a certain average level of “social influence”, the practice of copying your neighbours starts to overwhelm the initial benefits of learning from them. In the model, it isn’t clear to an individual whether its neighbours are doing what they’re doing because of independent assessment of the options, or because they are socially influenced by their neighbours. Or put another way, you might not know if the person to your right is stepping out into the road because she has carefully judged the risk of being hit by a bus, or simply because the person to the right of her started crossing the road just moments before. Of course, if all individuals copy everyone else instead of making their own assessments, then each neighbour’s decision is extremely uninformative about the truly correct thing to do.
Now, if we could just tweak the weighting parameters, we could make individuals a little less reliant on each other’s opinions, and force them to use their own judgement. That tweaking would force a bit more independence from our individuals, and a more accurate group would be the result. But of course, evolution doesn’t work like that. Once an individual has evolved dependence – i.e. a high weight on its neighbours’ behaviour – it becomes trapped. A small decrease in dependence (at the slow rate we can expect in a naturally evolving population) is unlikely to make an individual better at guessing the right answer, because it has already invested so much in watching its neighbours that it isn’t very good at assessing options for itself any more.
“Our findings challenge the notion that we should expect animal groups to be ‘tuned’ to respond optimally to environmental information,” say the authors, throwing a small spanner into the scientific and media works churning out articles about the remarkable evolution of swarm intelligence and the wisdom and the crowds. “In the context of human interaction, our results suggest that when social information is available, individuals will over use it.”
It might be calming to imagine that this study could help to return some independence to the individual in our highly networked world, where every purchase is reviewed and every decision publicly scrutinized on Facebook or Twitter. We could return to a more isolated state, where reflection and personal judgement win out over fashion and social influence…
But of course, such an opinion would have to be approved by the masses for it to have any real effect.