I’m indebted for this insight to Duncan J. WATTS (2011): Everything is obvious (once you know the answer). How common sense fails. Atlantic Books, London (p. 76 ff.)
A bunch of people are separately made to listen to a series of songs and then asked to rank them. Individual rankings are combined to yield the Top 5 songs for the group. And this for many groups.
Now assume that, as each participant steps up to listen, he is informed about the “provisional” Top 5, as established by aggregating the opinions of those in his group who came before him. In making his choice each participant may then rely on his own taste as well as the aggregate taste of his “predecessors”. In this way individual opinions may resonate down the sequence of choices.
If people have firm tastes in music, knowing what other people like would not affect choices. If, however, people are swayed, we get cumulative effects. Early opinions impact on those coming next in cascade fashion – especially early votes “matter” for the subsequent evolution, for the resonate longer. The question is: what will be the cumulative effect?
Thanks to internet it is possible nowadays to run such experiments in statistically meaningful manner. Depending on the sequence of predecessors, the “provisional Top 5” history will vary for each experimental run. This unique history will impact on each subsequent participant, and so on down the path of choices, in cumulative fashion. The details are complicated – and the results surprising.
What is the statistically measurable outcome? The choices become more unequal: votes tend to agglutinate more readily around this or that song. When influences resonate, we conclude, runaway hits are more likely. The choices also become more unpredictable: when we repeat the experiment the top choice can look quite different each time.
The taste of participants does not change. The only thing that changes is the nudge each participant receives from the chained opinion of his predecessors as he takes his turn. Such chains change by chance from one run to the other. Just as with riots, the initial conditions matter, and over the full length of the experiment they bring about quite different outcomes. When opinions resonate, chance can dominate group choices irrespective of what each individual opined to begin with.
The first lesson is that when opinions resonate it is no longer possible to predict the outcome, nor is it possible to work backward to the source to identify the “cause” – the whole chain matters.
The second lesson is that even full knowledge of individual opinions does not allow us to predict the outcome. Rather, the outcome emerges from the interplay of opinions – and their chanced sequence. Why should it matter? Well, it just does. It is an “emergent” property.
Counter-intuitive? Yes! We believe we DO recognize song quality, don’t we all?
We get into a circular argument. Song X won out because that’s what people preferred. And how do we know that people really liked song X? Because song X won.
There is a third lesson to be drawn. Success is often the result of “learning by doing” – and cannot be predicted. Take the ubiquitous “post-it” yellow sticker. When first sold – from a catalogue – sales were miserable. Who wants to buy “scrap paper”? To get sales going those yellow pads had to be given away for free to potential users. It was the users who discovered their usefulness, and made them irreplaceable.
Even the inventor does not image the uses his invention will be put to. Famously, when mainframe computers were first developed, IBM’s president predicted the company would not sell more than five machines…
 Why do people riot? https://deepdip.wordpress.com/2012/03/05/why-do-crowds-riot/