When to imitate, and when to experiment

Take a look at this picture:

It’s a picture of an iPhone next to a picture of several Samsung phones. It was submitted by Apple as evidence in a lawsuit claiming Samsung copied Apple’s design.

Whatever position you take on the lawsuit, you can recognize the notion behind it, conveying something like: “Don’t copy, create your own, create something better and thrive.”

Now check out this table:


This table is a comparison of bow-and-arrow hunting technology from two groups of nomadic foragers – the Hadza of Tanzania and the San of the Kalahari. The table is from a chapter (“A Cultural Species”) by Joseph Henrich in the book Explaining Culture Scientifically. Henrich included the table to highlight the differences between the groups, as an example of inter-group variability in the human species. (These are two groups with presumably little or no interaction between them – so I don’t mean the table to be comparable to the Apple-Samsung comparison).

Now here’s a passage discussing the table that I found thoroughly thought-provoking:

“Clearly, the skills, detailed knowledge, and procedures that go into manufacturing this equipment…is acquired predominately through some kind of imitative learning process. There is no way an individual can figure out all the details (such as where to find the beetle larva, or which branches to boil) that go into making a successful hunting kit, without learning extensively from others. In all the ethnography on Kalahari San foragers, we see no evidence that Kalahari hunters have experimented with longer (more than two meters) bows and fletched arrows, only to later reject these alterations. Woodburn (1970: 14) reports that enterprising Hadza have occasionally manufactured their bow staves from woods other than mutateko, but have always returned to mutateko –that is, most Hadza have never experimented with alternative woods. Kalahari foragers have not been observed to routinely test a range of beetles, seeds and branches for their poison-making possibilities – they just learn to gather and process chrysomelid beetles from other group members…Instead, although individual variation certainly exists, much of the variation is between groups. More importantly, detailed ethnographic observations corroborate this inference by showing that such manufacturing skills are acquired through a process of imitation and practice, not by free-ranging individual experimentation (Fiske 1998).” 

Of course most of us are still a whole lot more like the average Hadza and San (or the Samsung engineers?) than we are like the Apple engineers who designed the iPhone.  But I think it’s kind of fascinating how generally recognizable and intuitively attractive the process of experimentation and the introduction of variation have become.

Even though experimentation and variation are attractive, straightforward imitative learning makes much more immediate sense. Learning is simply a lot more efficient. We go with what others do because we know it works (and for other reasons too). And since we never ever have enough time to do all the things we’d like to do, knowing something works is a huge advantage that isn’t to be dismissed lightly. So when do we overcome the inertia-induced and sure-thing appeal of imitation in favor of the risk of experimentation?

This is a decision I find myself facing all the time when dealing with new large datasets. When something needs to be fixed, or standardized, or cleaned, there’s usually the direct and manual way of proceeding (for example, using Excel (ahhh!)…or worse, asking a research assistant to use Excel) and sometimes there’s the possibility of creating a program or lines of code for doing it faster, ideally something automated. But the automated approach means taking time that isn’t directly contributing to the research objective to create the automation, and it means the possibility that it might not work anyway. Sadly, I take the manual and laborious but safe route far, far too often.

Taking that route makes sense. It’s more interesting, because less understandable, that I tend to feel embarrassment about doing so, especially when there are clear examples of alternative contexts where I wouldn’t have felt any problem with, say, collecting the berries for poison the same why my father and neighbors do. “Create something new that saves time!?” It’s interesting to think about how my embarrassment, and Samsung’s embarrassment (which though different, probably share a common foundation), emerged.

For certain reasons I’m fairly frequently exposed to the start-up world in Cambridge and Boston. In that world it’s “build something new or die” (professionally). That tendency to experiment within the community is awesome. But there have got to be accessory conditions that make it possible. For example, it’s also the case that start-ups hold certain glamour. So in a way, professional experimenters don’t just take on the risk of failure in exchange for the exceedingly slight possibility of creating some new successful technology. In the short-term they get some return on their effort from the value associated with their role, and that short-term value probably helps them sustain their risky experimentation.

Here’s how Henrich describes what the Hadza would look like if “the acquisition of the adaptive repertoire were principally a product of individual learning” (the hypothesis he’s rejecting):

“every person would have to go through a trial and error process in which fletching, various potential poisons, and different bow sizes were tested. This stochastic process, based on a small number of trials and riddled with errors, would generate massive within-group inter-individual variation, as different hunters would find different poisons and different-sized bows most effective.” 

In the general population, that would be a nightmare. It isn’t even possible. But if you squint, that description looks like Boston’s start-up world (or the R community!). And I’m glad we have it.