> You can’t throw your hands up simply because you’ll never capture everything in your models.
You can however, decide that the key drivers in your domain are essentially impossible to capture quantitatively and decide not to model the domain scientifically. This applies especially well to cases where 'tacit knowledge' is important. Because that knowledge is hard to formulate, let alone formalize, it is really hard to quantize.
Taleb makes an argument in Black Swan that a bad model can do more harm than no model at all, and that "we can't throw our hands up" is not a valid excuse either: sometimes that's exactly the right call.
Right. It is also false to say there are no useful numbers in this space. For example: Dunbar’s number is 150.
But it is misleading to expect an employee to maintain relationships with 150 people—they also have a family and friends.
From the linked wikipedia:
> However, enormous 95% confidence intervals (4–520 and 2–336, respectively) implied that specifying any one number is futile.
what’s more interesting about dunbar’s number is that it suggests a potential maximum for the size of an effective organization (say, 520) rather than pinpointing an optimum. the idea of a maximum like this appeals to intuition, so it’s worth studying more (and more quantitatively), but opposes ambition, which is probably why we don’t have plentiful research in this area already.
I worked in quant finance for many years so I’m very familiar with low signal to noise in complex systems. You can’t throw your hands up simply because you’ll never capture everything in your models.
This field is so far from my areas of expertise but I imagine there are lots of smart people investigating and putting structure around these questions.