The Good Employee, a story about how you can explain companies with graph theory

1) First, let's look at the premise:

> But this time, let’s analyze the things from a mathematical perspective - because today if you don’t speak about data and models, nobody hears you. Let’s give the theoretical machine learning definition of a modern AI company: a modern AI company is a WUG - a.k.a. an Weighted Undirected Graph - that links at least three different families of nodes - Employees, Processes and Projects - in a chain of skills, troubles, goals, beers, prizes, achievements, careers, promotions, pizza, parties, lies and God only knows what else, with the overall goal to solve (or introduce) dependencies, investigate (or ignore) consequences but, by the end of the day, feed the so hungry desire to “change something” - also referred to as “bring something in production”.

This is nonsense, not to mention completely arbitrary. If you're going to do this kind of thing (which game theorists actually do routinely), you need to very carefully define your terms. I have no idea what a process is; I have no idea what a project is; I have no idea what an employee even is (are contractors employees? they naturally have different incentives than full-time or part-time employees). Etc. I could make a just-as-valid post arguing that a company is actually a n-dimensional topological space (and then using vectors instead of edges/nodes, while getting totally different "results"), but what the hell does that mean?

2) Next, we have some (yet again) arbitrary definitions:

I'm especially confused about why things are proportional to other things, namely budget, cost, or headcount. Again, if you're going to try to provide a mathematical model for social behaviors, you better have some solid justification behind your definitions.

3) The assumption that humans are robots:

> Ok, so in a digital Forest like the one just described, you cannot move with weapons to defend yourself against pumas - or whatever else lives inside a real forest with the desire of eating you and your backpack full of energy bars. Instead, the good Employee inside the WUG. Sorry, the Forest. Sorry again, in the Company, uses his ability and applies the Prim’s algorithm. Now, before going ahead, it could be useful to remind some concepts about graph exploration.

On the face of it, this seems (?) right, I guess. But when you think about it for more than 30 seconds, you realize what an insane claim it actually is. Humans are famously non-rational agents. The idea that humans subconsciously apply Prim's theorem is experimentally wrong... in fact, people are notoriously awful long-term decision makers, but I digress.

4) The weights table

Listen, I get it. Sometimes, you need to model social behavior. And to do that, you need to assign weight values to decisions. Again, game theorists do this all the time. But there's a few differences from a GT paper you'd read and this blog post. First of all, dude... like 15 weights? Really? This goes beyond speculative.

And second of all, we need justifications! Why is "pizza" -15, but "parties" are -25? These systems are actually pretty sensitive to initial conditions; in fact, the more weights/nodes, the more sensitive we'll be to initial conditions.

> Of course, you can add as many weights as you want.

Yeah, you could, but you don't. In fact, the idea when creating these kinds of models is to try to minimize these kinds of arbitrary weights.

5) People are robots.. again

Author claims "...yes, some of them apply Djistrka[sic]..." -- again, this is simply experimentally wrong. And hearing it out loud just sound so awkward. I don't even think I "apply Dijkstra" when I walk back to my car.

6) Making policy based on math is stupid

> This state of confusion leads to the fear of hiring people who are not expert in something, that is not able to solve company’s actual problem: that problem arose yesterday, come out months before, without nobody looking in the right direction, because the minimum_spanning_tree rules them all

No, the minimum_spanning_tree does not rule them all. Making company (or worse, political) policy based on these kinds of analyses is a plague. And sadly, these kinds of reductive and abstract models (often hailed as "data-driven") are tone-deaf and completely wrong.