top of page
  • Writer's pictureEvan Warfel

People are Systems Too (Against "Systemizing Vs. Empathizing")

Updated: Dec 19, 2018


The topic today is that Google Memo. There have been many good things written about it, including in The Economist, as well as at Wired, and at Slatestarcodex.

Minor points first: For a memo which attempts to appeal to science, it's interesting that the author rarely, if ever, uses the concrete numbers and precise descriptions of what he cites as evidence.

Also, the conclusions of memo are wrong; the memo isn't that well written and riddled with cherry picked as well as factually incorrect claims; it contains numerous imprecisions and dubious logic; the footnotes are only vaguely associated with the text in question; various parts of the memo are impossible to follow; and the author's pretense of neutral objectivity is easily seen through.

For example early on in the memo, the author writes that that liberal bias is evidenced via "compassion for the weak." The number of incorrect things presupposed by this sentence alone... For the record, It's not compassion if it's a chore. It isn't compassion if you are thinking about "weak" vs. "strong." (If I had to guess, I would bet that this definition of compassion is reflective of an earlier Loevinger stage of ego development.)

Also, generally speaking, if you are male and there is a gendered title for your role, try flipping the label around -- how would it feel to be called a frontierswoman, or a priestess?

Part 1.

Anyways, while several aspects of the Google memo stood out, I want to focus on one. The author makes repeated references to a "systemizing vs. empathizing" or "things vs. people" orientation.

This is an apparent reference to a theory proposed by Borat's cousin, Simon Baron-Cohen, and some of his colleagues. Briefly, they have postulated that there are two dimensions of people worth considering -- the ability to empathize with another person, and something called "systemizing." According to Baron-Cohen and friends, these dimensions are completely independent from each other. You can be oriented towards one, or the other, or both, or neither. Right from the start, the Google author gets this wrong and assumes that "empathizing" and "systemizing" are opposites (as well as "fixed"), not independent and malleable ones. So much for that.

Baron-Cohen theorize that Autism can be thought about as a location in this two-dimensional space, of being oriented towards systemizing, and separately, away from empathizing. This theory is also alleged to "explain cognitive differences" between men women.

It turns out the definitions that underpin the theory and its conclusions are weak, and I want to explore why.

In general, when considering a term to used describe people, the best way to understand what some person-oriented researcher means by it (be they a psychologist, a sociologist, an economist, and educator or business-school type, etc.) is to look at what they actually measure.

To see what "Systemizing" might refer to, here are some statements (technically known as "items") from the "Systemizing Quotient" scale:

* When I look at a mountain, I think about how precisely it was formed. (SIC)

* When traveling by train, I often wonder exactly how the rail networks are coordinated.

* I find it easy to grasp exactly how odds work in betting.

* I know very little about the different stages of the legislation process in my country. (Reversed)

* I am interested in knowing the path a river takes from its source to the sea.

* When I listen to a piece of music, I always notice the way it's structured.

(I put "sic" in capital letters because that particular item is either missing a key comma or absurd, and I wanted to express my feelings about sloppy psychometric items.)

Note that these items attempt to measure preferences, not differences behavioral tendencies, nor in how a person sees the world or how they think (unlike Locus of Control). And if my experience researching and creating psychometric scales is any guide, each of these levels (differences in preferences > differences in behavioral tendencies > differences in how a person thinks) is more fruitful than the previous one.

But let me make some things explicit. Taken individually, these items don't seem to measure much more than "certain types of cerebral curiosity." While this may nor many not be a problem for Baron-Cohen et al., it is a problem for the google memo.

To pick four items in particular:

Grasping exactly how odds work in betting is something covered in every undergraduate math program and likely the math requirements for high school.

I know very little about the different stages of the legislation process in my country... Are there no females in politics, nor female lawyers?

I am interested in knowing the path a river takes from its source to the sea... Are there no female geographers?

When I listen to a piece of music, I always notice the way it's structured... Are there no female musicians or composers?

Don't these items reflect a sort of curiosity which is likely filtered for by Google's interview and selection process anyways?

Furthermore, the Baron-Cohen measure intimates that a "systemizing quotient" is an inherent property of humans, even though preferences like the kinds measured here are socially conditioned! Moreover, if the authors wanted their items to the measure the thing they are claiming to measure, they probably would have benefited from interviewing female philosophers, female lawyers, female politicians, female writers, and so on.

Part 2.

Taken as a whole, the items and the scale becomes a little more convincing. Overall, it seems that they are measuring a coherent construct. (At this point, though, it still might not be anything than "Specific and Socially Conditioned Cerebral Curiosity.")

Reading through the scale, the following two things become transparent:

A) the author of the google memo is using a theory about Autism to stand in for a theory about people that itself doesn't seem well-defined (as the best definition of what you are working with is what you measure, not what you want to be measuring) to make statement about a company with a hefty selection process.

B) It's as if Baron-Cohen et al. don't know too much about systems, nor about measuring any "skill" relating to them.

Maybe that's the point. Maybe, for Baron-Cohen et al.'s theory, a preference for thinking about "systems" is what matters when thinking about Autism and "cognitive gender differences." But the measure isn't titled the "Cerebral Curiosity about 'Systems' Quotient." So we are left to contemplate what the authors think their scale is measuring.

Luckily for us, the authors give several definitions of "systemizing."

The first definition they give is clearly hyperbolic and seemingly inappropriate for a scientific journal:

"Systemizing is held to be our most powerful way of understanding and predicting the law-governed inanimate universe." Please.

The second definition is better, though it still seems to be about curiosity:

"Systemizing is the drive to analyze the variables in a system, to derive the underlying rules that govern the behavior of a system." (Emphasis added.)

The third is self-referential and thus complete crap as a definition:

Systemizing also refers to the drive to construct systems. (Which begs the question, what's a system but the product of systemizing? Or is it the other way around?)

The fourth is so imprecise it's useless:

Systemizing allows you to predict the behavior of a system, and to control it. Here, "allows", "predict", "behavior", "system" and "control" all could stand to be defined.

So far, the only remotely useable definition is the second one. Following this, the authors write that

A system is defined as something that takes inputs, which can then be operated on in variable ways, to deliver different outputs in a rule-governed way.

Call me naive, but if your definition could conceivably pick out every known thing in the universe, it isn't a great definition. A rock can take inputs (like blows from a hammer), variably distribute the force, and deliver outputs "in a rule-governed way." By their definition, a goldfish, the planet neptune and probiotic yogurt are all systems; any further distinctions are irrelevant. Moreover, the connection between inputs, rule-governed processes, and outputs is time. You know what exists in time? Everything.

Anyways, the authors then go on to illustrate their idea that systems can be described in terms of "Input" "Operation" and "Output." They give the following example of a "a tennis stroke" system:


Hit ball → Top spin → Ball bounces left

Hit ball → Back spin → Ball bounces right

Hit ball → No spin → Ball bounces forward

To call this a drastic oversimplification of "the racket-ball-ground-physics-motor-and-nervous-system tennis system" is put it too mildly. This "system" is even simpler than the model used in Mario and Wii Tennis. It's a cartoon of a cartoon system. "Hit ball" as an "input"? Come on.

Also, this system is either factually incorrect, or the authors should fire their tennis teacher. Top spin makes the ball appear to bounce back to the ground more quickly than gravity alone would dictate. Back spin causes the ball to hang in the air longer after it bounces at you. No spin causes it to bounce normally.

They also give the following math example:


3 → Squared → 9

3 → Cubed → 27

3 → Inverse → 0.3

Note that this is not an oversimplification, that you really can do (a limited form of) math with only one input.

The authors' claim seems to be that "systemizing" involves enumerating possible states of some defined set of variables, and then thinking about what the consequences of each state might be. (In other areas of psychology, this is called Disjunctive Thinking. The best example of disjunctive thinking I've seen is from the seventh season of Game of Thrones, when in episode three (spoiler alert), Littlefinger counsels Sansa Stark. I have not seen anything connecting disjunctive thinking to either Autism or gender.)

Minor quibbles aside (like that the only systems with "one input," like the math example, are people-made), this claim is decently precise. Why the authors then fail to make a test that operationalize their examples, and instead fall back to measuring preferences, escapes me.

Even more confusing, however, is their next claim, where they give the following as an example of not a system:


Jane → Birthday → Relaxes

Jane → Birthday → Withdraws

Jane → Birthday → Laughs

Jane → Birthday → Cries

The authors claim that the same input is producing different outputs. Clearly, the inputs aren't the same, though the label that refers to them is. Also, how is "Birthday" an operation? Surely "birthday" would be a conditioned input? Between this and the tennis example, it's as if the authors are using "system" as a metaphor for a system and not even realizing this is what they are doing. (This makes me speculate: Do these authors not know how to code?)

The authors also acknowledge that one would need to take into account Jane's mental states in order to make accurate predictions. Yet if math is a system of invisible, conceptual operations, what's so privileged about invisible mental states? Remember, this whole "systemizing" idea is supposed to be different and independent from "empathizing," even though they end up using one to define the other.

Part 3.

In truth, any distinction between "systems" and "empathy" is, at best, a poorly named one. Because people are systems too. Tell any medical student otherwise and you'll be laughed at. Complex and confusing systems we may be, but we are systems (or systems of systems, etc.) nonetheless.

Speaking of which, do you know what's even more complex than interacting with a person? Interacting with a person who doesn't yet speak any language. Yet despite this, it is apparently rare to be a male in Pediatrics.

If both people and computers are systems (as opposed to what the Google Memo and possibly Baron-Cohen et al. would have you believe), it actually makes more sense to think about (systemic) differences between these kinds of systems. I can think of three off the top of my head, and I will use "people" and "computers" to illustrate them.

While not an exhaustive list of "important dimensions of systems" it's conceivable that many different systems could be located in this three dimensional 'systems-space.' In addition, people could also be located in this space via their preferences for working with different kinds of systems. Perhaps people on the Autism spectrum would cluster in one corner of the space. Perhaps different preferences for different types of experiential-information-theoretic different systems will one day be explained by neuroscience.

In the meantime, it seems safe to say that people are just more complex systems, and people who work with people are doing something more complex. How do reverse gender disparities in the teaching and healing professions look now?

(Again, not only are preferences often conditioned, they can often be changed.)

My point is that "Systemizing" and "Empathizing" as defined by Baron-Cohen et al. is a set of concepts not worth thinking about. Despite their claims, it's not clear if these two things refer to two ends of a spectrum (in which case being unempathic gives us trouble) or if they are orthogonal (in which case there are four quadrants worth exploring, something which the authors barely did explicitly, and seemed to have confused implicitly), or something else entirely.

Part 4.

Just because a scientist uses a measure that someone else criticizes shouldn't mean one should necessarily doubt the scientist's acumen. This property (the merits of a proposed theory) can be abstracted from any judgement about the individual. (Not everyone is practiced at doing this, and so don't ask my why jury selection processes don't explicitly take this skill into account, even though it can be measured.)

In the spirit of charity, let's now give the Baron-Cohen theory the best possible argument and evaluate it on those merits. After all, sometimes (but not always) people reach the right conclusions even though they are unable to precisely articulate how they got there.

First, let's assume that "Systemizing" actually means "orientation towards thinking in terms of, as well as an ability to articulate novel frameworks." (Notice this isn't a drive, it's a skill). Second, let's assume that being skilled at "Frameworking" has nothing at all to do with being skilled "Empathizing."

Now we can explore the following four quadrants.

* High on both Frameworking and Empathizing: such people will likely come up with structured theories about how people work. (Any stage based model of human development comes to mind.)

* High on Frameworking and Low on Empathizing: What Baron-Cohen et al. allege captures Autism and STEM related things

* High on Empathizing and Low Frameworking: An empathic type who likely cares less about the abstract principles behind things? Do artists not have principles? I'm having trouble putting "artists" in this category without splitting Frameworking into subdimensions.

* Low on both: someone who is unable to understand others or the world? Maybe if you miss the critical windows for learning languages and have an underdeveloped capacity for abstraction you'd fall into this quadrant?

Parts of this theory sounds alright, but there are clearly problems with it as a whole. One of which is that empathy requires frameworks (like, say a framework of emotions!) in order to relate to what another person is going through. Another is that though "frameworking" is better than "systemizing," it still isn't clear.

Okay, so 'frameworking' is out as a better version of "Systemizing." But if you remember, "Systemizing" is different from what the items in the "Systemizing Quotient" scale attempt to measure. So maybe there are better ways to capture what Baron-Cohen et al.'s Systemizing measure points to. Assuming one can control for societally conditioned preferences, would it be more fruitful to think about...

... an Orientation Towards Applied Enumeration? This could explain things like a tendency to enumerate types of plants or trains.

... or an Orientation Towards the Complex Application of Relatively Simple Starting Rules?

... or how about natural systems vs. people-made systems?

These seem more fruitful, but they bring us back to the dimensions of systems I've proposed in the previous section. "Relatively Simple Rules" could be defined in terms of information complexity / density, dimensionality, and level of abstraction. Similarly, "natural systems" vs. "people-made systems" could also be classified along these three dimensions, too.

I ask these questions because despite the imprecise and unhelpful name, there seems to be some sort of conceptual structure that the Systematizing Quotient is attempting to measure. It would be fascinating if preferences for learning or for certain experiences or for thinking with certain types of conceptual structures were found to be due to neuronal branching and connection patterns.

Caveat: I am not familiar with the recent literature on Autism, and the content of this post is purely derived from the material mentioned in it.


bottom of page