Deserialized Cognition

Part Of: [Deserialized Cognition] sequence
Followup To: [Why Serialization?]

You’ve heard the phrase “pre-conceived notion” before. Ever wonder what it means? Let’s figure it out!

Cognitive Style: Conceptiation

Your mind is capable of generating concepts. Let us name this active process conceptiation.

How does this process work in practice? There are only two ways concepts are created: from oneself (inference conceptiation), or from others (social conceptiation):

Deserialization- Conceptiation

Inference Conceptiation attempts to get at self-motivated, non-social cognition. You process information, you form a conclusion (a result), you save this result to memory, and then you pass it along to other cognitive process. Examples:

  • A scientist trying to make progress in string theory
  • An artist teaching herself to speak Spanish

Social Conceptiation summarizes the thought process of someone immersed in a more social setting. Examples:

  • An engineer picking up an proverb (e.g., “no analogy is perfect”) from Facebook, without thinking about it much.
  • A socialite half-listening to some guy at a dinner party describing nuanced work tasks.

During both types of conceptiation, concepts are saved to your long-term memory. Call this serialization.

Cognitive Style: Deserialization

You are a lazy thinker.  Don’t take it personally, though – so is everyone else. How can we explain our inner cognitive miser? It turns out that there are at least two biological reasons for this failing:

  • Brains are slow because they rely on chemical synapses; they run at 100Hz (vs the 2 billion Hz of computers)
  • Brains are metabolically expensive, burning 800% more energy than other organs (20% of total organismic load)

Serialization techniques (discussed previously) allow our brains to be lazy. Not all concepts need to be created from scratch; if, at some point in the past, you have acquired the requisite mindware, you can always resurrect it from long-term memory, in virtue of your built-in deserialization mechanism:

Deserialization- Deserialization

Two Inputs

As mentioned, deserialization (loading) only works if the requisite concepts have been serialized (saved) at some point in the past. Since serialization comes in two flavors, we can now refer to two different kinds of deserialization:

Deserialization- Deserialization Modes (1)

Call the former inference deserialization, and the latter social deserialization.

Application: “I Love You”

Imagine you were raised to believe in the importance of regular expressions of affection to your significant other (SO). So, you say “I love you” to him/her every day. At first, you are eager to tell them the reasons behind your feelings, but after a while, novelty becomes increasingly effortful. Eventually, you settle into a simple “I love you” before falling asleep. Fast forward two years, and your SO says “I don’t feel like you are being affectionate enough”. How can we explain this?

We are now equipped to describe the “I love you” pattern as an instance of deserialized cognition, no? This form of cognition (more specifically, a behavioral pattern) was established previously, and no longer requires active conceptiation to perform. Why should your SO wish for you to employ active processing, especially if such processing yields content very similar to your habituated behavior?

Why would your SO wish you reject serialized cognition? Here’s one path an explanation may take: such an override goes against the instinct of the cognitive miser. Costly signaling is a staple concept in ethology: effort filters between those who truly hold the recipient in high regard and those who only wish to appear that way.

Speaking more generally, it seems to me that our itch for originality come from precisely this will to demonstrate rejection of deserialized cognition.


In this post, we explored how the brain uses concepts via two distinct mechanisms:

  1. In conceptiation, the brain actively constructs & uses novel concepts.
  2. In deserialization, the brain simply reuses pre-existing concepts.

The brain also employs two different ways to create concepts:

  1. Some concepts are constructed by one’s own mind.
  2. Concepts constructed in a social setting are constructed externally, but are (optionally) evaluated by the self.

Putting these together, we are now equipped to refer to four different types of cognition.

Deserialization- Cognition Taxonomy (2)

This new vocabulary opens many doors to explanation, including the question “why do people value originality?”

Credit: Some of the ideas of this post come from previous speculations about cached thoughts. However, compared to deserialization, caching has a weaker analogy strength: concept reuse has precious little to do with enforcing consistency within a memory hierarchy.

Until next time!

Why Serialization?

Part Of: [Deserialized Cognition] sequence


Nietzsche once said:

My time has not yet come; some men are born posthumously.

Well, this post is “born posthumously” too: its purpose will become apparent by its successor. Today, we will be taking a rather brisk stroll through computer science, to introduce serialization. We will be guided by the following concept graph:

Concept Map To Serialization

On a personal note, I’m trying to make these posts shorter, based on feedback I’ve received recently. 🙂

Let’s begin.

Object-Oriented Programming (OOP)

In the long long ago, most software was cleanly divided between data structures and the code that manipulated them. Nowadays, software tends to bundle these two computational elements into smaller packages called objects. This new practice is typically labelled object-oriented programming (OOP).

OOP- Comparison to imperative style (1)

The new style, OOP, has three basic principles:

  1. Encapsulation. Functions and data that pertain to the same logical unit should be kept together.
  2. Inheritance. Objects may be arranged hierarchically; they may inherit information in more basic objects.
  3. Polymorphism. The same inter-object interface can be satisfied by more than one object.

Of these three principles, the first is most paradigmatic: programming is now conceived as a conversation between multiple actors. The other two simply elaborate the rules of this new playground.

None of this is particularly novel to software engineers. In fact, the ability to conjure up conversational ecosystems – e.g., the taxi company OOP system above – is a skill expected in practically all software engineering interviews.

CogSci Connection: Some argue that conversational ecosystems is not an arbitrary invention, but necessary to mitigate complexity.

State Transitions

Definition: Let state represent a complete description of the current situation. If I were to give you full knowledge of the state of an object, you could (in principle) reconstitute it.

During a program’s lifecycle, the state of an object may change over time. Suppose you are submitting data to the taxi software from the above illustration. When you give your address to the billing system, that object updates its state. Object state transitions, then, look something like this:

OOP- Object State Transitions

Memory Hierarchy

Ultimately, of course, both code and data are 1s and 0s. And information has to be physically embedded somewhere. You can do this in switches, gears, vacuum tubes, DNA, and entangled quantum particles: there is nothing sacred about the medium. Computer engineers tend to favor magnetic disks and silicon chips, for economic reasons. Now, regardless of the medium, what properties do we want out of an information vehicle? Here’s a tentative list:

  • Error resistant.
  • Inexpensive.
  • Non-volatile (preserve state even if power is lost).
  • Fast.

Engineers, never with a deficit of creativity, have invented dozens of such information vehicle technologies. Let’s evaluate four separate candidates, courtesy of Tableau. 🙂

memory technology comparison

Are any of these technologies dominant (superior to all other candidates, in every dimension)?

No. We are forced to make tradeoffs. Which technology do you choose? Or, to put it more realistically, what would you predict computer manufacturers have built, guided by our collective preferences?

The universe called. It says my question is misleading. Economic pressures have caused manufacturers to choose… several different vehicles. And no, I don’t mean embedding different programs into different mediums. Rather, we embed our programs into multiple vehicles at the same time. The memory hierarchy is a case study in redundancy.

CogSci Connection: I cannot answer why economics has gravitated towards this highly counter-intuitive solution? But, it is important to realize that the brain does the same thing! It houses a hierarchy of trace memory, working memory, and long-term memory. Why is duplication required here, as well? So many unanswered questions…


It is time to combine OOP and the memory hierarchy. We now imagine multiple programs, duplicated across several vehicles, living in your computer:

OOP- Memory Hierarchy

In the above illustration, we have two programs being duplicated in two different information vehicles (main memory and hard drive). The main memory is faster, so state transitions (changes made by the user, etc) land there first. This is represented by the mutating color within the objects of main memory. But what happens if someone trips on your power cord, unplugging your CPU before main memory can be copied to the hard drive? All changes to the objects are lost! How do we fix this?

One solution is serialization (known in some circles as marshalling). If we simply write down the entire state of an object, we would be able to re-create it later. Many serialization formats (competing techniques for how best to record state) exist. Here is an example in the JavaScript Object Notation (.json) format:

{“menu”: {
“id”: “file”,
“value”: “File”,
“popup”: {
“menuitem”: [
{“value”: “New”, “onclick”: “CreateNewDoc()”},
{“value”: “Open”, “onclick”: “OpenDoc()”},
{“value”: “Close”, “onclick”: “CloseDoc()”}


So far, we’ve motivated serialization by appealing to a computer losing power. Why else would we use this technique?

Let’s return to our taxi software example. If the software becomes very popular, perhaps too many people will want to use it at the same time. In such a scenario, it is typical for engineers to load balance: distribute the same software on multiple different CPUs. How could you copy the same objects across different computers? By serialization!

CogSci Connection: Let’s pretend for a moment that computers are people, and objects are concepts. … Notice anything similar to interpersonal communication? 🙂


In this post, we’ve been introduced to object-oriented programming, and how it changed software to becoming more like a conversation between agents. We also learned the surprising fact about memory: that duplicate hierarchies are economically superior to single solutions. Finally, we connected these ideas in our model of serialization: how the entire state of an object can be transcribed to enable future “resurrections”.

Along the way, we noted three parallels between computer science and psychology:

  1. It is possible that object-oriented programming was first discovered by natural selection, as it invented nervous systems.
  2. For mysterious reasons, your brain also implements a duplication-heavy memory hierarchy.
  3. Inter-process serialization closely resembles inter-personal communication.