MBTI Isn't Rigorous Science, But It Isn't Astrology Either

Astrology classifies people from the outside by date of birth, ignoring far too many important, related factors. As far as the formation of personality/character goes, the influence of one’s birthday is probably negligible.
MBTI, at the very least, attempts to describe psychological preferences through questionnaires and theoretical dimensions, rather than being completely detached from measurement and ignoring obvious strongly-correlated factors the way astrology is.
Its problem is that the quality, stability, and theoretical foundation of its measurements are all fairly limited — not that it isn’t measuring anything at all.

Understanding MBTI’s Classification Scheme

I / E (Introversion / Extraversion)
S / N (Sensing / Intuition)
T / F (Thinking / Feeling)
J / P (Judging / Perceiving)

If you go just by these few letters, or by the handful of phrases that get passed around online, you’re naturally going to get serious distortion. It’s like if I say F=ma — even if I tell you F is force, m is mass, a is acceleration, that doesn’t mean the other person actually understands classical mechanics. That said, “popular understanding is often distorted” doesn’t, by itself, imply “MBTI theory is therefore correct.” We still have to evaluate things on their merits.

For ease of understanding, we can use this kind of approximate division to describe a person’s mental activity:

Self-related / external-related;
More concrete / more abstract;
More logical (people-”independent”) / more emotional (people-related);

Of course, you might say we could subdivide “others” further into: family, friends, coworkers, superiors… and sure, you could cook up a 64-personality system, right? But the fact that classification can be made finer and finer doesn’t mean that simple classification is pseudoscience.

I/E

If we follow this simplified framework, MBTI roughly treats I / E as: whether a person’s attention and psychological energy more often returns to the subjective interior, or more often heads outward into interaction.

But this is just a useful approximation, not MBTI’s strictest definition. The controversy around MBTI isn’t about “whether people can be classified,” but about whether these dimensions are accurately measured, whether they’re stable, and whether they actually have enough explanatory power.

This “psychological energy” thing is easy to write off as mysticism, but it’s really not mystical at all. For example: a day at the office is exhausting, and an I-type can come home, lie down for a bit — even without scrolling their phone — and feel “recharged.” But if you told an E-type to lie there at home without sleeping, they’d just get more and more irritable and restless. Send them out with friends instead, and suddenly they have energy again. That’s hard for an I-type to imagine (when an I-type feels mentally drained, they almost never feel like socializing). Even if an I-type might appear quite talkative in social settings, the moment they get home, they feel: finally, I can rest, this is so nice.

That said, this really doesn’t have a solid systematic foundation in neuroscience theory and practice, so it does feel a bit like a castle in the air — but it isn’t mysticism.

S/N OR T/F

These two pairs are easy to confuse with T/F, because S (Sensing) sounds a lot like F, doesn’t it? And N sounds a lot like T. But there are many fundamental differences between them.

The Official Explanation

N and S are closer to “perceiving preferences.”
T and F are closer to “judging preferences.”

The common framing is that N / S isn’t about people having different sensory organs, but about whether, when taking in and attending to information, a person leans more toward concrete facts, details, and experience, or more toward patterns, associations, and possibilities.

T / F also isn’t “rational vs. irrational.” It’s about whether, when making judgments, a person leans more toward consistency and logical standards, or more toward values, relationships, and how things affect people’s feelings.

My Take

The First Difference

A frequently cited explanation goes: N and S describe how you take in information; T and F describe how you process it. That framing is too crude and easily misleads people. Because we humans all have roughly the same sensors, and we take in information in similar ways — vision, hearing, touch… There may be differences, but that’s not really the issue here. In fact, both pairs describe processing of information; the difference with S/N is more about what aspects of information you focus on and how you organize it.

N and S belong to “passive” modes (tendencies) of processing information.
T and F belong to “active” modes (tendencies) of processing information.

What does “passive” mean? I used to find this very hard to grasp.

Until AI (LLMs)! An LLM has inputs and outputs, but we don’t understand how internally it converts the input into the output (note I’m saying we don’t understand, not we don’t know — humans, especially people in the field, of course know what tensor operations and nonlinear computations the LLM’s input goes through to produce the output, but we don’t really get it, don’t comprehend, don’t grasp why this should produce that output). Before LLMs, though, traditional algorithms could be understood. Some people might fail to learn them, but someone always could — we break the problem down step by step, this way and that, and then solve it. That’s something that can be understood.

Even though the human brain is like a giant nonlinear LLM, somehow we have a kind of thinking that operates “actively,” more like a traditional algorithm. The classic way to distinguish it is by asking questions: e.g., someone asks me, what do you think about MBTI? This article is my answer — that’s my active thinking, a process with a traceable path that can be output as human language. And there’s another kind of thinking, which we can think of as more like how an LLM operates. In fact, most subsystems of the brain run this way. For example: after meeting someone once, we can recognize them from a photo. How? Because they had double eyelids? A big nose? Dark skin? No — none of that even crossed our mind, we just recognized them! You can think of it as a dedicated VLM that gets trained while we sleep and does inference during the day, handing the recognition result over to the brain’s “traditional algorithm” computation module for later analysis.

So from this angle, you can understand face blindness as: this “VLM” module probably hasn’t developed well, so they have to use traditional algorithms for face recognition, and you can imagine how well that works — humans still aren’t very good at it even today, and ultimately even the VLM model itself doesn’t quite know how it recognizes faces. From this same angle you can also understand (though not necessarily correctly): the ASD spectrum may correspond to the dedicated neural network (a more general name for the LLM-style modules) for analyzing and recognizing facial expressions and emotions not having developed well. So-called high-functioning Asperger’s IQ compensation is when the CPU and traditional algorithms are so strong that the recognition results rival a dedicated neural network running on GPU.

Back to S/N: N means our brain tends to choose, or is good at choosing, the information that has been abstracted by these dedicated neural networks from the raw inputs (visual, auditory, tactile) — for example, judgments about a person’s character. S may mean a tendency to focus on a person’s concrete, detailed features.

But is this a matter of being good at it, or just of preference? MBTI theory doesn’t really make this clear. But often, preference and skill feed back into each other.

An example will make it obvious: if you’ve met someone just once and have some impression of them, an S-leaning person tends to remember things like “they had double eyelids.” An N-leaning person might tend to remember their character — for example, they’re someone quiet, who seems thoughtful but not great at expressing it.

In other words: more S-leaning people are more likely to remember concrete features — looks, clothing, speaking habits; more N-leaning people are more likely to remember an overall impression — the vibe a person gives off, their thinking style, their expressive style.

The Second Difference

N and S represent abstract vs. concrete, and the “concrete” can be either people or things, just as the abstract can be people or things.
T and F represent thinking vs. feeling. Both are abstract; it’s just that feeling is the people-related kind of abstraction, while thinking can be understood as people-unrelated abstraction (abstraction that doesn’t treat people as people).

Which means the example in the previous section wasn’t quite accurate. More precisely: holding N constant, NF types may more easily notice things like a person’s character, sense of relationship, emotional state — abstract content that’s people-related; NT types may more easily notice things like a person’s reasoning, structure of opinions, mode of analysis — abstract content that’s more about how things work. Similarly, holding S constant, SF types tend to more easily notice concrete, people-related content, e.g. someone’s tone of voice, politeness, expression, whether they make others feel comfortable; ST types more easily notice concrete, thing-related content, e.g. whether someone is dressed appropriately, whether they’re expressing themselves clearly, whether they’re efficient at getting things done, whether the details are reliable.

Combining the two big differences above, you can see that N/S and T/F are fundamentally quite different from each other, right? So the various combinations of these are completely different.

P/J

P/J is also easy to misread as “self-disciplined vs. procrastinator” or “good at planning vs. bad at planning,” but those aren’t accurate either.

A more appropriate reading is: when facing incomplete information and unsettled situations (externally), is a person more inclined to converge and lock things down, or more inclined to keep things open and look at more possibilities.

J types tend to form judgments quickly, set standards, and push things toward closure. It’s not that their execution is necessarily strong — it’s that psychologically they prefer “let’s nail it down.” P types tend to keep observing, preserve flexibility, and defer conclusions. It’s not that they’re necessarily disorganized — it’s that psychologically they prefer “let’s wait and see.”

So J/P isn’t about ability level; it’s about preference for “certainty” vs. “openness.” J types can procrastinate too, and P types can be highly efficient; it’s just that the former usually find it more uncomfortable to leave things hanging, while the latter usually find it more uncomfortable to be locked in too early.

Why Does I/E Have to Pair With the Others?

If S/N is more about “what you tend to pay attention to,” and T/F is more about “what standard/preference you tend to judge by,” then I/E is about: where these mental activities are mainly directed.

Because even among N types, some are more about constantly seeing new possibilities in external information, while others are more about constantly generating associations and overall pictures internally. Same with F: some care more about external relationships and the vibe of interactions; others care more about their own inner sense of values.

In other words, S/N and T/F only answer “which mode of processing do you lean toward,” and I/E adds in “is this processing mainly directed inward or outward.” So the eight basic “modes of brain activity” — Cognitive Functions — are Ni, Ne, Si, Se, Ti, Te, Fi, Fe, and then P/J plus the balance principle below determine each person’s priority ordering of these eight.

Why Balance?

The auxiliary-function balance principle in type dynamics has two balances: one between I/E, and one between S/N and T/F. Of course, this “balance principle” is more of a theoretical setup; its practical foundation isn’t very strong.

The I/E Balance

If you’re an I, your dominant is i and your auxiliary is e. Why aren’t both dominant and auxiliary i? Because then the person would probably become depressed, ha.

A person needs both inward reflection and outward action, so the dominant and auxiliary have to be one inward and one outward.
If you’re an E, this “outward-facing function” is the dominant function.
If you’re an I, this “outward-facing function” is the auxiliary function, and the dominant is the inward-leaning version of the other middle letter.

The S/N and T/F Balance

If the dominant is always S/N, the person keeps taking in information and has trouble converging;
If the dominant is always T/F, the person keeps making judgments and has trouble updating their input. So the auxiliary has to fill in the other category.

Test Results Are Unstable?

A lot of people reject MBTI on the grounds that it’s unstable — test results change. This criticism can’t be dismissed with a simple analogy; you have to break it down.

Are the test items and scoring methods reliable?
Does the test-taker understand the questions clearly?
Is the test-taker answering as honestly as possible?
More crucially: even when the person’s state hasn’t changed much, is the same person likely to flip across the binary boundary when retaking the test in the short term?

The first three points show up in many psychological measurements. Just because a lot of people use it, and therefore a lot of people misunderstand it, and therefore it gets called garbage — that’s not a good reason. Just like the iPhone has billions of units; tens of thousands break, but nobody concludes from that that the quality is bad. Many things get artificial, fabricated controversy precisely because so many people use them.

But the fourth point happens to be one of MBTI’s core controversies. Because it’s not a small fluctuation in a continuous score — it can flip you to an entirely different type.

Even at a top-tier hospital, the misdiagnosis rate for many illnesses is fairly high. Does that mean modern medicine is pseudoscience? No. On the other hand, people change too — for example, my resting heart rate this year might be 42, next year 48. Does that mean heart-rate measurement isn’t science? So “things change” by itself isn’t enough to invalidate all psychological measurement; but “type often changes upon retesting” does indeed weaken MBTI’s credibility as a typology tool. We have to honestly acknowledge this.

That said, when it comes to psychology, there aren’t that many theories that hold up to scrutiny. The Big Five? Honestly, I think that’s pretty bogus too.

Does the Theory Hold Up Under Scrutiny?

Psychology being more complex than physics doesn’t mean any psychological classification tool can be considered a reliable theory just because “it sounds plausible.” MBTI, at most, counts as a coarse-grained, discussable framework with limited evidential strength. But you also can’t blindly endorse it or dismiss it just because some individual found it accurate or inaccurate — TCM is the classic case. Unlike other scientific fields, though, a large portion of psychological theory is exactly like this (built on classification and induction, without reasonably reliable hypotheses or quantitative predictive mathematical models, so quantitative testing is even further out of reach).

Whether something holds up under scrutiny also has to be looked at dialectically. If every theory that fails to hold up to practical testing got expelled from science, there wouldn’t be many theories left. In reality, strictly speaking, outdated theories really should be “expelled” — Newtonian mechanics, for example. And yet we still teach it and use it heavily. Why?

Because in practice, we have to make trade-offs between a theory’s computational cost, ease of comprehension, and predictive accuracy. So we generally don’t use relativity or quantum mechanics to compute a simple pendulum or a vibrating spring; but when computing satellite GPS, we use special relativity rather than just Newtonian mechanics, because the error has already grown beyond what we can tolerate. When we compute voltages and currents, we generally don’t reach for quantum mechanics or Maxwell’s equations, even though those are obviously the more correct theories of how electrons and electromagnetic fields behave. So nobody calls us out and says: “you’re using a ‘wrong’ theory to compute voltage and current — that’s pseudoscience! Practice has long shown your theory is wrong, garbage!”

Of course, this kind of phenomenon isn’t very common, purely because the bar for physics and math is high enough that it filters out a lot of binary thinkers from picking sides and dunking. The bar for computing is lower, so in choices of programming languages you see a lot of obvious side-picking and dunking. In fact, the use and selection of programming languages is structurally similar to the trade-offs in physical-theory selection above: if you want top-tier performance and ultra-low latency, use C or even assembly; if you want fast development speed and don’t care so much about performance, use TypeScript or Python; if you want industrial software that balances stability and performance, use Qt; if you want a backend service with decent performance and good stability, use Java or Go. People who really understand computers don’t go around dunking on languages — they learn and use whichever language is most appropriate for their project. If you’ve only half-learned one language and, carrying that shame into territory you’re not strong in, you may end up turning it into side-picking and dunking, just to prop up some fragile self-esteem and avoid the shame.

MBTI’s rigor really is a lot lower than physics’s or computing’s, but measuring psychology by the rigor of physics and computing is itself a stupid-or-malicious dunking move. Because human research into the workings of the human brain is so insufficient, there’s a huge theoretical gap between the phenomena of the brain and its physical structure. But how can we tell different phenomena of the brain apart? At the moment, probably the only way is to summarize them through classification and induction. In any case, I can’t think of a better method.

Does the Theory Make Sense?

The section “Understanding MBTI’s Classification Scheme” already showed that it’s actually fairly reasonable — it does capture some common differences in attention, information preference, and judgment style.

But “seems to make sense” and “has been rigorously verified” are two different things. Compared to physics, math, and computing — fields with a much higher level of formalization — MBTI’s foundations, rigor of reasoning, and testability are all much weaker. The more appropriate stance is neither to deify it nor to lump it in with astrology, but to acknowledge that it can be illuminating, while its limits are also obvious.

What Use Is MBTI for Yourself?

I think MBTI does have some guiding value for the individual, but only on the condition that you treat it as a heuristic tool, not a diagnostic conclusion. First, you have to understand roughly what those dimensions are saying.

If you do, you can treat it as a low-cost entry point for self-observation, helping you summarize your preferences, work style, and communication style, and reduce some internal friction. Could you do this without MBTI? Of course — if you understand your own details, preferences, and habits well enough across the board. MBTI just lowers the complexity of organizing these questions to some degree, and necessarily comes with underfitting — i.e., distortion.

What Use Is MBTI for Socializing?

The biggest use for socializing is helping you realize: how others pay attention to information, how they express themselves, what they weigh when making decisions — these may be different from how you do it, and these differences usually come with strengths and weaknesses both.

It’s normal for people to have very different modes of thinking.
Strengths and weaknesses come as trade-offs. You can’t have it all yourself, so don’t demand it of others either.
You can use this to quickly gauge whether someone’s mode of thinking can mesh with yours, but you can’t apply it as a rigid verdict. Compared to judging by appearance, this is at least a bit more reliable.

But once you have someone’s type and then turn around and use it to explain everything they do, even pinning them down with it, distortion sets in. It’s there to remind yourself “don’t assume others are like me,” not to be used as a label-maker for other people.

In the same way that knowing about ADHD, ASD, NPD, BPD, OCPD, and other psychological issues helps me better understand myself, others, and the human brain — that complex and interesting thing — MBTI, for me, has a similar significance.

At bottom, understanding your own preferences, weak spots, and modes of mental activity helps you reduce pointless internal friction and play to your strengths while avoiding your weaknesses. And understanding what others pay attention to, how they express themselves, and how they tend to behave helps us reduce misunderstandings and conflicts and get along more peacefully. If MBTI has enough accuracy to offer a bit of insight in this sense, then it has already justified its value.