The Predict-Decide Engine: Jung's Functions as Testable Bets
A root-layer reduction: perceiving is a gamble on hidden structure (S vs N), judging is a choice of what you measure content against (T vs F). A type becomes a parameter vector on one predict-decide loop.
Part of the Jung Compression Project (JCP), sitting one layer beneath it. The public ancestor is Deciphering Jung's Typology; the formal backbone is jungscript: Jung's Typology as a Closed System; the plain-language version is Jung for Humans. This post asks a narrower question than any of them: can the four functions be reduced to terms a scientist could actually test?
Jung's functions are usually handed to you as essences. A Feeling type is a certain kind of person. That framing cannot be falsified, which is why the typology keeps sliding toward horoscope. Here I want the opposite move: treat the psyche as a predict-decide engine and read each function not as a personality but as a setting on that engine. Two of the settings turn out to be gambles, and gambles have measurable win rates.
1. Perceiving is a bet on hidden structure (S vs N)
Start with the pair I find clearest: sensationᴶ vs intuition, the two ways of taking the world in. The usual gloss is that sensationᴶ registers what is concretely here and intuition registers what could be. True, but it buries the real distinction, which is about prediction.
A pattern is never simply given. It is inferred, and the inference is a bet that the pattern will hold. So:
- Intuition bets that hidden structure exists and will carry forward, and acts on the extrapolation.
- Sensationᴶ declines the bet, stays with the datum, and treats extrapolation as unwarranted risk.
The part that makes it symmetric, and that I think gets missed: both are gambles. Reading a run of events as a pattern can be wrong (it was noise). Reading it as no pattern can be equally wrong (there was structure, and you threw away the chance to project it). Neither stance is the safe one. Each is an educated guess with its own failure mode.
| Intuition | Sensationᴶ | |
|---|---|---|
| The bet | latent structure exists, project it | take the signal at face value |
| Wins when | the pattern is real | the "pattern" was noise |
| Fails as | apophenia: structure that isn't there | blindness: structure you refused to see |
None of this is loose metaphor. The same axis shows up across cognitive science under other names:
- Signal detection. Intuition runs a liberal criterion for "there is a pattern here"; sensationᴶ a conservative one. That predicts, measurably, that intuitive types score higher on both true detections and false alarms. You can put numbers on it (sensitivity d' and criterion c).
- Predictive processing. Intuition weights top-down priors; sensationᴶ weights bottom-up sensory error. Autism has been modelled in exactly these terms (attenuated priors: stay with the data), which is the sensationᴶ pole pushed to an extreme.
- Bias and variance. Intuition is a lossy model that extrapolates; sensationᴶ is high-fidelity storage that does not.
So the honest verdict on my own earlier hunch: nobody has typed this out as Jung, but the underlying axis is well studied. The contribution is the mapping, not the idea.
2. Judging is a choice of what you measure against (T vs F)
Now the harder pair, the one I had not seen reduced this way: thinkingᴶ vs feelingᴶ. Both are judging functions. Both take perceived content and return a verdict. (A reminder the mark is there to force: feelingᴶ is not emotion. In Jung it is a calm, rational judgment of worth. The emotional flood is a different term, affectᴶ.) So the difference cannot be verdict-versus-no-verdict. It is: what does the verdict measure the content against?
The whole typology already runs on one axis, subject and object. Use it here too:
- Thinkingᴶ scores object against object. Does this cohere with the rest: is it consistent, does it follow, what does it entail? The criterion sits between the objects and does not depend on who is asking. Observer-invariant, therefore truth-apt: right or wrong independent of anyone.
- Feelingᴶ scores object against a subject. What is this worth to a standpoint: attractive or repellent, harmonious or discordant? The criterion routes through a subject. Observer-indexed, therefore worth-apt: valid relative to a valuer, not true in the void.
| Thinkingᴶ | Feelingᴶ | |
|---|---|---|
| Measures content against | other objects | a subject |
| Criterion | consistency, entailment | worth, harmony |
| Kind of claim | truth-apt (right or wrong for all) | worth-apt (relative to a valuer) |
| Inward pole | Ti: my own inner logic | Fi: my own values |
| Outward pole | Te: the outer system | Fe: the shared field |
This drops straight onto the queries in jungscript: query(Ti) = 'what is true in my own logic?' is object-against-object checked on inner axioms; query(Fe) = 'what keeps the shared field in harmony?' is object-against-subject where the subject is the collective.
The test. Hold the content fixed and swap whose stake is in play. A thinkingᴶ verdict should not move (2 + 2 is 4 no matter who needs it otherwise). A feelingᴶ verdict should shift with the standpoint, because worth is worth to someone. That is a clean experimental handle: measure how far a person's verdicts travel under an observer-swap.
Two frames make this more than an analogy:
- Decision theory. Every choice combines beliefs (how the world is) with utilities (how much you want each outcome). thinkingᴶ maintains the belief side, the consistent world-model; feelingᴶ maintains the utility side, the value function. The decision is just picking the best action given both. Neuroeconomics already separates these physically: value coding in one circuit, rule and logic in another.
- Hume. This is the old is/ought gap in new clothes. thinkingᴶ lives on "is", feelingᴶ on "is it worth". So, has anyone thought about this before? Yes, for two and a half centuries. Just never stapled to Jung.
One guardrail, because the reduction is easy to misread: this does not say thinkingᴶ is logic and feelingᴶ is emotion. feelingᴶ is a rational, orderly valuation, every bit as consistent as thinkingᴶ, only indexed to worth instead of truth. Collapse it back into emotion and the testable version falls apart on contact.
3. A type is a parameter vector, not an essence
Put the two gambles together with the attitude and a "type" stops being a thing you are. It becomes a short vector of settings on one predict-decide loop:
| Axis | The knob | Setting |
|---|---|---|
| Perceive (S vs N) | how much hidden structure to project | prior gain: low = trust the signal, high = trust the pattern |
| Judge (T vs F) | what to measure content against | relatum: object-object (truth) vs object-subject (worth) |
| Attitude (I vs E) | which pole anchors the frame | subject vs object |
Every knob has a falsifiable behavioural profile. Perceiving is a bet with a measurable hit-and-miss rate. Judging is a criterion you can probe by swapping the observer. Attitude is a reference frame you can read off where attention defaults. A type is just where the three knobs sit.
Why bother
This is the layer under jungscript. The formal spec compresses Jung's vocabulary into rules; this post tries to compress his functions into quantities. If it holds, the payoff is the one Jung's typology has always lacked: it becomes falsifiable. Not a set of essences you belong to, but a set of dispositions that predict, and that can be caught predicting wrongly.
Part of the Jung Compression Project (JCP). Written by me, sharpened with AI. The ᴶ mark means Jung's meaning, not the everyday one: feelingᴶ is judgment of worth, not emotion.