The privilege of doing little

I do not hate conceptual art because “it is not art”. That is the museum-cafe critique: convenient, loud, usually pronounced in front of something that really does look like a fire extinguisher forgotten there. The point is almost the opposite: conceptual art interests me precisely because it takes seriously a radical possibility, namely that a work can be a gesture, a device, a question, a torsion of context more than a display of technique.

The part that irritates me is more political than aesthetic. When visible technique withdraws, the work has to be read as meaning. But not everyone has the same right to be read that way. A chair moved by an already consecrated artist can become an intervention on the status of the object. The same chair moved by an unknown person risks remaining a moved chair, perhaps even in the wrong place, while someone from security arrives very slowly.

The precise target is not all modern art, which historically contains enormous and very different worlds. The target is the conceptual and contemporary zone in which the value of the work depends very strongly on the gesture, the frame, the genealogy, the name and the institution that makes it legible. Here the problem begins: if the work is above all embodied thought, who decides that this thought deserves to be heard?

The answer does not remain inside the museum. It starts from a chair, from a minimal gesture, from the privilege of being able to seem profound while doing very little; then it reaches resumes, degrees, PhDs, institutional brands and all those cases in which a society must judge something it cannot read well. The path is deliberately broad: first the aesthetic intuition, then the Bayesian model, then epistemic distance, the phase diagram and finally the possible exits from credentialism. The idea to keep fixed is simple: asking for more meritocracy is not enough; we have to build instruments capable of seeing merit.

Technique as collateral

A technically complex painting carries part of its own defense. Not all of it, because market, taste, social class and luck never go on vacation. But something is visible: composition, color, control of the medium, difficulty, rhythm, tension. A young unknown can still force attention because the surface of the work shows, at least in part, the quality of the gesture.

In conceptual art, by contrast, technique can be deliberately denied. The work does not say “look what I can do”; it says “look what this act means”. It is a wonderful liberation from virtuosity, but it has a cost: meaning no longer sits only in the object. It sits in the object plus the gesture, plus the author, plus the place, plus the history of art to which it responds, plus the critic who decides not to treat it as a maintenance accident.

Simplicity then becomes a luxury. First you have to prove that you can do a lot; then you are allowed to do little. First you pay the collateral of technique, curriculum, recognition; then you are granted the privilege of essentiality. This does not make conceptual art false. It makes it socially delicate, because the right to subtraction is not distributed evenly.

This resembles a feudalism of expression. Not in the sense that someone owns the wheat fields, but in the sense that someone owns legibility. With the same material gesture, the weight changes depending on who performs it. One person performs an operation on the language of art. The other has put something on the floor and now has to explain himself.

From the museum to the resume

The same mechanism leaves the museum and goes to a job interview in a restrained blazer. In work, research, university, startup selection and even public conversation, we often say we want to evaluate competence. In practice we evaluate compressed signals: degree, university, PhD, previous company, advisor, accelerator, journal, prize, network, follower count, the general air of someone already verified by someone else.

This does not happen only because the world is snobbish, although the world often offers excellent evidence for that hypothesis. It happens because many observers do not have enough epistemic resolution to read the result directly. A non-technical recruiter may not know how to evaluate a repository. A generalist investor may not understand whether a deep-tech demo is a breakthrough or a performance with a glossy dashboard. A non-specialist reader may not distinguish a real idea from a well-combed sentence.

When the result becomes opaque, the title becomes convenient. Not because it is always false: a degree, a PhD or selective experience does contain information. The trouble is that it also contains access, family, money, time, stability, cultural capital, knowledge of the rules, adults who already knew how the game worked when you still thought “orientation” had something to do with compasses.

Here we should be precise: the word is credentialism. It should name the thing clearly: a society that mistakes credentials for the dominant measure of value.

Latent quality, result and credential

Build the model from the bottom. A person $i$ has a latent competence $q_i$. It is not human value, not total intelligence, not a verdict on the soul. It is the quality relevant to a certain task $T$: technical ability, mathematical rigor, reliability, conceptual depth, artistic sensitivity, design ability and so on.

This person produces a result:

$$x_i$$

The result may be an article, software, a proof, a diagnosis, a composition, a startup, a portfolio, a work of art. The person also has a credential:

$$c_i$$

The credential may be a degree, a prestigious university, a master’s degree, a PhD, a previous company, an exhibition, a gallery, a prize, an institutional brand.

An observer $O$ wants to estimate $q_i$, but does not see it directly. The observer sees the result through their own cognitive apparatus. What they receive is:

$$y_{i,O} = q_i + \varepsilon_{i,O}$$

with

$$\varepsilon_{i,O} \sim N(0,\sigma_O^2)$$

The crucial point is that $\sigma_O^2$ does not belong only to the result. It depends on the observer. An expert looks at a repository and sees architecture, tests, technical debt, abstraction choices, fragile points. A non-expert sees that “the demo starts”, which is a useful criterion for buying a toaster, less so for evaluating a complex system.

In the case of conceptual art, the ambiguity sits partly in the work. In the case of social credentialism, the noise sits above all in the coupling between result and observer. The same performance can be very clear to someone with the instruments and almost mute to someone without them.

Bayesian estimation and the weight of the title

Put a general prior on quality:

$$q_i \sim N(\mu_0,\tau^2)$$

The credential is a noisy signal of quality:

$$c_i = q_i + \zeta_i$$

with

$$\zeta_i \sim N(0,\sigma_C^2)$$

Define precisions, the inverses of variances:

$$p_0 = \frac{1}{\tau^2}, \qquad p_X(O) = \frac{1}{\sigma_O^2}, \qquad p_C = \frac{1}{\sigma_C^2}$$

Precision is a technical word for how reliable a signal is. If variance is high, precision is low. If variance is low, precision is high. The world would be more orderly if meetings also worked this way, but let us not ask too much.

With independent Gaussians, the posterior mean is a precision-weighted average:

The weight of the direct result is:

$$w_X(O) = \frac{p_X(O)}{p_0 + p_X(O) + p_C}$$

The weight of the credential is:

$$w_C(O) = \frac{p_C}{p_0 + p_X(O) + p_C}$$

This is the formula that supports the whole piece. If the observer understands the result well, then $\sigma_O^2$ is small and therefore $p_X(O)$ is large. In that case:

$$w_X(O) \approx 1, \qquad w_C(O) \approx 0$$

The result dominates. If, instead, the observer does not understand the result, then $\sigma_O^2$ is large and $p_X(O)$ is small. In that case $w_C(O)$ grows.

The blunt sentence is this:

Not because the credential is magically true, but because it is legible.

With the same perceived result

Take two people, $A$ and $B$. In the frame of observer $O$, their results are indistinguishable:

$$y_{A,O} = y_{B,O} = y$$

But $A$ has a stronger credential:

$$c_A > c_B$$

The difference between the observer’s estimates is:

$$\Delta \hat q_O = \hat q_O(A) - \hat q_O(B)$$

Write the two estimates:

$$\hat q_O(A) = \frac{p_0\mu_0 + p_Xy + p_Cc_A} {p_0 + p_X + p_C}$$ $$\hat q_O(B) = \frac{p_0\mu_0 + p_Xy + p_Cc_B} {p_0 + p_X + p_C}$$

Subtracting:

$$\Delta \hat q_O = \frac{p_C(c_A-c_B)} {p_0 + p_X + p_C}$$

Therefore:

With the same perceived result, the social difference is produced by the credential. But the important part is not just “the degree counts”. The important part is:

The title counts to the extent that the result is illegible to whoever is judging.

This distinction moves the problem from the evaluated person to the social apparatus of measurement. We are not only saying that some people have a title and others do not. We are saying that a system poor in instruments uses the title because it cannot read the rest well enough.

Epistemic distance

Every observer has an epistemic frame:

$$e_O$$

Every evaluated subject has their own frame:

$$e_S$$

These vectors can contain education, technical language, experience, inferential habits, conceptual tools, available time, familiarity with the domain. The task $T$ decides which dimensions really matter. For this reason we introduce a metric:

$$G_T$$

The relevant epistemic distance is:

There is no absolute epistemic distance. Two people may be close when judging a novel and very far apart when judging mathematical physics; close in marketing and far in programming; close in theory and far in debugging a real system at three in the morning, which is a literary genre of its own.

Let observational noise depend on this distance:

$$\sigma_O^2(T,S) = \sigma_{\min}^2 + \kappa d_T(O,S)^2$$

If $d_T(O,S)$ is small, the observer sees well. If it is large, the result becomes opaque. This is the Euclidean version of the model. It works, but it grows too politely. The more useful intuition is to use a structure inspired by relativity.

The epistemic cone

In special relativity not all separations are equal. Some events can communicate causally; others sit outside the light cone. Here we use that structure as a mathematical analogy, not as a physical identity. Nobody is saying a recruiter curves spacetime, although some job descriptions do raise the suspicion.

Define the observer’s bridge capacity:

$$B_O(T)$$

This quantity represents how much the observer can bridge epistemic distance in that domain: training, time, attention, instruments, abstraction capacity, patience, willingness to have things explained without interrupting after twenty-eight seconds.

Define:

$$\beta_E(O,S,T) = \frac{d_T(O,S)}{B_O(T)}$$

If $\beta_E < 1$, the subject is inside the observer’s epistemic cone. Understanding is possible. If $\beta_E = 1$, we are on the edge: understanding requires effort, translation and a certain mutual mercy. If $\beta_E > 1$, the subject is outside the observer’s epistemic cone: direct evaluation does not work in the current frame, so the system leans on proxies.

For $\beta_E < 1$, define the epistemic Lorentz factor:

$$\gamma_E = \frac{1}{\sqrt{1-\beta_E^2}}$$

and epistemic rapidity:

$$\eta_E = \operatorname{artanh}(\beta_E)$$

Then:

$$\beta_E = \tanh \eta_E, \qquad \gamma_E = \cosh \eta_E, \qquad \sqrt{\gamma_E^2 - 1} = \sinh \eta_E$$

Now let observational noise grow with rapidity:

$$\sigma_O^2(\eta_E) = \sigma_{\min}^2 + \sigma_L^2\sinh^2\eta_E$$

For small epistemic distances:

$$\sinh \eta_E \approx \eta_E$$

so:

$$\sigma_O^2 \approx \sigma_{\min}^2 + \sigma_L^2\eta_E^2$$

Noise grows gently. But for large distances:

$$\sinh \eta_E \sim \frac{e^{\eta_E}}{2}$$

and therefore noise grows almost exponentially. The social translation is simple: beyond a certain distance, people do not understand each other “a little less”. Often they are not even measuring the same object.

The credential’s weight grows with distance

Substitute Lorentzian noise into the precision of the result:

The credential’s weight becomes:

Show that it grows with $\eta_E$. Set:

$$A = \sigma_{\min}^2, \qquad B = \sigma_L^2$$

Then:

$$p_X(\eta) = \frac{1}{A + B\sinh^2\eta}$$

Differentiate:

$$\frac{dp_X}{d\eta} = - \frac{ 2B\sinh\eta\cosh\eta }{ (A + B\sinh^2\eta)^2 }$$

Since:

$$2\sinh\eta\cosh\eta = \sinh(2\eta)$$

we have:

$$\frac{dp_X}{d\eta} = - \frac{ B\sinh(2\eta) }{ (A + B\sinh^2\eta)^2 }$$

For $\eta > 0$, $\sinh(2\eta) > 0$, so:

$$\frac{dp_X}{d\eta} < 0$$

The precision of the result decreases as epistemic distance grows. Now:

$$w_C(\eta) = \frac{p_C}{p_0+p_C+p_X(\eta)}$$

Differentiating:

$$\frac{dw_C}{d\eta} = - \frac{ p_Cp_X'(\eta) }{ (p_0+p_C+p_X(\eta))^2 }$$

Since $p_X'(\eta) < 0$, we obtain:

$$\frac{dw_C}{d\eta} > 0$$

Result:

This is the mathematical version of a very common scene: the farther the decision-maker is from the real work, the more they love the resume.

Grafico interattivo

Quando il titolo prende peso

regime ibrido

Rapidità epistemica ${\eta }_E$: 1.20 Misura diretta ${\Pi }_A$: 1.00 Accesso incorporato $\delta$: 0.60 Selettività $\beta$: 2.00

Peso credenziale 0%

Peso misura diretta 0%

Bias da accesso 0.00

Rapporto opportunità 1.00x

Regime: regime ibrido. ${\Pi }_D$ = 0.00.

The chart places three pieces of the model in the same view: the curve of $w_C$, the point chosen by epistemic rapidity and the dynamic amplification due to $\delta$ and $\beta$. If you increase $\Pi_A$, the quality of direct measurement apparatuses, the curve drops. If you increase $\eta_E$, the point moves right: the result becomes less legible and the credential absorbs more weight.

The credentialist horizon

We can define a threshold beyond which the credential weighs more than the direct result. The result dominates if:

$$w_X > w_C$$

Because the two weights have the same denominator, this condition is equivalent to:

$$p_X > p_C$$

The credential dominates when:

$$p_C > p_X$$

Using:

$$p_X(\eta) = \frac{1}{A + B\sinh^2\eta}$$

the condition becomes:

$$p_C > \frac{1}{A+B\sinh^2\eta}$$

Inverting:

$$A+B\sinh^2\eta > \frac{1}{p_C}$$

therefore:

$$\sinh^2\eta > \frac{1/p_C - A}{B}$$

If $1/p_C > A$, the threshold is:

For $\eta > \eta^\*$, the credential weighs more than the result. This threshold is the credentialist horizon: inside the horizon, the observer can still evaluate the result; outside the horizon, the result is too opaque and the system clings to the title like a handrail.

In words:

$\eta^\*$ measures how far a society can go before replacing competence with title.

The phase diagram

The model can be read as a phase diagram. Put epistemic rapidity $\eta_E$ on the horizontal axis: the farther right you go, the farther evaluator and evaluated are from each other. Put on the vertical axis:

$$\Pi_A = \frac{P_A}{p_C}$$

that is, the quality of direct measurement apparatuses relative to the precision attributed to the credential.

Define:

$$P_D = P_A+p_X(\eta_E)$$

where $P_D$ is total direct precision: external apparatuses plus spontaneous legibility of the result. The order parameter is:

The main transition occurs when:

Above that threshold, direct measurement is more precise than the credential. Below it, the title takes command.

The main regimes are:

$$\Pi_D > 3$$

Direct-meritocratic regime: measurement of the result clearly dominates. The title may exist, but it does not decide.

$$1 < \Pi_D < 3$$

Hybrid regime: result and credential compete. The system may be relatively healthy, but it depends greatly on who is evaluating and how much time they have.

$$\frac{1}{3} < \Pi_D < 1$$

Credentialist regime: the credential weighs more than direct measurement. Here a mediocre but well-packaged person can look more reliable than a competent but less legible person.

$$\Pi_D < \frac{1}{3}$$

Social-feudal regime: direct measurement is so weak that the system lives on proxies. Degree, institution, family, network, status and reputation become almost ontological coordinates of social value. If $\beta$ is also high, meaning opportunities are winner-takes-all, the system brutally amplifies small initial differences.

The worst movement is:

The more epistemic distance grows and measurement apparatuses worsen, the more the title replaces the result.

The credential also contains access

So far we have been generous: we treated the credential as a noisy but neutral signal:

$$c_i = q_i + \zeta_i$$

Socially, this is too kind. A credential also measures access: family, income, time, cultural capital, stability, psychological safety, schools, language, a network of informed adults, the ability to afford years of training without having to monetize every afternoon.

Introduce a variable $a_i$, access advantage:

$$c_i = q_i + \delta a_i + \zeta_i$$

The parameter $\delta$ measures how much the credential is contaminated by access. If $\delta = 0$, the title measures only competence plus noise. If $\delta > 0$, it also measures privilege.

The observer, however, often uses $c_i$ as if it were a clean signal of $q_i$:

$$\hat q_O = \frac{ p_0\mu_0 + p_Xy + p_Cc }{ p_0+p_X+p_C }$$

Substitute:

$$c = q + \delta a + \zeta$$

We get:

$$\hat q_O = \frac{ p_0\mu_0 + p_Xy + p_C(q+\delta a+\zeta) }{ p_0+p_X+p_C }$$

Take the expected value conditional on $q$ and $a$. Since:

$$E[y \mid q] = q, \qquad E[\zeta] = 0$$

it follows:

$$E[\hat q_O \mid q,a] = \frac{ p_0\mu_0 + p_Xq + p_Cq + p_C\delta a }{ p_0+p_X+p_C }$$

Collecting:

$$E[\hat q_O \mid q,a] = \frac{ p_0\mu_0 + (p_X+p_C)q + p_C\delta a }{ p_0+p_X+p_C }$$

The part due to access is:

Two people with the same competence and the same perceived result, but different access advantage, receive different evaluations:

$$\Delta \hat q_O = w_C(O)\delta \Delta a$$

Since $w_C(O)$ grows with epistemic distance:

This is the hardest formula in the model. It does not merely say “the system is unfair”. It says how it produces unfairness:

A society can become classist without every single decision-maker waking up in the morning with classist intentions. It is enough for distant observers to use proxies contaminated by access.

Society as a set of observers

A society is not one observer. It is a distribution of observers with different weights. A recruiter, an investor, a professor, an executive, a journalist, an official or a manager does not count as much as a random person in a thread at two in the morning, however important the thread may feel.

Let $\rho(O)$ be the social weight of observer $O$. The social value assigned to person $i$ is:

$$V_i = \int \rho(O)\hat q_O(i)\,dO$$

The average credentialist component is:

This is the credentialism index of a society or institution. If $C \approx 0$, direct results are evaluated above all. If $C \approx 1$, credentials are evaluated above all.

If the credential contains access:

$$c = q + \delta a + \zeta$$

then the average social advantage due to access is:

This formula is unpleasant but useful:

$$\text{credential injustice} \propto \text{access contamination} \times \text{institutional credentialism}$$

The problem is not only that some people have more access to the degree. The problem is that an epistemically poor society multiplies that advantage because it cannot measure alternatives well.

Direct measurement apparatuses

Saying “let us stop looking at degrees” is satisfying for about eleven seconds. Then the problem remains: if you remove a proxy without building a better measurement, you have not obtained a more just society. You have obtained a more confused one, which is different and often more quarrelsome.

We need to increase the precision of direct measurement.

Introduce an evaluation apparatus $A$: practical tests, portfolios, peer review, work samples, blind auditions, benchmarks, trial periods, rubrics, technical committees, expert translators.

Suppose the apparatus produces $K$ signals:

$$m_1,m_2,\ldots,m_K$$

Each signal is:

$$m_k = q + \epsilon_k$$

with:

$$\epsilon_k \sim N(0,\sigma_k^2)$$

The precision of signal $k$ is:

$$p_k = \frac{1}{\sigma_k^2}$$

If the signals are independent, the total precision of the apparatus is:

$$P_A = \sum_{k=1}^{K}p_k$$

The estimate becomes:

$$\hat q_A = \frac{ p_0\mu_0 + p_Cc + \sum_{k=1}^{K}p_km_k }{ p_0+p_C+P_A }$$

The weight of the credential is:

$$w_C(A) = \frac{p_C}{p_0+p_C+P_A}$$

This is the formula of institutional design:

If you want the credential to weigh at most a fraction $\varepsilon$, impose:

$$w_C(A) \leq \varepsilon$$

Using:

$$w_C(A) = \frac{p_C}{p_0+p_C+P_A}$$

the condition is:

$$\frac{p_C}{p_0+p_C+P_A} \leq \varepsilon$$

Multiplying:

$$p_C \leq \varepsilon(p_0+p_C+P_A)$$

Expanding:

$$p_C \leq \varepsilon p_0 + \varepsilon p_C + \varepsilon P_A$$

Moving $\varepsilon p_C$ to the left:

$$p_C(1-\varepsilon) \leq \varepsilon(p_0+P_A)$$

Dividing by $\varepsilon$:

$$\frac{p_C(1-\varepsilon)}{\varepsilon} \leq p_0+P_A$$

Therefore:

Example: if you want the title to weigh at most ten percent, meaning $\varepsilon=0.1$, you need:

If the credential is socially considered very precise, the alternative apparatus must be very good. This explains why so many institutions say “we look for talent”, then return to resumes as soon as the room fills with candidates.

Grafico

Quanto deve essere buona la misura diretta?

$\epsilon$ = 10%

Why anonymity alone is not enough

A blind evaluation temporarily removes the credential. Formally it sets:

$$p_C = 0$$

in the first phase. The blind estimate is:

$$\hat q_{\text{blind}} = \frac{ p_0\mu_0 + P_Am }{ p_0+P_A }$$

It can work very well if $P_A$ is large. But if $P_A$ is small, blind evaluation remains uncertain. The posterior variance without the credential is:

$$\operatorname{Var}(q \mid A) = \frac{1}{p_0+P_A}$$

With the credential it becomes:

$$\operatorname{Var}(q \mid A,c) = \frac{1}{p_0+P_A+p_C}$$

The informational value of the credential, in terms of variance reduction, is:

$$\Delta \operatorname{Var} = \frac{1}{p_0+P_A} - \frac{1}{p_0+P_A+p_C}$$

Compute it:

$$\Delta \operatorname{Var} = \frac{ p_0+P_A+p_C-(p_0+P_A) }{ (p_0+P_A)(p_0+P_A+p_C) }$$

therefore:

$$\Delta \operatorname{Var} = \frac{ p_C }{ (p_0+P_A)(p_0+P_A+p_C) }$$

This quantity decreases when $P_A$ grows. Translated:

The better the direct test is, the less you need to know the title.

The healthy rule is:

$$\text{result} \rightarrow \text{direct measurement} \rightarrow \text{possible credential}$$

The unhealthy rule is:

$$\text{credential} \rightarrow \text{filter} \rightarrow \text{maybe result}$$

In the first case the title helps when uncertainty remains. In the second it eliminates people before their real signal is measured.

Epistemic translators

The Lorentzian metaphor also helps explain the role of mediators. Suppose observer $O$ and subject $S$ are very far apart. The direct epistemic rapidity is:

$$\eta_{OS}$$

The direct noise is:

$$\sigma_{\text{direct}}^2 = A+B\sinh^2\eta_{OS}$$

Introduce a mediator $M$: someone who understands the subject enough and knows how to translate enough for the observer. It may be a technical person who explains a startup to an investor, a professor who evaluates an autodidact, a competent reviewer, a mentor, a serious popularizer, a community capable of bridging frames.

We have two rapidities:

$$\eta_{SM}, \qquad \eta_{MO}$$

In a Lorentzian structure, rapidities add:

$$\eta_{SO} \approx \eta_{SM}+\eta_{MO}$$

But noise grows as $\sinh^2\eta$, not linearly. One large rapidity is far more destructive than two small rapidities. If we split a distance $\eta$ into $n$ equal steps, each step has rapidity:

$$\frac{\eta}{n}$$

The total noise, assuming additive noises, is approximately:

$$\sigma_n^2 = nA+nB\sinh^2\left(\frac{\eta}{n}\right)$$

The direct noise is:

$$\sigma_1^2 = A+B\sinh^2\eta$$

For large $\eta$:

$$\sinh^2\eta \sim \frac{e^{2\eta}}{4}$$

while:

$$n\sinh^2\left(\frac{\eta}{n}\right) \sim n\frac{e^{2\eta/n}}{4}$$

For large $\eta$:

$$ne^{2\eta/n} \ll e^{2\eta}$$

so, if the fixed cost $nA$ is not too high:

Translation: a chain of competent intermediaries can make legible a signal that would be invisible in the direct passage. A good mentor is not a clientelist recommendation in clean clothes. It is a measurement apparatus that reduces effective epistemic distance.

Many people are not underestimated because they lack quality. They are underestimated because there is no bridge between their frame and the frame of whoever decides.

Why compressed signals win

A real result is high-dimensional:

$$x_i \in \mathbb{R}^n$$

A credential is almost scalar:

$$c_i \in \mathbb{R}$$

or even binary:

$$c_i \in \{0,1\}$$

This makes it poor, but portable. A degree occupies one line. A good portfolio requires time, tools, domain and attention. The world, unfortunately, has an administrative passion for things that fit on one line.

A distant observer cannot read $x_i$. They read a projection:

$$y_{i,O} = P_Ox_i+\varepsilon_O$$

If $P_O$ has low rank, many dimensions are destroyed. Two different results can become equal in $O$‘s frame:

$$x_A \neq x_B$$

but:

$$P_Ox_A \approx P_Ox_B$$

At that point the credential wins not because it is truer, but because it survives compression.

In information terms, the full result contains:

$$I(q;x)$$

But the observer accesses only:

$$I(q;P_Ox)$$

The credential contains:

$$I(q;c)$$

Credentialism begins when:

$$I(q;c) > I(q;P_Ox)$$

even if in absolute terms:

$$I(q;x) > I(q;c)$$

This is one of the key sentences:

A just society is not one that denies all value to credentials. It is one that increases the amount of real information accessible to observers.

Dynamics: reputation and opportunity

So far the model is static. But social life does not assign a score once and then go to sleep. Opportunities received today produce reputation tomorrow, and tomorrow’s reputation produces other opportunities the day after tomorrow.

Let $A_i(t)$ be the level of opportunities received by person $i$ at time $t$. Reputation evolves like this:

$$R_i(t+1) = (1-\lambda)R_i(t)+\gamma A_i(t)$$

where $\lambda$ is reputational forgetting and $\gamma$ measures how much opportunities increase reputation.

Opportunities depend on social value:

$$A_i(t) = \frac{ e^{\beta V_i(t)} }{ \sum_j e^{\beta V_j(t)} }$$

The parameter $\beta$ measures how winner-takes-all the system is. If $\beta$ is high, small differences in perceived value produce large differences in opportunity. It is the world in which coming second is often an elegant way not to exist.

Take two people with the same competence but different access:

$$q_A=q_B, \qquad a_A>a_B$$

The difference in social value due to the contaminated credential is:

$$\Delta V = \delta(a_A-a_B)C$$

The opportunity ratio is:

$$\frac{A_A}{A_B} = e^{\beta \Delta V}$$

therefore:

The initial advantage is not merely added. It is exponentiated by the market of opportunities. If $\beta$ is high, if $\delta$ is high and if $C$ is high, a difference in access produces an enormous difference in trajectory.

The chain is:

$$\text{access} \rightarrow \text{credential} \rightarrow \text{perceived value} \rightarrow \text{opportunity} \rightarrow \text{reputation} \rightarrow \text{more opportunity}$$

Credentialism is not only an error of evaluation. It is a dynamic accelerator of inequality.

Goodhart joins the party

If society assigns value according to:

$$V = w_Xx+w_Cc$$

rational individuals will invest in the channels with the highest return. Suppose a person can invest energy in real competence $q$ or in credential $c$:

$$q = q_0+\alpha_qe_q$$ $$c = c_0+\alpha_ce_c$$

with constraint:

$$e_q+e_c=E$$

The expected social value is:

$$V = w_X(q_0+\alpha_qe_q) + w_C(c_0+\alpha_ce_c)$$

The marginal return on investment in competence is:

$$\frac{\partial V}{\partial e_q} = w_X\alpha_q$$

The marginal return on investment in credential is:

$$\frac{\partial V}{\partial e_c} = w_C\alpha_c$$

The person will invest more in credential than in competence when:

$$w_C\alpha_c > w_X\alpha_q$$

This is a formalization of Goodhart: when a measure becomes a target, it ceases to be a good measure. If the title opens doors, people optimize for the title. If the brand weighs more than the work, they optimize for the brand. If the biography must contain certain words, those words will appear with the spontaneity of a fake plant in a waiting room.

This is where title inflation begins. When everyone has the degree, the master’s degree is needed. When many have the master’s, the PhD is needed. When many have the PhD, the famous university is needed. Then the publication, then the network, then the recommendation letter, then the right contact, then a very well-dressed kind of exhaustion.

Formally, society raises the threshold:

$$c_i > c^\*(t)$$

and $c^\*(t)$ grows as the average distribution of credentials grows. The proxy becomes more and more expensive and less and less informative.

The problem of distant decision-makers

Return to the index:

$$C = \int \rho(O)w_C(O)\,dO$$

A society may have many experts, but if decision-making power is not in their hands, credentialism remains high. The problem is not only the distribution of observers; it is the distribution of power among observers.

If $\rho(O)$ is concentrated on epistemically distant observers, then $C$ is high. If $\rho(O)$ shifts toward observers close to the domain, then $C$ falls.

A society becomes credentialist when selection power is held by observers far from the signals they must evaluate.

Examples:

• a non-technical manager selects engineers mostly by looking at resumes;
• a non-technical investor evaluates deep tech by looking at universities and advisors;
• a bureaucratic body selects research through poor numerical indicators;
• a generalist public evaluates a thinker by looking at title, TED talk and publishing brand;
• an artistic system recognizes the conceptual gesture especially when it already arrives validated by gallery, museum and genealogy.

In all these cases the problem is not the proxy alone. It is the distance between whoever decides and what is being decided.

How credentialism is reduced

The goal is not to abolish every credential. The goal is to reduce:

$$C = \int \rho(O)w_C(O)\,dO$$

We can act on four levers.

First: increase the precision of direct measurement.

$$P_A \uparrow \Rightarrow w_C(A)=\frac{p_C}{p_0+p_C+P_A}\downarrow$$

This means work samples, practical tests, portfolios actually evaluated, peer review, benchmarks, blind auditions, trial periods, good rubrics.

Second: reduce epistemic distance.

$$\eta_E \downarrow \Rightarrow \sigma_O^2 \downarrow \Rightarrow p_X \uparrow \Rightarrow w_C \downarrow$$

This means translation, training decision-makers, competent intermediaries, better explanations, common standards, evaluators who know how to read the signal.

Third: move evaluative power.

$$\rho(O_{\text{competent}})\uparrow$$

Having experts around is not enough if decisions are then made by someone looking at the signal from too far away.

Fourth: reduce the credential’s contamination by access.

$$\delta \downarrow$$

This means scholarships, broader access, paths less dependent on family, better schools, less opaque procedures, lower costs, better distributed information.

An anti-credentialist apparatus must therefore do at least one of these things:

Saying “more meritocracy” is not enough. We have to build instruments that see merit. Otherwise meritocracy remains a good word for conferences, and good words for conferences have already done enough damage.

The ideal social detector

We can formulate the problem as optimization. An institution chooses an apparatus $A$, which produces signals $M_A$. We want those signals to contain a lot of information about $q$, little information about $a$, cost little enough to be usable and be legible to observers.

One possible objective function is:

where:

$$I(q;M_A)$$

is information about competence;

$$I(a;M_A)$$

is information about access advantage;

$$\operatorname{Cost}(A)$$

is the cost of the apparatus;

$$\operatorname{Noise}_O(A)$$

is the average observer noise in reading it.

A good system maximizes information about competence and minimizes information about privilege. It does not erase all social signals. It builds signals with high competence content and low access content.

The degree is often ambiguous because it contains both:

$$c = q+\delta a+\zeta$$

A better apparatus must reduce $\delta$ and reduce $\zeta$.

The final form of the theory

The social estimate of an observer is:

$$\hat q_O = \frac{ p_0\mu_0 + p_X(O)y_O + p_Cc }{ p_0+p_X(O)+p_C }$$

The credential’s weight is:

$$w_C(O) = \frac{p_C}{p_0+p_X(O)+p_C}$$

The precision of the direct result is:

$$p_X(O) = \frac{1}{ \sigma_{\min}^2+ \sigma_L^2\sinh^2\eta_E(O,S,T) }$$

Therefore:

$$\eta_E\uparrow \Rightarrow p_X\downarrow \Rightarrow w_C\uparrow$$

If the credential contains privilege:

$$c=q+\delta a+\zeta$$

then:

$$\text{bias}=w_C\delta a$$

Aggregating over socially powerful observers:

$$C = \int \rho(O)w_C(O)\,dO$$

and therefore:

$$\Delta V_{\text{access}} = \delta\Delta a\,C$$

In a competitive market of opportunities:

$$\frac{A_A}{A_B} = e^{\beta\delta\Delta a\,C}$$

The initial privilege is amplified when:

$$\beta \text{ is high}, \qquad \delta \text{ is high}, \qquad C \text{ is high}$$

Translated:

Two mistakes to avoid

The first mistake is to say that a degree means nothing. That is not true. In general:

$$I(q;c)>0$$

The credential contains information.

The second mistake is to say that, since it contains information, it is therefore right to use it as the dominant filter. That is not true either, because:

$$c=q+\delta a+\zeta$$

The credential also contains noise and access. The correct position is:

The credential is an informative but socially contaminated signal.

A proxy is acceptable as support. It becomes unjust when it becomes destiny.

Returning to the gesture

Conceptual art makes the problem visible because it pushes it to the limit. If the work is almost all idea, who guarantees that the idea will be read as an idea? If the answer is too often “the name”, then the free gesture hides an entry tax.

The same structure repeats wherever a complex result passes through an observer with poor instruments. Software passes through the recruiter. Research passes through the bibliometric indicator. The startup passes through the pitch deck. The person passes through the resume. The work passes through the wall label.

The privilege of doing little is not the right to be simple. Simplicity can be an enormous achievement: three notes played by someone who knows everything they are keeping silent, a clear page after years of thought, an ordinary object placed in the exact point where it becomes a question. The privilege begins when only some people can be read that way, while everyone else is asked to display evidence, muscles, titles, receipts, blood and a small notarized declaration.

The minimal gesture should not be abolished. What must be democratized is the right to be taken seriously. And that is not obtained by proclaiming that everyone has talent; it is obtained by building instruments good enough to recognize talent even when it does not arrive already dressed as authority.