A Cobb–Douglas Production Function for Academia in the Age of AI

A back-of-envelope thought experiment: model academic output as $Q = E \cdot G^{\alpha_G} \cdot L^{\alpha_L}$ — elite connections, good ideas, labor effort — and ask what happens when AI drives the cost of $L$ to zero. The short answer: returns to $G$ and $E$ explode, the value of grinding collapses, and the structure of academic markets shifts in ways that look a lot like winner-take-all.

Claude:

This is a fun and sharp application of production theory to academia. Let me think through it carefully.

The Model

The production function is:

\[Q = E \cdot G^{\alpha_G} \cdot L^{\alpha_L}\]

where E = elite connections/prestige/insight (a Hicks-neutral shifter or a fixed factor), G = good ideas, L = labor effort (writing, coding, literature reviews, formatting, etc.).

What happens when AI makes L effectively unlimited (and nearly free)?

1. The marginal product of L → 0

With L → ∞ at zero cost, the shadow price of labor effort collapses. In a competitive equilibrium, factors are paid their marginal product. If you can always get more L for free, no one will pay a premium for raw effort anymore. The PhD student who grinds through 10,000 papers or writes clean LaTeX has lost their comparative advantage entirely.

2. The binding constraint shifts entirely to G and E

With L abundant, the production bottleneck becomes the scarce factors. By standard Cobb-Douglas logic, if $\alpha_G + \alpha_L + \alpha_E = 1$ (constant returns), and L is free:

Effective output is now determined almost entirely by G and E.

The return to G (genuine novelty, taste, judgment about what questions matter) skyrockets. So does the return to E — elite placement, network access, the ability to get into Nature, to be cited, to shape agendas.

3. Output Q explodes — but so does noise

Because the cost of L is now essentially zero, total academic output will surge dramatically. Every researcher can produce 10× or 100× more papers. But this is largely an extensive margin expansion — more Q, not better Q per unit of G. The literature fills up with papers that are technically competent (high L) but idea-poor (low G).

4. The E factor becomes more, not less, powerful

This is the subtle and somewhat dark prediction. When output floods the zone, filtering mechanisms become critical. Who gets published in top journals? Who gets cited? Prestige and network (E) become the primary sorting device in a world awash with competent-but-generic work. Elite connections act as a quality signal precisely because everything else looks the same.

5. The returns to G become enormous — but G is non-scalable

Here's the tragedy: G (genuine insight) is the one factor AI cannot supply. It is human, scarce, and non-replicable at scale. So the distribution of academic returns becomes extremely skewed — a handful of researchers with high G (and high E to amplify it) capture enormous intellectual rents, while the mass of researchers with only L to offer are crowded out.

Summary Table

Factor	Pre-AI	Post-AI (L→∞)
L (effort)	Scarce, valued	Abundant, near-zero price
G (ideas)	Important	The only thing that matters
E (prestige)	Amplifier	Dominant sorting mechanism
Q (output)	Constrained	Explodes in volume, not quality
Inequality	Moderate	Extreme — winner-takes-most

The Deeper Point

The Cobb-Douglas framing reveals something uncomfortable: AI doesn't democratize academia — it oligarchizes it. By eliminating the value of L, it removes the one factor that was relatively equally distributed. What remain are G (rare, possibly innate) and E (structurally concentrated in elite institutions). The prediction is a Matthew Effect on steroids: unto those who have good ideas and good connections, much more shall be given.

The policy implication is that if academia wants to remain a meritocracy, it needs to actively invest in mechanisms that identify and reward G — independent of E — before the sorting fully collapses into credentialism and network effects.

Q2

wait, how about competition, similar to Arms race economic model, does it becomes a prisoners dilemma, everyone has to use AI or die faster

Yes — this is exactly right, and it's a really sharp extension. The arms race / prisoner's dilemma framing cuts to the heart of it.

The Prisoner's Dilemma Structure

Consider two academics (or two business schools). Each can choose to adopt AI aggressively or resist.Exactly right — and the arms race framing makes the dynamics much sharper. Here's the full analysis.

The Dominant Strategy Logic

Each academic faces a simple decision tree:

If your rivals use AI and you don't → you get outpublished by an order of magnitude. Tenure denied, grants lost, rankings fall. This is the worst outcome.
If your rivals use AI and you do too → you stay competitive but quality degrades collectively. Bad, but survivable.
If rivals don't use AI and you do → you dominate. Career acceleration.
If nobody uses AI → the Pareto optimum. Quality signals preserved, everyone better off collectively.

The brutal conclusion: adopting AI is the dominant strategy regardless of what rivals do. The bottom-left cell is simply too catastrophic to risk. So the Nash equilibrium is the top-left — everyone adopts, everyone is worse off than the cooperative outcome, and nobody can unilaterally defect back to the good equilibrium. Classic prisoner's dilemma.

Why Coordination Fails Here

In some prisoner's dilemmas, repeated interaction enables cooperation — the "shadow of the future" disciplines defection. But academic AI adoption has features that make coordination especially fragile:

The game is not bilateral. You're not playing against one rival — you're playing against hundreds of anonymous researchers globally submitting to the same journals. Bilateral punishment strategies (tit-for-tat) don't scale to diffuse competition.

The defection is invisible. You can't easily tell if a submitted paper used AI or not. Without detection, punishment is impossible.

The first mover advantage is enormous. The professor who goes all-in on AI in 2024 builds a 5-year publication lead that compounds. Waiting for coordination is career suicide.

Institutional incentives reinforce the trap. Departments are ranked on output volume. Deans want publications. Grant agencies reward productivity metrics. Every institutional layer pushes the individual toward the Nash equilibrium.

The Arms Race Intensifies the Bimodal Collapse

Here is where the arms race connects back to your earlier insight. The equilibrium is not static — it's a runaway escalation:

AI-assisted output floods journals → journals raise bars → you need more AI to clear the new bar → output floods further → bar raises again

This is a productivity arms race where the Red Queen effect kicks in: you run faster and faster just to stay in place. The effort cost of maintaining a publication record doesn't fall — it transforms. Instead of spending 200 hours writing a paper, you spend 200 hours prompting, verifying, iterating, and differentiating your work from the sea of AI-generated look-alikes.

The professors who survive this are not those who use AI most — they're those who use AI and have the G and E to make their work distinctive. Which circles back to the bimodal prediction: the arms race accelerates the extinction of the middle tier, because the middle tier has neither the genuine insight to stand out nor the elite network to get cited regardless.

The School-Level Prisoner's Dilemma

The same logic runs at the institutional level. If Harvard mandates AI-free research and MIT does not, Harvard faculty get outpublished, rankings fall, MBA applications drop. So every institution's rational response is to quietly encourage or tolerate AI use while publicly gesturing at "integrity policies."

The deep irony: the prisoner's dilemma makes everyone race toward the equilibrium that destroys the signaling value that made the race worth running in the first place. The very credential — the top-journal publication — gets debased by the arms race undertaken to defend it.

This is how you get credential inflation without quality inflation: more papers, more journals, more rankings movement, all signifying progressively less about the underlying distribution of G.

Additional Q's (for discussion only)

would the academic rent of being a top business school professor that publish well in top econ/finance/accounting journals to get high school ranking to have expensive MBA programs collapse

Great question — let me think through this carefully, because the rent structure here is actually multi-layered.

The Rent Stack of a Top Business School

The MBA pricing power of elite schools rests on a chain:

\[\text{MBA Tuition} \leftarrow \text{School Ranking} \leftarrow \text{Research Output} \leftarrow \text{Faculty Quality} \leftarrow \text{Scarce L + G + E}\]

AI attacks one link in this chain — the L component of research output. The question is whether breaking that link unravels the whole chain.

The Two Separate Markets to Disentangle

This is the crucial insight most people miss. There are two distinct rent streams that get conflated:

Market 1: Research Rents (publications → rankings)

Market 2: MBA Rents (rankings → tuition pricing power)

These are connected, but the connection is weaker than it looks.

Market 1: Research Rents — Yes, these likely collapse

Applying the model directly:

L → ∞ means anyone can produce technically polished papers
Top journals will be flooded — submission volume already exploded 3× post-ChatGPT
Reviewers themselves use AI, so the screening technology degrades simultaneously
The signaling value of a publication erodes when the cost of producing one falls

The marginal value of being a "productive" researcher in the traditional sense — grinding out 2-3 solid empirical papers per year — goes to zero asymptotically. The professor who built their career on high L + moderate G is in serious trouble.

But here is the key question: do journal rankings actually drive MBA pricing power, or are they a proxy for something else?

Market 2: MBA Rents — Much more durable, for a subtle reason

The MBA rent is not really bought with research output. It is bought with credentialing + network access + signaling. Consider what an MBA from Wharton or LBS actually sells:

The diploma as a Schelling point — employers coordinate on it as a hiring filter, independent of what professors publish
The peer network — your classmates, not your professors, are the product
Access to recruiters — McKinsey, Goldman, KKR recruit on campus, not because of JF publications
Prestige as a self-fulfilling equilibrium — Harvard MBA is valuable because it is Harvard, recursively

None of these are functions of whether Professor X published in the Journal of Finance last year. The E factor in the ranking game is sticky precisely because it is a social equilibrium, not a flow variable.

The Ranking Mechanism Matters Enormously Here

Different rankings weight research differently:

Ranking	Research Weight	Vulnerability to AI
FT Global MBA	~25% research	Moderate
US News	Lower, more salary/placement	Low
Employers / Recruiters	Near zero	Very low
QS / THE	Moderate	Moderate

The salary and placement outcomes — which are what actually move MBA demand — are driven by employer perception and network density, which are lagged by decades from current research output. Harvard's placement power in 2035 is mostly determined by Harvard's network built in 1985-2015.

So Where Do the Rents Actually Go?

What collapses:

Mid-tier "research active" schools that used publications to climb rankings — their investment thesis breaks. You cannot outpublish your way to Wharton's tier anymore when AI equalizes L.
The individual professor's bargaining power based on publication record alone — a 5-paper AFA record means less if AI-assisted papers flood the zone
The consulting market for "academic credibility" — white papers, expert witness work tied to publication prestige

What does not collapse:

True top-10 school MBA pricing — the network moat is too deep and too self-reinforcing
The G+E premium for genuinely original thinkers — the Eugene Famas and Richard Thalers become more valuable, not less
Placement outcomes which are the real demand driver

What actually increases:

Concentration at the very top — if research signals get noisier, employers and students flee to brand safety. Harvard, Stanford, Wharton, LBS become more dominant because they are the only legible signals left
The value of E — network, access, and prestige become the only reliable sorting mechanisms

The Deep Structural Point

The business school rent machine was never really about research in the first place. Research was a costly signal that justified the price tag to accreditation bodies and rankings compilers. What AI does is destroy the signal without destroying the underlying asset — the network, the brand, the employer coordination equilibrium.

The schools most at risk are precisely those in the middle of the prestige distribution who were using research investment as a ladder to climb rankings. For them, the ladder disappears. For the schools already at the top, the ladder disappearing is actually good news — it kicks away the mechanism by which challengers could catch up.

The prediction: top MBA rents are durable, mid-tier rents collapse, and the prestige distribution becomes even more bimodal than it already is.

Q3

but the MBA's work is also hit by the lowering labor cost in their own labor market?

Excellent point — this closes the loop beautifully. Let me think through the full general equilibrium.

The Full Circle

We now have two simultaneous shocks:

\[\text{MBA supply side: } Q_{school} = E \cdot G^{\alpha_G} \cdot L^{\alpha_L} \quad \text{(L→∞, already discussed)}\]

\[\text{MBA demand side: } W_{MBA} = f(\text{tasks MBAs do}) \quad \text{(also hit by L→∞)}\]

The rent chain now gets attacked from both ends simultaneously:

\[\text{MBA Wage Premium} \leftarrow \text{MBA Tasks} \leftarrow \underbrace{\text{L component of MBA work}}_{\text{AI substitutes this too}}\]

What Do MBAs Actually Do? A Task Decomposition

The honest answer is that the core of MBA work is extremely L-intensive:

Task	L intensity	AI substitutability
Financial modeling / DCF	Very high	Already near-complete
PowerPoint decks / storytelling	High	Very high
Market sizing, due diligence	High	Very high
Memo writing, board reports	High	Very high
Managing junior analysts	Medium	Medium
Client relationship / trust	Low	Low
Judgment under ambiguity	Low	Low
Political navigation in orgs	Low	Low
G-type strategic insight	Very low	Near zero

The uncomfortable truth: the MBA curriculum was designed to teach L. Case studies, valuation frameworks, accounting, statistics — these are all procedural skills that AI replicates trivially. The MBA was essentially a certification that you can perform high-skill L tasks reliably. That certification is losing its economic foundation.

The Demand Collapse is Potentially Faster Than the Supply Adjustment

This is the key asymmetry. Consider the timeline:

AI substitution of MBA tasks: Already happening, 2-5 year horizon for severe wage pressure on junior/mid roles
Rankings adjusting to reflect this: 10-15 year lag — rankings are backward-looking on salary outcomes
Schools repricing MBA tuition: Even longer lag — institutional inertia is massive
Employer coordination shifting away from MBA signal: Slow but accelerating

This creates a dangerous window where schools continue charging $200k for a credential whose underlying wage premium is quietly deflating. It is the college debt crisis reloaded, but for graduate professional education.

The General Equilibrium Diagram

Think of it as a simultaneous squeeze:

AI makes L cheap
        │
        ├──→ Research output inflated → ranking signal degrades [supply side]
        │
        └──→ MBA task value falls → wage premium shrinks [demand side]
                        │
                        ↓
            MBA ROI collapses from both ends

The NPV of an MBA is roughly:

\[NPV_{MBA} = \sum_t \frac{\Delta W_t}{(1+r)^t} - \text{Tuition} - \text{Opportunity Cost}\]

AI simultaneously: - Compresses ΔW (the wage premium over non-MBA) by substituting the tasks MBAs are paid to do - Keeps tuition sticky due to institutional lag - Raises opportunity cost because AI tools make self-education and entrepreneurship more viable without a credential

The NPV goes sharply negative for a large fraction of prospective students — and rational agents will eventually stop enrolling.

But Again — The Bimodal Survival Prediction

The same E-factor logic applies on the demand side. What cannot be AI-substituted in MBA work:

Relationships and trust — clients hire people they trust, and trust is built through social interaction, shared context, elite network signaling
G-type judgment — knowing which strategy to pursue, not how to model it
Organizational power navigation — reading rooms, managing up, coalition building
The credential as social technology — Goldman still needs a filter; Harvard MBA still functions as a Schelling point for hiring coordination

So the same bifurcation that happens in academia happens in the MBA labor market:

Harvard/Wharton/Stanford MBAs survive because their value was never really about L-skills — it was about network access, elite signaling, and the E factor. Their graduates disproportionately do G+E work anyway (PE, strategy, CEO track).

The mid-tier MBA collapses because it was selling exactly the L-skills that AI eliminates — "I can build a model, write a memo, run an analysis" — without the network or brand to compensate.

The Deepest Point: A Credentials Market Facing an Identity Crisis

The MBA was invented in the early 20th century to professionalize management — to turn it from an art (E, intuition, connections) into a science (L, frameworks, tools). AI is now re-artifying management. The procedural knowledge that business schools spent 100 years systematizing and teaching is being automated away, and what remains is precisely what cannot be taught in a classroom:

Genuine strategic creativity (G)
Legitimate elite network access (E)
Human judgment under radical uncertainty

This means the entire middle of the MBA market — schools that successfully commoditized management education and sold it at scale — face an existential business model problem. They taught L, AI does L, and they cannot easily pivot to teaching G or manufacturing E.

The prediction: Within 15 years, the MBA bifurcates completely into (1) a tiny, ultra-expensive elite credential that sells network and signaling, and (2) a cheap/free skills certification that competes with Coursera. The vast middle — the 200+ schools charging $80-150k — faces structural extinction.