Introduction
The architecture of market allocation has undergone three structural transformations in modern economic history. First, the emergence of physical marketplaces centralized exchange and enabled price discovery through face-to-face bargaining. Second, the printing press and subsequent information technologies reduced search costs and enabled price comparison across geographically separated options. Third, digital platforms created algorithmic intermediation, coordinating millions of buyers and sellers through centralized matching systems.
We may be entering a fourth structural transformation: the emergence of AI systems as allocative interfaces. When machine learning models construct consideration sets—filtering, ranking, and presenting options before human decision-makers engage—the very architecture of competition may shift.
Foundational Question
How does price formation change when consideration becomes computational?
Traditional price theory assumes that competition occurs among all available options in a market. Buyers are assumed to have access (perhaps at some search cost) to the full set of available alternatives, and prices emerge from the interaction of supply and demand. But when AI systems construct consideration sets under bounded inference, competition may effectively occur only among options that are computationally admissible—that is, options that can be inferred, evaluated, and included by the allocating system.
If this structural transformation proceeds, price may no longer emerge solely from supply and demand. It may also emerge from allocative eligibility and inferential accessibility. This possibility motivates the theoretical framework we develop.
Caveats and Scope Limitations
Important Scope Clarifications
Before proceeding, we must establish explicit scope limitations:
- This is theoretical work. No empirical evidence is presented. All claims about market behavior are speculative and should be treated as hypotheses requiring validation.
- This is not about dynamic pricing. We do not discuss algorithmic pricing optimization, surge pricing, or real-time price adjustment based on demand signals. Our concern is with the structural conditions that make competition possible, not with how individual prices are set.
- This is not about platform economics. We do not discuss network effects, two-sided markets, or platform monetization strategies, except as background context.
- This is not predictive. We do not claim that AI-mediated markets will definitively exhibit the patterns described. We argue that these patterns are possible and warrant investigation.
- This is not normative. We do not argue that AI-mediated allocation is desirable or undesirable. We examine potential structural effects, not policy prescriptions.
Part I — Historical Pricing Models
Foundational assumptions of classical price theory
Classical Price Formation
The classical economic model of price formation rests on several foundational assumptions:
1. Value Precedes Competition
Buyers are assumed to have stable preferences and can assign utility values to available options. These values exist prior to and independently of the market interaction.
2. Competition Is Universal
All sellers in a market compete for all buyers. Frictions may exist, but structurally, competition is assumed to be n-to-n (all sellers to all buyers).
3. Price Equilibrates Markets
Prices adjust until supply equals demand. The price mechanism clears markets by reconciling buyers' willingness to pay with sellers' willingness to accept.
4. Information Flows Freely (Eventually)
While classical economics recognizes information asymmetry, the baseline model assumes that information about prices and quality is, given sufficient search, accessible to all market participants.
These assumptions served as the foundation for marginal utility theory, general equilibrium theory, and much of modern microeconomics. They remain valid under the conditions they were designed to describe: markets where buyers can access available options and competition operates across the full set of alternatives.
The Accessibility Assumption
All classical models share a common assumption: options that exist in a market are, in principle, accessible to buyers who search for them. Frictions exist—search costs, information asymmetry, transaction costs—but these are barriers to access, not barriers of access. Given sufficient resources, a buyer could find and consider all available options.
The AI-Mediated Allocation Challenge
The AI-mediated allocation challenge begins when this assumption fails.
Part II — Why AI-Mediated Markets May Differ
The consideration set problem and the K < n constraint
The Consideration Set Problem
Human decision-making operates through consideration sets—small subsets of available options that decision-makers actually evaluate before choosing. A homebuyer might visit five properties despite 500 being on the market. A traveler might compare three hotels despite 100 being available.
In traditional markets, consideration sets are buyer-determined. Each buyer, through their own search process, constructs a set of options to evaluate.
When AI Constructs Consideration Sets
AI-mediated allocation reverses this structure. Instead of buyers searching and constructing consideration sets, allocative systems construct consideration sets on behalf of buyers:
- A travel assistant presents three hotel options from thousands
- A property search system shows five homes from tens of thousands
- A product recommender suggests one item from millions
The critical shift is that consideration becomes computational. The allocative system, not the buyer, determines which options enter the consideration set.
The K < n Constraint
Formal Constraint Definition
Let n be the number of available options in a market.
Let K be the number of options presented to the buyer.
K < n
The system can only present a subset of available options. Inclusion becomes scarce.
In traditional markets, K is determined by the buyer's search process. The buyer could, in principle, expand K by searching more extensively (up to n).
In AI-mediated markets, K is determined by the allocative system's inference constraints:
- Compute budget: Time and resources available for search and evaluation
- Context window: Information the system can process in a single inference
- Model architecture: Limits on how many options can be compared in parallel
- Quality standards: Minimum confidence threshold for inclusion
Admissibility Premium
Definition: Admissibility Premium
The potential pricing advantage that computationally admissible options may enjoy over computationally inadmissible options, all else equal.
The admissibility premium emerges from a simple mechanism:
- Option A is computationally admissible (included in consideration sets)
- Option B is computationally inadmissible (excluded from consideration sets)
- Even if B offers better price or quality, it cannot compete because it is not considered
- A faces weaker effective competition and may sustain higher prices
From Value to Price: A New Chain
In traditional markets, the causal chain from value to price is:
In AI-mediated markets, the chain may become:
This longer chain introduces new structural positions where pricing power may accumulate.
Part III — Computational Price Formation
Representation friction, inferential cost, and allocative efficiency
Representation Friction
Definition
Representation Friction is the cost imposed on an option by poor, incomplete, or non-standard machine-readable representation.
Representation friction arises from:
- Data Sparsity: Insufficient structured data for evaluation
- Data Non-Standardization: Inconsistent formats requiring additional processing
- Representation Lag: Outdated or stale information
Inferential Cost
Definition
Inferential Cost is the computational resources required to evaluate an option for inclusion in a consideration set.
Inferential cost includes:
- Discovery Cost: Resources required to locate the option
- Evaluation Cost: Resources required to assess relevance and quality
- Comparison Cost: Resources required to compare against alternatives
Computational Liquidity
Definition
Computational Liquidity is the ease with which an asset enters machine-generated consideration sets across allocative systems.
Computational liquidity is the allocative analog of financial liquidity. It measures how easily an asset can be considered by allocative systems.
Part IV — New Pricing Primitives
Formal characterization of computational pricing mechanisms
Representation Friction
Rᵢ = φ(d(fᵢ))Higher representation friction correlates with lower prices, all else equal.
Hypothesis: ∂pᵢ/∂Rᵢ ≤ 0
Inferential Cost
Iᵢ = cᵢ / C(allocation)Higher inferential cost correlates with lower prices, all else equal.
Hypothesis: ∂pᵢ/∂Iᵢ ≤ 0
Allocative Efficiency
αᵢ = P(i ∈ S | i ∈ Ω)Higher allocative efficiency correlates with higher prices, all else equal.
Hypothesis: ∂pᵢ/∂αᵢ ≥ 0
Computational Liquidity
Lᵢ = aᵢ / |A|Higher computational liquidity correlates with higher prices, all else equal.
Hypothesis: ∂pᵢ/∂Lᵢ ≥ 0
Admissibility Premium
π = E[p | p ∈ S] - E[p | p ∈ T]Computationally admissible options may sustain higher prices, all else equal.
Hypothesis: π ≥ 0
Part V — Sector Examples
Speculative illustrations of theoretical concepts in practice. These examples are illustrative, not predictive.
Hospitality
- Hotels with structured, complete data preferentially included
- Admissibility premium for AI-friendly representation
- Computational liquidity across booking platforms
Real Estate
- Properties with rich specifications preferentially included
- Standardized representation protocols reduce friction
- Allocative efficiency affects time-to-sale
E-Commerce
- Products with standardized specifications preferentially included
- Brand recognition may bypass allocative systems
- Platform incentives shape system behavior
Services
- Providers with structured service descriptions preferentially included
- Regulatory requirements limit allocative discretion
- Professional networks persist alongside systems
Caveat
These sector examples are speculative and intended to illustrate theoretical concepts, not predict market outcomes.
Part VI — Implications for Infrastructure
If the theoretical mechanisms described above manifest in markets, allocative infrastructure may become economically significant.
Representation Infrastructure as Market Infrastructure
Representation infrastructure—systems that standardize, structure, and distribute machine-readable option data—may become analogous to financial market infrastructure:
- Financial: Exchanges, clearinghouses, and data providers enable price discovery
- Allocative: Representation protocols, schema standards, and data pipelines enable consideration
If representation affects allocative efficiency, and allocative efficiency affects pricing, then representation infrastructure becomes pricing infrastructure.
Infrastructure Concentration Risks
If allocative infrastructure becomes economically significant, concentration risks emerge:
- Representation infrastructure concentration
- Allocative infrastructure concentration
- Protocol capture risks
These risks are familiar from financial markets and warrant similar scrutiny.
Part VII — Research Agenda
Empirical questions for investigating computational pricing mechanisms
Can Representation Affect Pricing Outcomes?
Methods: Regression analysis, natural experiments, field experiments
Can Inferential Cost Be Measured?
Methods: Observational analysis, experimental manipulation, simulation
Can Admissibility Premiums Be Observed?
Methods: Matching analysis, regression discontinuity, difference-in-differences
Can Computational Liquidity Be Quantified?
Methods: Cross-platform analysis, network analysis, experimental expansion
Conclusion
This paper has introduced Computational Pricing Theory—a framework for understanding how price formation may change when AI systems become the primary allocative interface between buyers and sellers.
Key Contributions
- Historical context on classical pricing models and the accessibility assumption
- The K < n constraint and inferential scarcity as new structural conditions
- Five pricing primitives: Representation Friction, Inferential Cost, Allocative Efficiency, Computational Liquidity, and Admissibility Premium
- Sector examples illustrating theoretical mechanisms
- Infrastructure implications and research agenda
Limitations and Future Work: This paper is entirely theoretical. No empirical claims are advanced. All proposed mechanisms require validation. The research agenda outlined in Part VII represents a path from theory to evidence.
The questions raised by Computational Pricing Theory are not merely academic. If AI systems become dominant allocative interfaces, the structure of competition itself may change. Understanding these structural changes in advance—and developing the measurement and governance frameworks to assess them— may be essential for ensuring that AI-mediated markets remain competitive, fair, and efficient.