Enterprise AI Analysis
Choice via AI
Authored by CHRISTOPHER KOPS & ELIAS TSAKAS
Published: February 5, 2026
This paper proposes a model of choice via agentic artificial intelligence (AI). A key feature is that the AI may misinterpret a menu before recommending what to choose. A single acyclicity condition guarantees that there is a monotonic interpretation and a strict preference relation that together rationalize the AI's recommendations. Since this preference is in general not unique, there is no safeguard against it misaligning with that of a decision maker. What enables the verification of such AI alignment is interpretations satisfying double monotonicity. Indeed, double monotonicity ensures full identifiability and internal consistency. But, an additional idempotence property is required to guarantee that recommendations are fully rational and remain grounded within the original feasible set.
Keywords: WARP, preferences, AI | JEL Codes: D01, D91
Executive Impact & Key Metrics
Our analysis of "Choice via AI" reveals critical insights for optimizing enterprise AI deployments, ensuring alignment and verifiable rationality.
0
Research Papers Analyzed
0
Decision Alignment Achievable
0
Efficiency Gain Potential
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
AI Choice Modeling
This paper introduces the AI Agent's Choice (AIC) model, where an AI misinterprets a menu before recommending a choice. The interpretation operator I is monotonic (if S ⊆ T, then I(S) ⊆ I(T)). A choice function c is an AIC if there exists a strict preference > and I such that c(S) is the >-best element in I(S).
Rationality & Interpretability
A key finding is that a single No Shifted Cycles (NSC) condition characterizes AIC choice functions, guaranteeing rationalizable recommendations. However, initial AIC models only allow partial identification of underlying preferences and interpretation, making full verification of AI alignment challenging.
Double Monotonicity & Identifiability
The Rational AI Agent's Choice (RAIC) model introduces Double Monotonicity for the interpretation operator (I(S) ⊆ I(T) ⇔ S ⊆ T). This ensures full identifiability of both the AI's preferences and its interpretation operator, making AI alignment verifiable through behavioral axioms like No Binary Cycles, C-Contraction Independence, and Noticeable Difference.
Grounded AI Decisions (WARP)
The most robust model, Grounded and Rational AI Agent's Choice (GRAIC), incorporates an Idempotence property for the interpretation (I(I(S)) = I(S)). This prevents interpretive loops, ensuring choices are always from the actual feasible set. GRAIC is characterized by the Weak Axiom of Revealed Preference (WARP), achieving traditional economic rationality and full identification of AI's internal logic.
Core AI Choice Model Characterization
Acyclic ChoiceA single condition (NSC) characterizes AI Agent's Choice, ensuring rationalizable recommendations despite misinterpretations.
Enterprise Process Flow
DM seeks advice
→
AI misinterprets menu (I(S))
→
AI chooses best (x>y) from I(S)
→
DM receives recommendation (c(S)=x)
| Feature | AIC (Monotonic I) | RAIC (Double Monotonic I) |
|---|---|---|
| Preference Identification | Partial | Full |
| Interpretation Clarity | Distorted | Order Isomorphic |
| Behavioral Axioms |
|
|
Grounded and Rational AI Agent Choice (GRAIC)
The GRAIC model ensures AI recommendations are fully rational and grounded, aligning with traditional economic WARP. This requires an idempotent interpretation operator (I(I(S))=I(S)), preventing interpretive loops and ensuring choices are always from the actual feasible set. This ensures the AI's reasoning is both sound and transparent.
Key Takeaway: WARP as a characterization ensures complete identification of preferences and interpretation, providing a robust framework for verifiable AI rationality.
Quantify Your AI Impact
Calculate the potential annual savings and reclaimed human hours by deploying rational and grounded AI agents in your enterprise.
Potential Annual Savings
$0
Human Hours Reclaimed
0
Your Enterprise AI Roadmap
A phased approach to integrate rational AI decision-making into your operations, ensuring verifiable alignment and optimal outcomes.
Phase 1: Diagnostic Assessment & Model Definition
Duration: 4-6 Weeks
Identify key decision points, define initial AI choice models, and establish baseline performance metrics.
Phase 2: Interpretation Operator Calibration
Duration: 6-10 Weeks
Train and fine-tune AI interpretation operators for monotonicity and double monotonicity, focusing on accurate menu understanding.
Phase 3: Preference Alignment & Rationalization
Duration: 8-12 Weeks
Implement and validate AI preference relations, ensuring consistency with behavioral axioms like NSC, NBC, CCI, and ND.
Phase 4: Grounded Deployment & WARP Verification
Duration: 6-8 Weeks
Integrate idempotent interpretation and verify WARP satisfaction, ensuring AI recommendations are truly rational and grounded in feasible sets.
Phase 5: Continuous Monitoring & Optimization
Duration: Ongoing
Establish monitoring frameworks to track AI performance, identify potential misalignments, and continuously optimize choice models for evolving business needs.
Ready to Build a Rational AI Strategy?
Unlock the full potential of your enterprise AI with a strategy rooted in verifiable rationality and perfect alignment.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat