Human surrogate
Must match human behavior at the relevant level
93 primary studies coded with this role
Three decision levels and a validation layer
Fig. 2 of the review as an interactive diagram. Select a level to see how the review defines it, how much evidence sits there, and where the review discusses it.
Decision levels and cross-cutting validation
Aggregation
Pipeline outputs are the orders others face
Decision rules enter pipelines as components
Level 1. Individual decision
Individual decisions refer to choices, estimates, forecasts, or advice elicited based on a specific prompt and instrument.
In Section 4: Individual financial behavior
- 4.1 Consistency without human-likeness
A widely replicated individual-level finding is that capable models choose consistently when tasks are presented in a standardized format. - 4.2 Which biases transfer?
Bias batteries often find human-direction effects, whereas the incidence, size, and occasionally the sign of those effects vary across tasks. - 4.3 Reference points in the prompt
Prompted numbers usually move judgments, but their effects vary by model, source, and task and can be absent (ten designs are compared in Online Resource 1). - 4.4 Priors, identity cues, and advice
Models share some priors but differ within them. - 4.5 Beliefs, expectations, and calibration
Synthetic respondents often reproduce aggregate patterns better than dispersion.
Level 2. Coordination within a decision system
Coordination within a decision system combines specialized LLM roles within a pipeline to serve a single objective.
In Section 5: Coordination within decision systems
- An LLM agent entering a market is often a pipeline whose aggregation rule shapes its decisions.
- The LLM selected the final action in 171 of the 295 systems, shared the selection with another policy, optimizer, or rule in 58, and supplied only signals or features in 54; 12 were unclassified.
- The 54 signal-only systems stay within scope because the LLM’s judgment, forecast, proposed strategy, or advice is evaluated, either directly or through the decision it feeds.
Level 3. Interaction among participants
Interaction among economic participants concerns separately motivated agents that affect one another through prices, payoffs, networks, or shared advice, including human–AI markets.
In Section 6: Interaction and collective outcomes
- 6.1 Replicating laboratory asset markets
Replications of canonical experiments show that the model populating a market determines its results. - 6.2 Institutions, auctions, and information aggregation
In one study, varying both the institution and the model family materially changed the outcomes. - 6.3 Simulated markets and systemic episodes
In larger simulated markets, aggregate patterns are sensitive to the imposed heterogeneity, prompt design, and information environment. - 6.4 Population composition and adaptation
Most replications vary the model across markets, each containing a single model, holding the institution and instructions fixed (Section 6.1). - 6.5 Dependence among agents as an aggregation channel
The interaction studies document dependence among LLM agents, and two of them also measure it outside the market, before any interaction.
Level 4. Validation across levels
Validation across levels asks whether a measurement supports its use at the next level; it cuts across the three decision levels rather than forming another aggregation stage.
It includes training data contamination, prompt and persona sensitivity, run and version variation, human benchmarks, evaluation design, and the inferential unit.
In Section 7: Threats to validity
- 7.1 Look-ahead, memorization, and contamination
Within training windows, recall and inference are difficult to separate. - 7.2 Prompt sensitivity, frames, and personas
The behavior of LLMs is frame-dependent in ways that matter for aggregation. - 7.3 Output variability, drift, and replicability
Repeated outputs can vary under a fixed model configuration, and setting the temperature to zero does not guarantee determinism. - 7.4 Hyper-rationality and the human benchmark
Human-likeness depends on the comparator, and a human comparator is necessary only for claims about human fidelity. - 7.5 Evaluation practice and the inferential unit
In audits that re-run published LLM trading systems over later or longer windows or with realistic costs, the reported gains shrink or disappear (Li et al., 2025b; Li et al., 2026a; Ye et al., 2026).
Aggregate outcomes
LLM agents alter two inputs to this machinery. First, their heterogeneity depends on the chosen configuration. Researchers specify or sample models, versions, personas, information sets, and endowments. For human-surrogate claims, the resulting distribution of types requires validation against a human population; for participant claims, it should represent the AI configurations being studied. Second, the construction may correlate their errors and actions. Shared weights, post-training, prompts, interfaces, or information sources can produce common systematic errors even when calls are sampled independently. The literature on algorithmic monoculture predicts that correlated decisions can lower welfare, even when each individual decision improves (Kleinberg & Raghavan, 2021). These two features, configured heterogeneity and dependence among agents, are the lens through which this review reads the interaction evidence.
Diagram text from Fig. 2; definitions quoted from Section 3.1; counts from Online Resource 1 (Fig. 3 rows; roles overlap).
Roles and the validation standard each implies
Agent roles imply different validation criteria. The human surrogate must reproduce relevant human behavior. A financial decision tool must demonstrate performance out of sample and net of relevant frictions. An AI economic participant must be accurately described, regardless of whether it resembles a human. An experimental instrument must stably measure the intended construct across different wordings, runs, and versions. A paper may claim several roles, but evidence meeting one standard need not meet another: profitable trades do not establish human fidelity, and a human-like answer does not establish stable measurement of a behavioral construct. These four roles correspond to four of the five claims in Section 8; the fifth, that the LLM component explains outcomes beyond simpler agent rules, can accompany any of them.
Financial decision tool
Must perform out of sample, net of frictions
158 primary studies coded with this role
AI economic participant
Must be described accurately as configured and deployed
29 primary studies coded with this role
Experimental instrument
Must measure what it claims, stably across wordings, runs, and versions
73 primary studies coded with this role