cs.GTMar 10, 2026

Noncooperative Human-AI Agent Dynamics

Authors: Dylan Waldner, Vyacheslav Kungurtsev, Mitchelle Ashimosi

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

This paper investigates the dynamics of noncooperative interactions between artificial intelligence agents and human decision-makers in strategic environments. In particular, motivated by extensive literature in behavioral Economics, human agents are more faithfully modeled with respect to the state of the art using Prospect Theoretic preferences, while AI agents are modeled with standard expected utility maximization. Prospect Theory incorporates known cognitive heuristics employed by humans, including reference dependence and greater loss aversion relative to utility to relative gains. This paper runs different combinations of expected utility and prospect theoretic agents in a number of classic matrix games as well as examples specialized to tease out distinctions in strategic behavior with respect to preference functions, to explore the emergent behaviors from mixed population (human vs. AI) competition. Extensive numerical simulations are performed across AI, aware humans (those with full knowledge of the game structure and payoffs), and learning Prospect Agents (i.e., for AIs representing humans). A number of interesting observations and patterns show up, spanning barely distinguishable behavior, behavior corroborating Prospect preference anomalies in the theoretical literature, and unexpected surprises. Code can be found at https://github.com/dylanwaldner/noncooperative-human-AI.

Figures & tables

ConfigurationAgent 1 TypeAgent 2 Type
AI-AIExpected Utility MaximizerExpected Utility Maximizer
AI-AHExpected Utility MaximizerAware Human (PT)
AI-LHExpected Utility MaximizerLearning Human (PT)
AH-AHAware Human (PT)Aware Human (PT)
AH-LHAware Human (PT)Learning Human (PT)
LH-LHLearning Human (PT)Learning Human (PT)
Table 1: Experimental Population Configurations
ParameterValue
Learning rate αQ\alpha_{Q}0.01
Initial exploration ϵ\epsilon0.3
Final exploration ϵmin⁡\epsilon_{\min}0.01
Exploration decay0.995 per episode
Belief update rate λb\lambda_{b}0.95
Reference update rate λr\lambda_{r}0.95
Table 2: Experimental Parameters
Figure 4: AH vs AH PD Joint actions by Ref Type
Figure 8: All Matchup Policies (Last 5,0005,000 steps)
Figure 9: Ref Point and CPT Transform Magnitude (EMAOR, Last 5,0005,000 steps)
Figure 13: LH Anomaly: Matchup Policies, CPT/EU L2, and Ref. Points (EMA, Last 5,0005,000 steps)
Figure 14: One Shot Anomalous LH Q Values and Q Value Change (EMAOR vs AH)
Figure 15: Repeated Anomalous LH Agents (Player 1, left; Player 2, middle) Q vals and Q value change (EMA)
Figure 16: AH vs AH Baseline (EMA)
Figure 20: Converged Policies One Shot MP (Last 5,0005,000 steps)
Figure 24: Action Changes One Shot MP (Last 5,0005,000 steps)
Figure 28: Policies Repeated MP (Last 5,0005,000 steps)
Figure 32: CPT Induced Action Changes Repeated MP (Last 5,0005,000 steps)
Figure 36: AH vs AH Baseline Stag Hunt
Figure 40: One Off Stag Hunt Matchup Policies (Last 5,0005,000 steps)
Figure 44: Repeated Stag Hunt Matchup Policies (Last 5,0005,000 steps)
Figure 48: One Off Battle of the Sexes AH vs AH Baseline
Figure 52: One Off Battle of the Sexes Matchup Policies (Last 5,0005,000 steps)
Figure 56: One Off Battle of the Sexes LH Q Values (vs. AI)
Figure 60: Repeated Battle of the Sexes Policies (Last 5,0005,000 Steps)
Figure 61: LH vs. AI Q values: Clear preference vs. Indifference (V-based)
Figure 65: AH vs. AH baseline Chicken
Figure 69: One Off Chicken Policies (Last 5,0005,000 Steps)
Figure 73: Repeated Chicken Policies (Last 5,0005,000 Steps)
Figure 77: Repeated Chicken Avg. Rewards (Last 5,0005,000 Steps)
Figure 81: One Off Crawford’s Counterexample Policies by Matchup (Last 5,0005,000 Steps)
Figure 85: Repeated Crawford’s Counterexample Policies by Matchup (Last 5,0005,000 Steps)
Figure 86: AH vs. AH Baseline Policy (EMA)
Figure 90: One Off Ochs’ Game Policies by Matchup (Last 5,0005,000 Steps)
Figure 94: One Off Ochs’ Game CPT Action Change Rates by Matchup (Last 5,0005,000 Steps)
Figure 98: Repeated Ochs’ Game Policies by Matchup (Last 5,0005,000 Steps)
More in this topic

Prospect Theory

See all 9 papers

Explore similar work

CardsList

  1. Cooperate to Compete: Strategic Coordination in Multi-Agent Conquest

    Apr 28, 2026Abigail O'Neill, Alan Zhu, Mihran Miroyan +2Multi-Agent CoordinationMulti-Agent Negotiation

    University of California, Berkeley

  2. Serious Games: Human-AI Interaction, Evolution, and Coevolution

    May 22, 2025Nandini Doreswamy, Louise HorstmanshofHuman-AI InteractionHuman-AI Decision Making

    Southern Cross University, Lismore, New South Wales, Australia · National Coalition of Independent Scholars · National Coalition of Independent Scholars (NCIS)

  3. Competition and Cooperation of LLM Agents in Games

    Apr 1, 2026Jiayi Yao, Cong Chen, Baosen ZhangLLM AgentsMulti-Agent LLM Systems

    Department of Electrical and Computer Engineering, University of Washington, Seattle. · Thayer School of Engineering at Dartmouth College.