Predicting Decisions of AI Agents from Limited Interaction through Text-Tabular Modeling

2026-05-12

AI Summary

This study addresses the challenge of accurately predicting the next move of an unknown AI opponent in natural language negotiation settings, where interactions are limited and the adversary’s internal mechanisms are opaque. The problem is formulated as a goal-oriented text-to-table prediction task, and the paper introduces an LLM-as-Observer framework: rather than using a large language model (LLM) directly for few-shot reasoning, its parameters are frozen and its hidden states are leveraged as an observation encoder to extract decision-relevant signals. By integrating structured game states, dialogue history, and a small number of interaction examples, the approach jointly models these inputs with a tabular foundation model and employs target-adaptive learning. Evaluated on 91 previously unseen AI agents, the method substantially outperforms baselines, achieving a ~4 percentage point gain in AUC and a 14% reduction in bargaining offer prediction error.

Abstract

AI agents negotiate and transact in natural language with unfamiliar counterparts: a buyer bot facing an unknown seller, or a procurement assistant negotiating with a supplier. In such interactions, the counterpart's LLM, prompts, control logic, and rule-based fallbacks are hidden, while each decision can have monetary consequences. We ask whether an agent can predict an unfamiliar counterpart's next decision from a few interactions. To avoid real-world logging confounds, we study this problem in controlled bargaining and negotiation games, formulating it as target-adaptive text-tabular prediction: each decision point is a table row combining structured game state, offer history, and dialogue, while $K$ previous games of the same target agent, i.e., the counterpart being modeled, are provided in the prompt as labeled adaptation examples. Our model is built on a tabular foundation model that represents rows using game-state features and LLM-based text representations, and adds LLM-as-Observer as an additional representation: a small frozen LLM reads the decision-time state and dialogue; its answer is discarded, and its hidden state becomes a decision-oriented feature, making the LLM an encoder rather than a direct few-shot predictor. Training on 13 frontier-LLM agents and testing on 91 held-out scaffolded agents, the full model outperforms direct LLM-as-Predictor prompting and game+text features baselines. Within this tabular model, Observer features contribute beyond the other feature schemes: at $K=16$, they improve response-prediction AUC by about 4 points across both tasks and reduce bargaining offer-prediction error by 14%. These results show that formulating counterpart prediction as a target-adaptive text-tabular task enables effective adaptation, and that hidden LLM representations expose decision-relevant signals that direct prompting does not surface.

Problem

• AI agent prediction
• limited interaction
• text-tabular modeling
• counterpart modeling
• decision prediction

Innovation

• text-tabular modeling
• LLM-as-Observer
• target-adaptive prediction
• counterpart modeling
• decision prediction