Abstract
This paper presents a computational process model for the reasoning underlying arguments of persuasion. Implemented in a program called PERSUADER, the system provides counsel for resolving impasses in collective bargaining by searching hierarchical goal trees to generate persuasive economic arguments.
Introduction: From Adversarial Logic to Persuasion
In the mid-1980s, artificial intelligence research into argumentation primarily focused on adversarial logic—the art of winning a debate. However, labor mediation requires a shift from "winning" to "persuading." A mediator’s objective is not to prove a party wrong, but to convince both sides to accept concessions they initially resisted. PERSUADER addresses this by modeling cooperative problem solving, where the goal is conflict resolution rather than total victory.
Methodology: Goal Trees and Qualitative Reasoning
The system represents the conflicting interests of companies and unions through hierarchical Goal Trees.
Goal Representation
Each party is assigned a tree where nodes represent economic quantities. Each node is assigned a sign: (+) indicates a goal to increase that quantity, while (-) indicates a goal to decrease it.
Causal Propagation
"Support links" within these trees demonstrate how subgoals affect higher-level objectives. For instance, decreasing "Labor Costs" (-) directly supports the higher-level goal of increasing "Profits" (+).
Generating "Threatening" Arguments
To nudge a party toward compromise, PERSUADER identifies "threatening" arguments. The process involves:
- Identifying Violations: Determining which of the company’s goals are violated by a union's demand.
- Searching for Compensating Actions: Finding actions the company might take to offset that violation.
- Evaluating Impact: Checking if these compensating actions violate a union goal that the union values more than its original demand.
The system ranks the "convincing power" of these arguments, ranging from appeals to universal principles to the high-stakes threat of strikes or lockouts.
Case Study: The Transit Company Impasse
The model was tested using a case study of a transit company dispute. In this simulation, the union demanded a wage increase (Importance: 6). PERSUADER traced this demand through the company's goal tree:
- Higher Wages $\rightarrow$ Higher Labor Costs $\rightarrow$ Reduced Profits.
- To compensate for reduced profits, the system identified that the company could reduce Employment (layoffs).
- Because the union valued "Employment" (Importance: 8) more than the "Wage" increase (Importance: 6), the system generated the persuasive argument: "If the company is forced to grant higher wages, then it will lay off workers."
Key Insights and Limitations
A central finding is that showing how a refusal threatens a more important goal is often more effective than showing how a proposal furthers a goal. Furthermore, the model demonstrates that qualitative reasoning—focusing on the direction of change (increase/decrease) rather than exact dollar amounts—is often sufficient for modeling logical flow in negotiations.
However, the system relies on manually encoded goal trees and importance values. Future work is needed to automate the generation of "reasonable values" for demands and to better model arguments that highlight the positive benefits of proposals.
Conclusion
PERSUADER demonstrates that complex, multi-step persuasive arguments can be generated through the systematic search of goal trees. By focusing on qualitative economic relationships, the system provides a framework for computational mediation in high-stakes labor disputes.
Summary updated