An Analysis of a Welfare Eligibility Determination Interview: A Planning Approach

Eswaran SubrahmanianAAAI 1983

Introduction

In 1983, research into artificial intelligence began exploring how formal planning theory could be applied to complex human interactions. This paper analyzes the welfare eligibility determination interview, treating it as a goal-oriented dialogue guided by structured plans. By mapping conversational protocols to AI frameworks, the study demonstrates how caseworkers navigate the tension between rigid legal requirements and the unpredictable nature of client responses.

The Planning Framework

The analysis draws on two primary AI theories to model professional discourse. First, the Hobbs and Evans (1979) planning mechanism treats conversation as a sequence of goals and actions, categorized into domain goals (the task), discourse goals (flow), and social goals (interpersonal dynamics). Second, the Agar and Hobbs (1982) theory of coherence evaluates dialogue across three levels:

  • Global Coherence: How an utterance fits the overall task.
  • Local Coherence: The relationship between successive sentences.
  • Thematic Coherence: Recurring content-oriented themes, such as "Verification."

The Application Form as a Global Plan

A central insight of the paper is that the physical application form serves as more than just a data entry tool; it acts as a "pre-structured plan" that ensures global coherence throughout the interview. While caseworkers do not always follow the form linearly, its hierarchical structure provides a roadmap for achieving high-level domain goals, such as data collection and benefit calculation.

Figure 1: Hierarchical structure of the eligibility determination interview, showing how domain goals like data collection and verification are organized.

Recursive Sub-plans and Local Coherence

Beyond the global structure, caseworkers employ "mental templates" or sub-plans to handle specific topics. For instance, when a client mentions "odd jobs," the caseworker initiates a recursive query structure to exhaustively gather necessary attributes like pay rate and frequency. This ensures that even "messy" real-world data is captured in a way that satisfies the underlying legal rules.

Figure 2: A planning tree illustrating how caseworkers identify topics and generate query plans to resolve verification problems.

The study identifies five primary domain goals: data collection, verification, explanation of rights/responsibilities, benefit calculation, and form completion. The author successfully maps distinct segments of the interview—such as employment and rights—into hierarchical planning trees. Verification is particularly noteworthy as it functions as both a goal (something to be achieved) and a recurring theme (a topic that resurfaces whenever new data is introduced).

Conclusion

The rigid nature of welfare statutes actually simplifies the AI planning problem by constraining the "search space" for conversational strategies. By identifying these pre-structured conversational plans, the research provides a blueprint for building expert systems that can drive planning-based user interfaces in social services. Future work involves verifying if these planning structures are universal across different caseworkers and adapting them to dynamic legal environments.

Summary updated

BibTeX
@inproceedings{subrahmanian1983analysis,
  author = {Eswaran Subrahmanian},
  title = {{An Analysis of a Welfare Eligibility Determination Interview: A Planning Approach}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {1983},
}
Figure