Introduction
In the early development of artificial intelligence, researchers prioritized physical reasoning and goal-oriented planning. While these systems could navigate spatial environments or solve logic puzzles, they often failed to grasp the invisible social and legal constraints that govern human behavior. In his 1980 paper, James R. Meehan argues that for AI to truly understand natural language and human narratives, it must possess a formal representation of authority structures—the complex web of laws, obligations, and power dynamics that define social interaction.
The Social Representation Gap
Meehan identifies a significant gap in the knowledge representation of the era. Consider a simple narrative: "Mary forgot to renew a library book; they sent her a bill." A standard goal-based AI might struggle to explain why the bill was sent. Without an understanding of the "library" as a social group with specific rules and the authority to impose fines, the connection between a forgotten book and a financial penalty remains opaque.
To address this, Meehan proposes a shift toward "social physics," where human behavior is analyzed not just through individual desires, but through the social forces—permissions, prohibitions, and requirements—that act upon them.
A Symbolic Framework for Authority
The paper adopts a symbolic knowledge representation approach, defining a taxonomy of concepts designed to be programmed into a reasoning system.
1. The Group
The fundamental unit of Meehan’s analysis is the Group. A group consists of participants, a "social net" (the attitudes and relationships between members), and "normal procedures" (the roles and activities expected within that context).
2. Six Types of Legality
Meehan categorizes every act into one of six legal statuses:
- Explicitly Legal: Acts protected by right (e.g., free speech).
- Permissible via License: Acts that require specific permission (e.g., practicing medicine).
- Commanded: Acts required by a superior authority (e.g., a child taking medicine).
- Agent-based: Acts performed on behalf of an authority (e.g., a police officer conducting a search).
- Legally Required: Acts dictated by the rules of a specific system (e.g., following suit in a game of bridge).
- Explicitly Forbidden: Acts that are prohibited.
3. Primitives of Authority
Meehan identifies five "primitives"—the core actions an authority can perform. These include defining the legal status of an act, enforcing decisions, and revising the rules of the group itself.
Power, Politics, and Social Indebtedness
A key insight of the paper is the distinction between Authority and Enforcement. Using historical examples like Tudor England’s "Letters of Reprisal," Meehan demonstrates that a court can grant legal permission for an act (such as seizing a ship) without possessing the physical power to enforce it.
Furthermore, Meehan defines Power as the ability to make things happen and Politics as the act of influencing authority-related decisions. He introduces the concept of Social Indebtedness, suggesting that power functions like a currency. When one individual acts to increase another's power, a "debt" is incurred, creating a social expectation of future reciprocation.
Implications for Natural Language Understanding (NLU)
By integrating these authority structures, Meehan demonstrates that an NLU program can achieve several critical functions:
- Categorization: Identifying the legal or social status of any action within a story.
- Explanation: Connecting seemingly unrelated events (like the library book and the bill) through the lens of social rules.
- Prediction: Forecasting future actions based on power dynamics. For instance, if a mother tells her son to go to bed, the system predicts he will comply; if a younger sister gives the same command, the system predicts he will not, based on the lack of an authority relationship.
Conclusion and Limitations
The primary challenge identified by Meehan is the knowledge engineering bottleneck. In 1980, every social nuance and legal rule had to be manually encoded by programmers. While authority structures are ubiquitous—appearing in everything from children's bedtime routines to international diplomacy—representing the full breadth of human social life remains a monumental task.
Despite these challenges, Meehan’s work remains a foundational exploration of how AI might move beyond "common sense" physics into the even more complex realm of "social physics."
Summary updated