Abstract
In reasoning under uncertainty in AI, there are (at least) two useful and different ways of understanding beliefs: the first is as absolute belief or degree of belief in propositions and the second is as belief update or measure of change in belief. Pignistic and plausibility transformations are two well-known probability transformations that map belief functions to probability functions in the Dempster-Shafer theory of evidence. In this paper, we establish the link between pignistic and plausibility transformations by devising a belief-update framework for belief functions where plausibility transformation works on belief update while pignistic transformation operates on absolute belief. In this framework, we define a new belief-update operator connecting the two transformations, and interpret the framework in a belief-function model of parametric statistical inference. As a metaphor, these two transformations projecting the belief-update framework for belief functions to that for probabilities are likened to the fire projecting reality into shadows on the wall in Plato's cave.
Introduction
Reasoning under uncertainty is a cornerstone of Artificial Intelligence. The Dempster-Shafer (DS) theory of evidence offers a robust alternative to standard Bayesian probability by using "belief functions" that can assign mass to sets of outcomes, effectively representing states of partial or total ignorance. However, for practical decision-making, these belief functions must often be transformed into standard probability distributions. Two primary methods have emerged: the pignistic transformation (for absolute belief and decision-making) and the plausibility transformation (for belief updating). Historically, these were viewed as distinct tools. This paper bridges this gap, establishing a formal link by treating them as different "projections" of a unified reasoning process—a metaphor inspired by Plato’s Cave.
Theoretical Framework
The authors introduce a novel mathematical framework for belief updates within DS theory. The centerpiece is a new binary operator, denoted as $\circ$, which combines a prior belief ($m$) with new evidence ($m_{up}$). Unlike the traditional Dempster’s rule of combination ($\oplus$), the $\circ$ operator is non-commutative, reflecting the distinct roles of a "starting state of mind" versus "incoming data."
The framework is built on three axiomatic pillars:
- Composition: Sequential updates are equivalent to a single combined update.
- Plato’s Projection: The transformation commutes with the update process.
- Proportionality: Updates from a state of total ignorance are proportional to the evidence's strength.
In this "Plato's Cave" model, belief functions represent a higher-dimensional reality. The pignistic transformation projects the state of belief, while the plausibility transformation projects the power of the evidence onto the "wall" of standard probability.
Key Results and Comparison
A significant contribution is the derivation of a generalized Bayesian Theorem for belief functions. The authors demonstrate that when the prior is a standard probability distribution, the $\circ$ operator reduces to the classical Bayesian parallel combination, ensuring backward compatibility.
The paper highlights the differences between the new operator and Dempster’s rule through Example 3.3. When starting from a state of total ignorance, Dempster’s rule often distributes mass equally, potentially losing the structure of the evidence. In contrast, the $\circ$ operator preserves this structure more faithfully.
As shown in Figure 1, the $\circ$ operator yields a different distribution than the $\oplus$ operator for the same evidence, particularly in how it handles the set $\{\theta, \bar{\theta}\}$, reflecting a more nuanced interpretation of uncertainty and ignorance.
Conclusion and Future Work
This research establishes a unified link between the pignistic and plausibility transformations, providing a more coherent theoretical foundation for DS theory. Future work includes extending this framework to other probability transformations and exploring its connections to Walley’s theory of imprecise probabilities. Additionally, investigating the computational efficiency of the $\circ$ operator in large-scale AI applications remains an important next step.
Summary updated