TL;DR
This paper introduces a decision theory for "Symbolic Probability" (SP), a framework that generalizes standard probability by using ordered symbols instead of numbers to represent uncertainty. This allows for more flexible reasoning in situations where precise numerical probabilities are unavailable or inappropriate, while still preserving desirable Bayesian inference patterns.
What Problem Does This Paper Address?
Standard probability theory, particularly in its Bayesian formulation, is incredibly powerful for decision-making under uncertainty. However, it relies on assigning precise numerical probabilities to events. This can be problematic in real-world scenarios where information is incomplete, leading to situations where exact numerical probabilities are impossible to determine or where human intuition systematically violates the axioms of probability (as seen in phenomena like the Ellsberg paradox). Symbolic Probability (SP) was developed to address this by using ordered symbolic supports instead of numbers. While SP provides a way to represent uncertainty and perform inference, it lacked a formal decision theory that explains how preferences over actions (acts) can be derived from these symbolic supports, analogous to how expected utility theory works with numerical probabilities. This paper aims to fill that gap by developing such a decision theory for SP.
Methods
The core of the paper is the development of a decision theory for SP based on a set of postulates for preference relations on acts. The authors define acts as contracts that reward a decision-maker with prizes based on the occurrence of events. They introduce the concept of "canonical acts," which are simplified representations of more complex acts, making them amenable to comparison.
Key methodological contributions include:
- Postulates for Preference Relations: A set of axioms (D1-D5) are proposed for a preference relation (▷) on acts. These postulates ensure consistency (transitivity), allow for substitution of equivalent acts, and establish a connection between preference and support comparison.
- Binary Utility Function: The paper shows that a preference relation satisfying these postulates can be represented by a "binary utility function." This function maps acts to pairs of symbolic supports, where one symbol represents the "gain" component (associated with the best event of a canonical act) and the other represents the "loss" component (associated with the worst event).
- Connection to Support Comparison: The decision theory explicitly links preference for acts to the comparison of their symbolic supports. Specifically, an act is preferred if its "gain" component is more supported and its "loss" component is less supported, mirroring intuitive decision-making principles.
- Formalization of Ellsberg Paradox: The paper demonstrates how their framework can represent and provide a decision-theoretic explanation for the Ellsberg paradox, a classic example of behavior that deviates from standard probability.
Datasets
Not discussed in the paper. The paper focuses on theoretical framework development and uses hypothetical examples like the Ellsberg paradox for illustration.
Key Insights and Ideas
- Subjective Interpretation for SP: The paper argues that SP, like numerical probability, can be given a useful and appropriate subjective interpretation. This is crucial because it provides a basis for eliciting uncertainty from behavior.
- Bridging Belief and Choice: The developed decision theory establishes a direct link between symbolic beliefs (supports) and preferences over actions. This duality is a core tenet of the approach.
- Flexibility Beyond Numerical Probability: SP, and consequently this decision theory, offers a more flexible way to model uncertainty when precise numerical probabilities are unavailable or when agents exhibit behavior that violates standard probability axioms.
- Binary Utility as a Generalization: The binary utility function is presented as a generalization of expected utility, capable of handling non-probabilistic uncertainty representations.
- Intuitive Decision-Making Captured: The postulates and the resulting binary utility function capture intuitive notions of decision-making: maximizing gains and minimizing losses, while being indifferent to neutral outcomes.
Key Results
- A preference relation on acts satisfying the proposed postulates can be represented by a binary utility function.
- This binary utility function's components correspond to the symbolic supports of the best and worst events in a canonical act.
- The decision theory provides a formal mechanism for comparing acts based on their symbolic supports, offering a way to make decisions in the SP framework.
- The framework successfully explains the preferences observed in the Ellsberg paradox, demonstrating its utility for situations that elude standard probability.
Limitations and Future Work
- Completeness of Preference: The paper does not require the preference relation to be complete, meaning that for some pairs of acts, a decision-maker might be indifferent or unable to decide between them. This reflects the inherent vagueness of symbolic uncertainty.
- Reverse Question: The authors acknowledge that they have not yet addressed the reverse question: what behavior characterizes the SP structure itself.
- Applications: The paper suggests future work in applying these techniques to domains like medical diagnosis, where eliciting preferences from experts to construct decision-making frameworks is a significant challenge. The goal is to use these frameworks as diagnostic aids.
- Computational Aspects: While the theory is developed, the computational implications and efficiency of applying this decision theory in practice are not deeply explored.
Summary updated