Functional Decision Theory: A New Theory of Instrumental Rationality

Abstract: This paper describes and motivates a new decision theory known as functional decision theory (FDT), as distinct from causal decision theory and evidential decision theory. Functional decision theorists hold that the normative principle for action is to treat one's decision as the output of a fixed mathematical function that answers the question, "Which output of this very function would yield the best outcome?" Adhering to this principle delivers a number of benefits, including the ability to maximize wealth in an array of traditional decision-theoretic and game-theoretic problems where CDT and EDT perform poorly. Using one simple and coherent decision rule, functional decision theorists (for example) achieve more utility than CDT on Newcomb's problem, more utility than EDT on the smoking lesion problem, and more utility than both in Parfit's hitchhiker problem. In this paper, we define FDT, explore its prescriptions in a number of different decision problems, compare it to CDT and EDT, and give philosophical justifications for FDT as a normative theory of decision-making.

• The paper introduces Functional Decision Theory (FDT) as a new theory of instrumental rationality, aiming to outperform existing Causal (CDT) and Evidential Decision Theories (EDT).
• FDT treats actions as the output of a decision function and demonstrates higher utility than CDT and EDT in classic problems like Newcomb's Problem and Parfit's Hitchhiker Problem.
• FDT leverages "subjunctive dependence" to handle logical interdependencies beyond simple causality or correlation, offering potential enhancements for AI decision-making in prediction-influenced scenarios.

Overview of Functional Decision Theory: A New Theory of Instrumental Rationality

The paper "Functional Decision Theory: A New Theory of Instrumental Rationality" by Eliezer Yudkowsky and Nate Soares presents an innovative approach to decision theory, termed Functional Decision Theory (FDT). This theory is posited as superior to both Causal Decision Theory (CDT) and Evidential Decision Theory (EDT), offering a unified framework for achieving higher utility across a range of decision-theoretic and game-theoretic challenges.

Core Proposition

FDT contends that actions should be treated as the output of a fixed mathematical function. This perspective contrasts with CDT and EDT by centering decision-making not on the physical act or evidence-based outcomes but on a decision function that produces the best outcome. The theorists propose that FDT delivers higher utility in classical problems where CDT and EDT traditionally underperform.

Key Decision Problems

The paper exemplifies the efficacy of FDT through various well-known decision problems:

• Newcomb’s Problem: FDT exceeds the utility of CDT by endorsing one-boxing, correlating actions with predictions in a manner that reflects logical dependency.
• The Smoking Lesion Problem: FDT avoids the pitfalls of EDT by smoking, breaking the non-causal correlation between action and lesion.
• Parfit’s Hitchhiker Problem: Demonstrating its comprehensive application, FDT attains higher utility than both CDT and EDT, reinforcing rational follow-through on commitments.

Comparative Analysis

Within the field of game theory, FDT offers a consistent strategy across single and multi-agent environments by accounting for predictor and agent interdependencies absent in CDT and EDT. Further, the paper critiques CDT’s reliance on causal independence and EDT’s inadequate handling of statistical correlations, highlighting their inability to resolve paradoxes optimally.

Subjunctive Dependence

A critical insight of FDT is the concept of “subjunctive dependence,” akin to causal dependence in CDT but extending to logical and mathematical interdependencies. This approach ensures that consistent decision-making aligns with the derived outcomes optimal for an agent's computational model, breaking traditional limits regarding statistical and causal rigidity.

Implications and Future Work

The theoretical implications of FDT suggest enhancements in AI decision-making efficacy by addressing prediction-influenced scenarios robustly. Practically, this theory advocates for computational models capable of broader reasoning beyond immediate cause-and-effect, rivalling conventional decision theories in decision-making depth. Furthermore, future research is encouraged to formalize the identification and assessment of subjunctive dependencies, enhancing the accuracy and applicability of FDT.

Conclusion

"Functional Decision Theory: A New Theory of Instrumental Rationality" redefines the landscape of decision theory with its novel approach, maximizing utility by treating decision functions as central to reasoning, thereby overcoming limitations encountered in CDT and EDT. Through FDT, the authors provide a framework that supports rational, utility-oriented decision-making, a significant milestone that necessitates further exploration and refinement in both theoretical and applied domains.