Research Summary: A Unifying Perspective on Probabilities as Model Predictions

This research from arXiv presents a unifying perspective on the understanding of probabilities, arguing that all probabilistic statements are outputs of specific prediction methods. It contends that probabilities are inherently model-dependent, even those traditionally considered objective, as they rely on particular abstractions and transformation processes. The paper suggests that meeting a finite calibration criterion allows for anticipation of utility distributions.

Why it matters

This research is strategically important because it fundamentally redefines the nature of probabilistic statements, which are central to decision-making across all domains. Understanding probabilities as model-dependent outputs impacts how organizations interpret risk, forecast outcomes, and develop strategies based on uncertain information, necessitating careful consideration of the underlying models and their limitations. It highlights the need for transparency and robustness in predictive modeling to ensure reliable strategic guidance.

Key insights

• All probabilities are conceptualized as outputs of a prediction method, involving both abstraction construction and transformation into predictions.
• This perspective unifies different types of probabilities, resolving foundational disagreements between schools of thought like Bayesians and frequentists.
• Probabilities, including those deemed 'objective', are inherently dependent on the specific underlying model.
• Meeting a finite calibration criterion enables the anticipation of utility distributions when acting on probabilistic statements.
• The inherent model-dependence of probabilities challenges traditional views of their objectivity and reliability.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2609.09855