Limits to Prediction

Abstract.  There is a relatively short list of high-level reasons that a system can be hard to predict: (1) the dynamics may be in some sense inherently unpredictable (e.g. chaotic), (2) the financial or energetic cost of measuring the system to sufficient accuracy may be prohibitive, (3) the computational problem of prediction may be difficult (e.g. NP-complete, or worse, uncomputable), even when provided with unlimited data, and/or (4) the state space itself may be unknown, as is the case in systems that adapt, evolve, or innovate. Traditional dynamical systems theory treats (1) and to some extent (2), computational complexity treats (3), and rigorous work on (4) is still ongoing, yet there is very little understanding of the trade-offs and interactions between these four classes of obstacles to prediction. These issues touch on the core limitations of prediction in almost every area of vital importance to society, ranging from weather prediction, to political prediction, to epidemiological forecasting. This workshop will bring together researchers who work on the mathematical, algorithmic, and practical aspects of prediction across these four classes of obstacles, and in a variety of areas, to share knowledge across fields and to try to begin to glean answers to fundamental questions such as:
- In what situations are there absolute limits to prediction, and how are those limits determined?
- What are the trade-offs between the different types of obstacles to prediction mentioned above?
- How can such trade-offs be leveraged to best spend human, financial, and energetic resources?