Uncertain<T>: Abstractions for Uncertain Hardware and Software

......Computing has entered the era of uncertain data, in which hardware and software generate and reason about estimates. New hardware sensors, such as those found in smartphones, fitness devices, cars, homes, and games, observe the physical world around them. Approximate computing deliberately exploits software robustness and unreliable hardware in the name of efficiency. Analog and neuromorphic systems perform computation on new hardware substrates. Machine learning helps make sense of large, complex data problems. Speech recognition, natural language processing, and other human–computer interactions face the ambiguity of human input. These data sources already produce estimates that millions of people rely on daily—but can we trust them? Despite their ubiquity, economic significance, and societal impact, building applications using these uncertain data sources is surprisingly ad hoc. Most current software and hardware abstraction layers ignore the error in estimates, which leads to uncertainty bugs. One potential solution that researchers are exploring is probabilistic programming languages, which provide abstractions for reasoning about uncertainty, but these languages are intended for programmers with statistical expertise. The richness and generality of these languages poses a high barrier to entry for programmers who lack such expertise. More broadly, the wide ranging and increasing use of estimates in modern software pose correctness, optimization, and programmer productivity problems that current programming languages do not adequately address. Here, we describe Uncertain, a simple programming language abstraction that lets programmers without statistics expertise easily and correctly compute with estimates. Uncertain’s semantics automatically propagate uncertainty in an estimate through computation on that estimate and define a statistical interpretation for conditionals that compute with uncertain values. The Uncertain runtime lazily evaluates James Bornholt