How can we imbue computational systems with a deeper understanding of differences in human experiences across populations and settings and provide tools that support designers recognizing and reasoning about such differences? Despite significant advances in machine sensing and machine learning technologies—and the availability of rich APIs for creating context-aware applications and conversational agents—there remains little support for creating AI-supported experiences that work well across diverse populations and settings. We fill this gap with _Differ, _a difference-aware computing platform that surfaces differences in how a human experience may be realized across people and places. Our work highlights the need and feasibility of building computational platforms that support designers of AI-supported experiences being sensitive to and reflective of our differences. 

Differ: A Difference-Aware Computing Platform Differ is a difference-aware computing platform that takes as input a machine-interpretable definition of a context-aware experience and outputs to a designer a set of visualizations that surface potential issues in realizing the experience for particular populations of users and in particular settings. Differ has four core computational abstractions that it provides: concept expressions, accountable perspectives,_ issues of concern, and visualizations._

Using these computational abstractions, our prototype provides implementations of (1)  a variety of accountable perspectives that capture meaning differences in geographies (urban/rural, states, city neighborhoods),  user demographics (age group, budget, wheelchair accessibility), and settings (time of day); and (2) a variety of issue functions such as prevalence, conceptual fit, popularity, safety, and affordability. Broadly speaking, implemented reference systems use a combination of Yelp data (e.g., metadata associated with businesses and places), Foursquare data (for popularity and affordability), public city crime datasets (for safety), city neighborhood data, and census data (for urban/rural). To generate the supported visualizations, Differ uses standard Python libraries such as geopandas, geometry, and plotly to generate a visualization for any given concept expression and reference system.