Algorithms used by the justice system, banks, and private companies are sometimes biased, according to researchers at the Massachusetts Institute of Technology (MIT). They disproportionately affect people of color and people in lower-income brackets when they apply for loans, apply for jobs, or even go to court to set bail, for example.
To restore fair equality, researchers have developed a new artificial intelligence (AI) programming language that can assess the fairness of algorithms more accurately and quickly than available alternatives. Their „Sum-Product Probabilistic Language“ (SPPL) is a probabilistic programming system at the intersection of programming languages and AI. It aims to make AI systems easier to develop. There have already been successes with SPPL in computer vision, data cleaning, and automated data modeling.
„SPPL is a probabilistic programming language that delivers exact solutions to a broad range of probabilistic inference queries. The language handles continuous, discrete, and mixed-type probability distributions; many-to-one numerical transformations; and a query language that includes general predicates on random variables.
Users express generative models as probabilistic programs with standard imperative constructs, such as arrays, if/else branches, for loops, etc. The program is then translated to a sum-product expression (a generalization of sum-product networks) that statically represents the probability distribution of all random variables in the program. This expression is used to deliver answers to probabilistic inference queries.“ (See here)
According to Feras Saad, a doctoral student in electrical engineering and computer science, similar systems exist, „but ours is specialized and optimized for a specific class of models, so it can deliver solutions 1,000 times faster.“ Fairness questions SPPL can answer include „How likely is the model to recommend credit to someone over 40?“ or „What is the likelihood of being hired if the candidate is qualified for the job and from an underrepresented group?“ (See also)
0 Kommentare zu “Fairness in Artificial Intelligence”