This article focuses on addressing the opacity issue in sub-symbolic machine learning predictors by promoting two complementary activities: symbolic knowledge extraction (SKE) and symbolic knowledge injection (SKI). SKE involves distilling the knowledge learned by a sub-symbolic predictor into a symbolic form, making it more interpretable. SKI, conversely, involves encoding symbolic knowledge into a sub-symbolic predictor to enhance its behavior with human-like commonsense. The authors propose general meta-models for both SKE and SKI and develop two taxonomies to classify existing methods. They analyze 132 SKE methods and 117 SKI methods, categorizing them based on purpose, operation, input/output data, and predictor types. The article aims to provide a systematic literature review, collect and categorize methods, and offer insights into the state-of-the-art and future research directions. The work is relevant for data scientists, researchers, and developers interested in improving the interpretability and control of sub-symbolic predictors.This article focuses on addressing the opacity issue in sub-symbolic machine learning predictors by promoting two complementary activities: symbolic knowledge extraction (SKE) and symbolic knowledge injection (SKI). SKE involves distilling the knowledge learned by a sub-symbolic predictor into a symbolic form, making it more interpretable. SKI, conversely, involves encoding symbolic knowledge into a sub-symbolic predictor to enhance its behavior with human-like commonsense. The authors propose general meta-models for both SKE and SKI and develop two taxonomies to classify existing methods. They analyze 132 SKE methods and 117 SKI methods, categorizing them based on purpose, operation, input/output data, and predictor types. The article aims to provide a systematic literature review, collect and categorize methods, and offer insights into the state-of-the-art and future research directions. The work is relevant for data scientists, researchers, and developers interested in improving the interpretability and control of sub-symbolic predictors.