Machine learning algorithms pervade contemporary society. They are integral to social institutions, inform processes of governance, and animate the mundane technologies of daily life. Consistently, the outcomes of machine learning reflect, reproduce, and amplify structural inequalities. The field of fair machine learning has emerged in response, developing mathematical techniques that increase fairness based on anti-classification, classification parity, and calibration standards. In practice, these computational correctives invariably fall short, operating from an algorithmic idealism that does not, and cannot, address systemic, Intersectional stratifications. Taking present fair machine learning methods as our point of departure, we suggest instead the notion and practice of algorithmic reparation. Rooted in theories of Intersectionality, reparative algorithms name, unmask, and undo allocative and representational harms as they materialize (American English sp) in sociotechnical form. We propose algorithmic reparation as a foundation for building, evaluating, adjusting, and when necessary, omitting and eradicating machine learning systems.

In type 1 diabetes (T1D), long-term blood sugar control measured by HbA1c can vary over a wide range. Although metabolites are commonly studied in type 2 diabetes (T2D) as they relate to HbA1c, similar studies in T1D are few. Since the underlying etiologies of T1D and T2D are different, we explored the overall pathways of metabolism in T1D subjects with tight vs. poor blood sugar control. T1D subjects (n=50) were divided into two groups (n=25 each) based on low vs. high HbA1c (Screen 1, low: mean 6.7 ± 0.1%; high: mean 8.7 ± 0.3%). Serum samples were analyzed on a metabolomic GC/HPLC/MS platform and analyzed for differences in biochemicals between the groups. The screen was repeated with a second group of 50 patients for verification (Screen 2, mean 6.3 ± 0.1%; mean 8.0 ± 0.2%) and then both screens combined. For each detected biochemical, average scaled intensity was determined and statistical significance of differences was calculated using a two-tailed, unpaired Student’s T test (p-values) and False Discovery Rate (q-values). The platform distinguished 690 components in Screen 1 and 623 biochemicals in Screen 2. In Screen 1, two biochemicals, including glucose, showed significant differences (≦0.05) for both p and q value. In Screen 2, none of the biochemicals reached threshold. When the screens were combined (100 samples), 10 biochemicals were significant for both p and q. Three components (glucose, mannose and 1,5-AG) were carbohydrates related to blood sugar control and thus expected in a high and low HbA1c analysis. Significant biochemicals commonly found in T2D and related to blood sugar control were not significant in T1D for both p and q in the screens. We conclude that few metabolites are correlated in T1D with tight vs. poor blood sugar control compared to T2D. Unlike T2D, metabolites related to intracellular sugar utilization, the Krebs cycle and purine synthesis are unrelated to blood sugar control in T1D cohorts. Disclosure W. Kuhtreiber: None. S.E. Janes: None. M.W. Yang: None. D.L. Faustman: None.