Engineers use supervised learning in AI/ML to build highly accurate, scalable tools for national planning, energy distribution, and language translation. Supervised regression models predict a continuous numerical value, such as a price or temperature. The mathematics behind regression models is completely neutral. However, because these algorithms try to find a “line of best fit” or a probability curve across an entire dataset, they are mathematically forced to prioritise the dominant Eurocentric narrative.
Simple Linear Regression
Simple Linear Regression uses Ordinary Least Squares (OLS) to draw a single straight line through a dataset to predict a relationship between an independent variable (X) and a dependent variable (Y). In AI/ML, this can force a straight, extraction-oriented trendline through complex, cyclical, or disrupted historical realities.
Since OLS works by minimising the squared distance (residuals) between the trendline and every data point, and because the internet is flooded with Eurocentric data, the trendline is pulled sharply in that direction.
When the model encounters a different narrative, this situation represents a statistical outlier. Because Simple Linear Regression squares its errors, it can treat a unique observation as a massive “error” that needs to be smoothed over. The algorithm therefore forces the data point to fit the standard Eurocentric trendline, potentially obscuring the underlying reality.
OLS minimises squared residuals, so a dataset with a disproportionate concentration of Western data points can exert substantial influence over the resulting trendline, potentially treating less-represented realities as statistical outliers.
Conclusion
Linear Regression uses Ordinary Least Squares (OLS) to minimise the sum of squared residuals. It is a mathematical optimisation problem designed to find the line of best fit for a given set of inputs and outputs.
Linear regression makes absolutely no assumptions about capitalism, growth, or exploitation. It simply states: “Given X, this is the most mathematically probable Y.”
The concern, therefore, is not that linear regression itself is biased towards Western narratives. Rather, the resulting trendline can be influenced by the composition of the dataset on which the model is trained. If Western data points vastly outnumber other perspectives, those observations can exert greater statistical influence over the resulting model.
Day 17/ 30 of the #AlRewardmaxxing Series.
Tomorrow in Part 18, we're stepping deeper into machine learning mechanics to explore how multiple regression and feature weighting amplify structural bias!
What's your take? When pure mathematical optimization treats unique truths as statistical errors, how do we fix the line of best fit? Let's discuss below!