This is more about using AI in the classroom. Often times family income is a major feature in determining low-performance. Simply not including this in predicti

This is more about using AI in the classroom. Often times family income is a major feature in determining low-performance. Simply not including this in predictive models isn't enough because you may find a proxy feature that is also highly correlated to family wealth. Current research in edtech is now looking at the transferability of AI models. For example, if I target a specifically low-income / low-performing school and use some type of process to improve their scores, will that model transfer to another school (with hetero- or homogeneous students). So the statement "Reduce childhood poverty and you’ll improve results." isn't directly referring to how wealth improves intelligence, but rather from a logistical regression perspective each "unit of wealth" is correlated to a higher performance.