The Model Structure That Needs Improvement

If we want to try and predict or model better understand the relationship between training and adaptation and performance it's really the model structure itself that needs improvement. If you have six data points you can get a perfect fit with a sixth order or fifth order six-order polynomial statistically speaking you need five to fifty independent measurements for every adjustable parameter that you have. The problem we run into is people love to over parameterize the other statistician now I'm blanking on the name George Box is all models are wrong but some are useful who was it who said give me enough parameters I can draw an elephant?