Statistics and Machine Learning: Foundations, Limitations, and Ethics
Refreshments at 3:30 PM, Davis 2nd floor
When should we use statistics? When should we use machine learning? When should we not use either one? The basic philosophical assumptions of statistics and machine learning can be hard to see amidst practice, but they are key for seeing their limitations, appropriate uses, and ethical implications. This talks gives a broad overview of the nature of statements in statistics and machine learning, covering historical roots (about the central tendency, use of probability, and the "reduction of data"), sociological critique (around forcing the world into mold of data matrices, and modeling being connected to the exercise of power), and limitations according to statistical theory itself (such as in the bias-variance tradeoff suggesting that a "false" model can predict better than a "true" one). Understanding these limitations, we can more responsibly decide how, and indeed if, to use statistical tools, and which tools to use.