Want to know what kinds of bias are fixable with statistics and how?
Read on... π§΅π
This is a simple mental map of how different biases affect the process of using algorithms to make changes to the physical world. The way we can fix each bias is as follows...
Read on... π§΅π
This is a simple mental map of how different biases affect the process of using algorithms to make changes to the physical world. The way we can fix each bias is as follows...
- Data selection bias: you need an accurate mathematical model of the data creation process
- Statistical bias: you need good statistics
- Bias due to generalization: you need an accurate mathematical model of the observations in the data and in the target population 2/7
- Statistical bias: you need good statistics
- Bias due to generalization: you need an accurate mathematical model of the observations in the data and in the target population 2/7
To fix the "bias due to causal assumptions", we need to fix all 3 smaller biases. At that point, if your model fits the data well then it should be a very close match to the world. In this case, correlation IS causation and we can say the inputs CAUSE the outputs. 3/7
Because we understand causation in the model, we can investigate the degree to which different variables cause different outcomes and this opens the door to the theoretical investigation of the causal effects of gender and race on the model outcomes. 4/7
We can ask how much they affect the behavior of the model and we can even modify the model to factor in the use of race and gender to the exact degree that we deem them necessary. There is one *big* caveat to this... 5/7
To use gender and race in a mathematical model, we have to quantify them. That is to say we have to capture these traits using numbers. Unfortunately, this process is complicated by the fact that quantification is partially subjective. (This requires a thread of its own!) 6/7
There are limits to what statistics can do. To use statistics, we must first translate reality into numbers. After applying our algorithms, we must then translate the numbers back. Unfortunately, statistics can only say so much about what's been lost in translation. 7/7
I hope you found this thread informative. If you would like to support this kind of content, follow me and also like and retweet the thread. π
β οΈ In the thread, I use the term "statistics" to cover both "causal inference" (the study of mathematical causation) and "statistics". I think these fields are so co-dependent that they're basically the same field and will eventually merge in a few decades but others disagree! π
β’ β’ β’
Missing some Tweet in this thread? You can try to
force a refresh
γ
