The document discusses how data is increasingly being used to make decisions but that data-driven decisions are not always good. It notes several cognitive biases that impair human decision making and how roads help navigate to locations while data helps navigate decisions. However, data has flaws like biases, errors, and being out of date. To help ensure data-driven decisions are good, the document recommends that decision systems be open to scrutiny, allow appeals and challenges, be tailored to minimize error impacts, enable reflection and correction of biases, and introduce diversity into decision making processes. The overall message is that decisions should be informed by data, not blindly driven by data.
45. Be open to scrutiny
Allow appeals & challenges
Tailor algorithms to the impact of errors
Enable reflection & correction
Introduce diversity into decision making
decisions that are pooled with other people's decisions to decide who makes decisions that have impact on a lot of us
This becomes more important as code is generated based on data (machine learning), because then the data creates the code. The flaws in the data leads to the bugs in the code.
We censor ourselves according to social norms, try to work against our own biases. Could algorithms do the same?
Decisions involving humans often involve shades of grey
Rich understanding of context
Empathy & compassion