- Bias (variance from the target) and noise (variance from others) appear is most organizations
- System noise compounds rather than balance (cancel) out
- Most people are content with maintaining a single world view
- Focusing on common language allows us to avoid developing common standards
- Conflict avoidance helps support this
- (Feedback is really important to making consistent decisions)
- Verifiability does not affect our estimates
- (Just because something can be verified does not mean we are better at estimating it)
- Decisions require predictive and evaluative judgements
- Bias and noise are independent errors
- Avoid mixing your values and facts
- Waiting a few weeks to make a second guess helps improve personal guess
- To get most accurate average estimates
- Get input from other people
- Ask yourself a second time, weeks later
- Ask yourself to guess again, assuming your first guess is wrong
- Summing up votes can have a huge boost in accuracy
- Crowds are only wise if people register views independently from each other
- Discussing ideas with others often intensifies original beliefs
- “We do not need more accurate weights than our measures.”
- “Frugal rules” (algorithms with few inputs) are often better than complex algorithms but not as good as ML
- People are often willing to trust an algorithm until it makes a bad decision
- Most things happen in “the Valley of the Norm” where it is easy to explain after the fact
- Using comparisons instead of labels for more accurate judgements at scale
- 7 is the magical number of categories
- We like casual stories of explanation, even if they are not rational
- Intelligence has a strong correlation with financial success
- Open minded is better than strong confidence
- Documenting each step helps to reduce bias due to information exposure
Leave a Reply
You must be logged in to post a comment.